Nishi Family › Compare › Graphics and Sovereign Render
Nishi Compare · capabilities, evidence and delivery
Graphics and Sovereign Render
Compare capabilities and delivery evidence. Source discovery, qualification, deployment and accepted outcomes are separate; an open contract names work still to verify.
Current evidence for Beach rendering, procedural characters and motion. Alpha playback and joint-stage observation are verified; photoreal character appearance and full-cast motion/contact remain unaccepted. Native procedural body generation is reproducible, but its geometry is not yet a qualified animated NPC. Historical scores are retained below and are not a current SOTA rating. Competitor entries require dated, versioned outcome comparisons.
Overview
The purpose, declared position and evidence coverage of this domain. Source presence and completed acceptance are different measures.
Where we are. Beach Alpha now renders prepared characters with declared shared scene lighting. The normal load/frame controls and Scene lighting -> neutral diagnostic -> Scene lighting interaction are verified on the published build. Directional shading is delivered; anatomy, skin realism, cloud/terrain shadows and night correctness remain unaccepted. Existing collision runtime and asset identities are preserved. Open the current Alpha below; exact build identities, images and limitations remain in the retained delivery history.
Where we need to go. Beautiful, reproducible procedural men, women, animals and environments, including fantasy and anime styles, in one coherent isekai world with shared geometry, materials, rigging, contact and VR. Use the highest available device capability for quality, then adapt explicitly to resource limits, battery preference or failure. Preserve meaningful features across quality levels and declare downgrades. Measure against current Houdini/MetaHuman-class references and independent user outcomes; anatomy, photoreal appearance and clinical predictive validity require separate evidence.
Research bar. Infinigen (Princeton, BSD-3) is measured on procedural nature, creatures, indoor rooms via constraint solve, materials, ground-truth annotation; 2.0 alpha adds controllable APIs and indoor materials and rooms; 331GB flying-indoor video dataset Aug 2026. Theirs: five genome classes, NURBS part lofting, single-surface creatures -- source-level read, grade-1. Ours: our expansion rungs GR4-GR6 and the ground-truth contract flip measured on this page.
Research bar. SIGGRAPH Advances in Real-Time Rendering is measured on twenty years (2006-2026) of the industry's shipped real-time technique. Theirs: the named-technique record. Ours: GR7 mirrors it into the library beside the 9762-paper index; rows then cite talks by name.
Research bar. MetaHuman (Epic) is measured on the external photoreal digital-human bar. Theirs: photoreal artist-authored rigs. Ours: the panel grades our figure against MetaHuman-class stills; our north star, the inside-and-out twin, is a HIGHER bar they do not enter.
Research bar. Modly workflow completeness (repository review 2026-09-19; runtime not exercised) is measured on Documented image-to-mesh-to-scene workflow; canonical automation lifecycle includes model, capability, workflow-run and process-run with status/cancel/export recovery. Nishi delivery target is a coherent playable scene: shared versioned characters, motion, environment, lighting, sound and interactions composed through Alpha, Beta and Production. Character inspection is one component work area, not the delivery unit. Native retained README knowledge/fetched/rf_c3f0d26b.raw SHA83e18fc0bc9776427a3b1dad9b0f2692ff4b433ac93bf5ade76598baa84ff853 and agent contract rf_8f1034dd.raw SHAfa0661bb83700967177ef441eb0aa39fcb7e17068ff8e98f97835d5a9a9784bd.. Theirs: Repository and CLI source inspected; runtime not exercised. Ours: Nishi complete-workflow comparator: select or generate, inspect, push pinned version to Alpha, approve Beta; shared renderer remains incomplete.
Research bar. NASA model and simulation credibility baseline (verified 2026-09-19) is measured on NASA-STD-7009B approved 2024-03-05 supersedes A with Change 1. Separate capability credibility from credibility of each result; preserve units, coordinate frames, source pedigree, uncertainty and defect history. Native researcher retained knowledge/fetched/rf_2f9edbe1.raw SHA7b95a2065325dc3035c35ce110557d817555d3f22210ce5f6ce0bb2fb4ccd8b4, 1417365 bytes.. Theirs: Official current standard metadata and PDF inspected; no Nishi conformance claim. Ours: Use as validation architecture baseline, not as proof that Beach is accurate.
Research bar. Medical simulation credibility by intended use is measured on FDA final guidance November 2023 provides a risk-informed credibility framework for physics and first-principles models. Visual likeness, interactive behavior and prediction of clinical tissue outcomes require separate evidence and contexts of use.. Theirs: Primary guidance verified 2026-09-19; no clinical qualification claim. Ours: Bind claims to context, consequences, comparator evidence and applicability.
Research bar. Measurement uncertainty and model development is measured on Official JCGM index includes GUM-1:2023, GUM-6:2020, GUM-5:2026 and JCGM100 amendment1:2026 on nonlinearity. Index currency verified; new 2026 documents require detailed review before adoption.. Theirs: Primary index reviewed 2026-09-19; full 2026 amendment not reviewed. Ours: Maintain uncertainty budgets and dated source lineage rather than universal accuracy claims.
Where we are. Measured 2026-09-20 native craft qualification job 1789924794: nx_wasm_craft 96/96 GREEN, gamefeel 17/17 GREEN, wasm VM 20/20 GREEN, visiting cast 4/4 TEXC+TEXM, page JavaScript 1,184,815 scanned bytes with 0 scope reports, and all recipe pages read back with loader markers. Source/config identities: nx_craft_emit sha256 de1a052ac7f679e72a89f8fd527d52c3f273b9dc6abc8498782288e3df0e9; nx_jsscope sha256 0054acfa1d9143c7aa4d06d1e028b2feeff9b6f37984ad02b1e4707a0d0f6210; world_recipes path contract sha256 cca9284d9119f63bdf25ca907b198cec7f3ce1e3bafbd5f91084f9c558d28102. Browser discovery returned no available browser, so served visual/interaction acceptance remains UNVERIFIED and no Beta or Production claim is made.
Where we are. Correction to the preceding receipt row: the authoritative nx_craft_emit hash is de1a052ac7f679e72a89f8fd527d52c3f273b9dc6abc849eff8782288e3df0e9 (the prior row contained a transcription typo); the page qualification result and all other evidence remain unchanged.
Where we are. Measured 2026-09-20
Where we are. Incident 2026-09-20
Where we are. Blocked 2026-09-20
Where we are. Evidence 2026-09-20
Where we are. Measured 2026-09-20
Where we are. Measured 2026-09-20
Where we are. Measured 2026-09-20
Where we are. Measured 2026-09-20
Where we are. Measured 2026-09-20
Where we are. Measured 2026-09-20
Where we are. Measured 2026-09-20
Where we are. Measured 2026-09-20
Where we are. Measured 2026-09-20
Where we are. Measured 2026-09-20
Latest recorded release
2026-10-04 · GR62
INTAKE CLASSES PUBLISHED (the 1791063220 queue row closed). Queued rows had not fired (builds predated source by ~1.6 h), so built by hand once admission granted (load1 3.4, blocked 0). nx_refarea_gate --qualify read 87/88 RED from CWD buildroot: section E 20 of 20 PASS incl. live names 6 of 6 and live oracle stems 2 of 2; the one RED was live-conf-files-cc-by-nc-nd-in-cc-nc-nd plus 0 live manifest rows, both because the live conf and manifest are serving-root relative and --qualify anchors CWD to buildroot (the conf does carry the NC+ND row). The same candidate staged and run from the serving root: nx_behaveprobe other_rc=0 (GREEN). Content ruler, every LOST line read: gate lost 0 (gained 189), regen lost 0 (gained 29), matrix lost 0 (gained 120), sota lost 0 (gained 122), all GREEN ONE-WAY-GAIN. Promoted with expect_sha256: gate f12f25eb, regen 664f55d4, matrix 32acd2d6, sota f91c638a; nx_offc_install promoted on all four, buildroot/_offc REFRESHED for regen, matrix and sota; regen FORK-FRESH on the matrix fork. Scoped republish aesthetictwin (index private_spans=20, api 0) and graphics (0 and 0), each GREEN and each rewriting feed.xml; nx_page_verify GREEN on both. Served-byte recount (nx_txtscan count, items named here by count only) over aesthetictwin index.html and api.json, graphics index.html and api.json, and feed.xml: every private.names name row in three case forms 0 in all five files; the third-party site name 0; stemnum spans (prefix plus each digit 0-9) 0, the bare prefix surviving only as function names and citation keys; the intake store path 0; the vault root 0; both oracle stems 0; optout.names carries no rows yet (0 needles). Vault cids: 1 in each aesthetictwin file, both the declared public gallery url; feed 1 prose mention with no hex. private-ref tokens on the pages: aesthetictwin index 371, graphics index 260. Residual, named: aesthetictwin index.html carries two non-oracle intake subject stems at 2 occurrences each (exempt by design, the allow-non-oracle-intake-file-stem control) and 16 plan-plane ids sharing a prefix with intake plan file names; one gallery image path under the NAS ai image tree (not the vault). Gate binary on its serving-root run is the evidence; a NAS /api/gate_run receipt with the live manifest rows was not taken this leg
Release entries describe recorded changes; they do not establish that every capability passed evaluation.
51 of 105 capability source symbols found|4 of them authored stronger-claims|54 open|coverage 485/1000|adoption 30 full / 21 partial
Evidence profile — what the gaps on this board actually are
Measured by nx_swcompare_evidence, read back by nx_evprofile_lib. Every figure is a count with its denominator — there is deliberately no score, no grade and no percentage anywhere in this band, because a stored scalar is a field a seat can edit and a counted partition is not.
evidence|grounded 44/44|unsupported 0|gates green 17/17|proven able to fail 0/17|never bitten 17|green at 0/0 0|open gaps 46|of them unnamed 0|of them proof withheld 0|flips ready 2
proven able to fail counts the gates that have a RECORDED RED — nx_gate_bite mutated the gate subject, rebuilt it, watched the gate go red, and that record is inside the shared TTL. never bitten is its complement over the same denominator: those gates ran and were green, and nothing has ever shown them able to detect anything, so their green is a statement about this run and not about the gate. green at 0/0 is a separate and much weaker observation — the gate printed GREEN on a zero denominator, so its own tooth counter says it examined nothing. A gate can be green, non-zero, and still never bitten; that is the common case and it is now visible instead of implied.
partition: grounded + unsupported = 44 vs present 44 · named + unnamed + withheld = 46 vs open 46 · both reconcile
liar-kill conj=GPQN · all four conjuncts held
graded document: BUILDROOT tree, 98494 bytes · gates map: PRIMARY · stamped 36d 3h ago · STALE against the shared TTL of 604800s — this is evidence about a build nobody can vouch for now · source ../knowledge/status/evstamp_graphics.verdict
| Gap class | What it is, and the work it names |
|---|---|
| FLIP-READY | A watch contract symbol is now present in the organ it named: the cell can flip on a pure data edit with no new code. These are the cheapest closures on this board and they belong before any new work. |
| STALE | This profile is older than the shared TTL, or its clock ran backwards. It is reported as unusable rather than as fresh, because being wrong in the direction of refusing is the safe direction. |
nx_swcompare_evidence graphics itself and are not carried on the stamp, so this page names the classes and the producer names the rows. That split is stated rather than hidden: a count without a worklist is not actionable, and this band is honest about which half of that it is.Production map
Follow the dependencies, declared acceptance criteria and recorded priorities. Inspect source binding before treating a rank as executable work.
Ranking source binding: PLAN_MATRIX_BOUND_ONLY. Recorded priorities require current acceptance evidence and resource checks before execution.
Ranking matches the captured plan and matrix only. Latest execution outcome, research freshness, accepted delivery and investment return are unverified.
Recorded priority estimates
Order from nx_compare_rank (nx_dr_ocm: (deficit + cost-of-delay + option + enables) x sponsor x self-sufficiency x momentum / cost). FINISH rows are rungs whose symbol is present but whose organ is short of full adoption: listed before new work by this heuristic. Priority is not measured delivery cost or execution readiness. Stamp: # asof=1791487931 domain=graphics target_version=0.1 rungs=63 done=14 open=49 finish=4 ranker=nx_dr_ocm
| # | Stage | Rung | Priority | Derivation |
|---|---|---|---|---|
| F | FINISH | Path-traced GI (GR10) gp_bounce_gi dossier | REGISTERED-UNAUTHORISED | registered, no cap ever minted; no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked) |
| F | FINISH | Skin-bound dynamics (GR13) dbs_displace_free dossier | REGISTERED-DARK | callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked); only referenced by nx_dyna_bind_skin_gate -- a mention, not a run |
| F | FINISH | Physically-based sky and aerial perspective (GR26) gpe_sky_atmos dossier | REGISTRY-UNCERTAIN | registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion |
| F | FINISH | World-quality referee built and wired (GR30) wqg_build_wired dossier | REGISTERED-DARK | callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked) |
| #1 | 0.1 | The leap: genome splat avatar (GR2) gs_avatar_genome dossier | 1000 | v=20 m=2 c=40 |
| #2 | later | Ground material: granular sand and wet-shore darkening (GR27) wc_mat_ground dossier | 3800 | v=19 m=2 c=10 |
| #3 | later | Linear-light pipeline (GR14) sr_linear_light dossier | 3000 | v=30 m=1 c=10 |
| #4 | later | The beauty tier's first frame: the beach scene rendered offline by a sovereign path tracer and published beside the live frame with its referee score (GR42) env_beauty_scene dossier | 2025 | v=54 m=3 c=80 |
| #5 | later | Sovereign strand system (TressFX-referenced) (GR15) hr_strand_sys dossier | 2000 | v=40 m=1 c=20 |
| #6 | later | Render the sections we already generate (GR19) ntb_view_sections dossier | 2000 | v=40 m=1 c=20 |
| #7 | later | One colour transform, two execution surfaces, ours (GR23) sr_view_transform dossier | 2000 | v=30 m=1 c=15 |
| #8 | later | The normal map already inside the asset is read (GR34) gpe_texm_normal_bind dossier | 2000 | v=10 m=1 c=5 |
| #9 | later | The JS-logic ratchet is pinned against class laundering (GR37) jl_class_pin dossier | 2000 | v=20 m=1 c=10 |
| #10 | later | Strand chain ordered root-to-tip (GR33) hg_seg_axis dossier | 1800 | v=9 m=1 c=5 |
| #11 | later | Spherical-harmonics irradiance from the physically-based sky for the real-time tier (GR45) sr_sh_irradiance dossier | 1600 | v=12 m=2 c=15 |
| #12 | later | Marschner-class hair shading (GR16) hr_marschner_lobe dossier | 1333 | v=20 m=1 c=15 |
| #13 | later | Second species canon (GR4) bp_species_canon dossier | 1000 | v=10 m=1 c=10 |
| #14 | later | Garment and hair penetration ruler (GR20) cc_garment_pen dossier | 1000 | v=15 m=1 c=15 |
| #15 | later | A second art style of emission beside the voxel raycast (GR31) wc_style_surface dossier | 1000 | v=20 m=1 c=20 |
| #16 | later | The rendered strand is the collided strand: character geometry leaves emitted JS for NishiLang lowered to wasm (GR48) gpe_char_geom_wasm dossier | 1000 | v=40 m=1 c=40 |
| #17 | later | The translation ladder is declared and counted per target: zero hops on NishiOS and Nishi browser, minimum hops elsewhere, an API layer only where a third party mandates one (GR49) tl_translation_census dossier | 1000 | v=10 m=2 c=20 |
| #18 | later | Skin-tone fidelity of the LIVE render is measured in Delta E and ITA error per class under named lights (GR55) skin_tone_fidelity_dE dossier | 1000 | v=15 m=1 c=15 |
| #19 | later | A camera model as data (GR61) _ABSENT_:gr_camera_row dossier | 1000 | v=30 m=1 c=30 |
| #20 | later | Native-resolution GPU delivery (GR11) cw_webgpu_compute dossier | 900 | v=9 m=1 c=10 |
| #21 | later | Palm flora reaches the SERVED beach (GR28) wc_veg_mesh dossier | 900 | v=9 m=1 c=10 |
| #22 | later | FACE and MORF are consumed, so the face is geometry and not paint (GR36) gpe_morf_consume dossier | 900 | v=9 m=2 c=20 |
| #23 | later | Skin micro-surface on the path the family looks through (GR29) gpe_skin_microsurface dossier | 866 | v=13 m=1 c=15 |
| #24 | later | A declared asset ladder, so finished work reaches the cast (GR35) gpe_asset_ladder dossier | 800 | v=4 m=1 c=5 |
| #25 | later | A character crop that contains a character (GR39) ws_char_crop_nonempty dossier | 800 | v=4 m=1 c=5 |
| #26 | later | A low-discrepancy sampler for the path tracer: Sobol with Owen scrambling, the blue-noise organ reused for the screen (GR44) gp_sobol_owen dossier | 800 | v=12 m=1 c=15 |
| #27 | later | The sovereign strand organ reaches a live world (GR32) wc_strand_bind dossier | 700 | v=7 m=1 c=10 |
| #28 | later | A ray-triangle bounding-volume hierarchy for the beauty tier: the path tracer hits the cast meshes (GR43) gp_bvh_ray_tri dossier | 600 | v=12 m=1 c=20 |
| #29 | later | Cascaded shadow maps on the kernel and WebGPU tiers: the cast casts a shadow on the ground (GR46) rk_shadow_cascade dossier | 600 | v=12 m=1 c=20 |
| #30 | later | Groom style zones with pinned dynamics (GR17) gr_style_zones dossier | 466 | v=7 m=1 c=15 |
| #31 | later | Genome asymmetry -- synthetic faces are too perfect (GR21) bg_asym_permil dossier | 400 | v=4 m=1 c=10 |
| #32 | later | Camera rejection sampling (GR22) ntb_cam_reject dossier | 400 | v=2 m=1 c=5 |
| #33 | later | Identity parameter space with correlated sampling (GR24) bg_identity_space dossier | 350 | v=7 m=1 c=20 |
| #34 | later | Placement solved, not authored (GR25) pl_fit_solve dossier | 350 | v=7 m=1 c=20 |
| #35 | later | Infinigen fork map as data (GR18) if_fork_map dossier | 200 | v=2 m=1 c=10 |
| #36 | later | Part-genome creatures (GR5) cg_genome_parts dossier | 150 | v=3 m=1 c=20 |
| #37 | later | The support surface and its cover (GR60) _ABSENT_:gr_support_cover dossier | 125 | v=1 m=5 c=40 |
| #38 | later | The board ranker measures every row (GR40) rk_measure_all dossier | 100 | v=1 m=1 c=10 |
| #39 | later | Real-versus-render indistinguishability bench with outside liar-killers (GR63) _ABSENT_:gr_realrender_bench dossier | 100 | v=1 m=3 c=30 |
| #40 | later | Sheer and stretched fabric over skin (GR57) _ABSENT_:gr_sheer_bsdf dossier | 75 | v=1 m=3 c=40 |
| #41 | later | 4D splat motion (GR3) g4_avatar_rig dossier | 66 | v=1 m=1 c=15 |
| #42 | later | Wet as a surface state on skin, hair and sand (GR58) _ABSENT_:gr_wet_state dossier | 66 | v=1 m=2 c=30 |
| #43 | later | The review viewers gain the sovereign wasm tier (GR41) nre_sov_tier dossier | 50 | v=1 m=1 c=20 |
| #44 | later | Our subsurface term is calibrated against a measured face BSSRDF and its RMSE is published (GR56) skin_bssrdf_calibration dossier | 50 | v=1 m=1 c=20 |
| #45 | later | The ship loop adjudicates itself against the SERVED artifact and refuses to report success for a provably inert promote (GR50) os_served_delta dossier | 0 | v=0 m=1 c=20 |
| #46 | later | The hair gate gains a tooth the broken output must fail: rendered strands inside the body are counted, not assumed (GR51) gpe_hair_penetration_tooth dossier | 0 | v=0 m=1 c=15 |
| #47 | later | The estate's searcher can read the artifacts the estate serves (GR52) st_longline_windows dossier | 0 | v=0 m=1 c=20 |
| #48 | later | The estate reads its own served pages with its own wire: a sovereign CDP witness composed from the shipped WebSocket stack (GR53) nx_cdp_eval dossier | 0 | v=0 m=1 c=15 |
| #49 | later | ONE HAIR SYSTEM: five parallel implementations collapse to one, and this rung is measured by DELETIONS not additions (GR54) hair_one_system dossier | 0 | v=0 m=1 c=40 |
Declared roadmap — contract, acceptance, executor, effort
| Rung | Closes with | Definition of done (pre-declared) | Executor | Est. |
|---|---|---|---|---|
| Referee panel (GR1) | cjc_panel | LANDED 2026-08-19 in one leg (estimate held): cjc_panel composes the statistics judge, the published Kovac skin-gamut ruler, image_beauty and nx_percept, forked per image; FP and FN measured on the labelled corpus v1 (n=10, 3 BAD by operator record, 7 GOOD = the six banked WebGL NXA rig stills captured on the real GPU + the published Mackenzie render) -- stats and symmetry proven non-separating and left report-only, skin and percept separable, floors 85/152 from the margins, composite FP=0 FN=0, ADMITTED under a floor-bound receipt; nx_cjc_panel_gate GREEN, bite-proven both ways. Open inside the rung: the VLM lens is UNOBSERVABLE until a sovereign VLM oracle exists (none on the estate, measured), and the floors need the second-corpus retest (MetaHuman-class stills) before they are believed | Organ | 1 u |
| The leap: genome splat avatar (GR2) after GR1 | gs_avatar_genome | RIGHT-SIZED 2026-08-18 to the published bar (the earlier budget understated by ~10x): v2 by canon-mode bodyfit search (16-dimensional ring-scale under the judge), the gated face pipeline composed in AT >=100K-gaussian face density (GaussianAvatars-class heads), strand-aligned splat GROOM hair at groom scale (85 percent of the industry poly budget), 8K-class skin with wrinkle response; done when the panel grades it above the ring figure AND the banked WebGL reference on the same framing -- declared before the experiment. THE BAR IS NOW DATA (leg 1, 2026-08-19, admitted panel on knowledge/cjc_gr2_framing.list, every row a figcrop): ring figure skin 7 percept -112 BAD; splat v1 skin 118 percept 123 BAD (skin-coloured, perceptually junk-ward); the six WebGL rig crops skin 279-661 percept 210-228 GOOD; Mackenzie skin 624 percept 281. DONE-RULE IN NUMBERS: v2 crop must read composite GOOD with percept above 228 (the best banked rig crop) and skin at or above 279 (the weakest), re-measured with the same conf and receipt | Organ | 4 u |
| 4D splat motion (GR3) after GR2 | g4_avatar_rig | The splat avatar moves with the same gait, breath and expression machinery the mesh uses; replay bit-identical | Organ | 1.5 u |
| Second species canon (GR4) | bp_species_canon | A horse canon through the identical pipeline; a bodybench-style ruler against a CC anatomy oracle of the species proves species-agnostic generation | Organ | 1 u |
| Part-genome creatures (GR5) after GR4 | cg_genome_parts | Typed interpolatable parts over the existing plans; distinct genomes must measure distinct and interpolation must stay anatomically legal | Organ | 2 u |
| Indoor rooms (GR6) | ig_room_solve | Constraint-solved rooms with materials through our engine, deterministic and bounded; REFUSES an unsatisfiable constraint set rather than overlapping furniture | Organ | 2 u |
| Advances archive mirrored (GR7) | gp_collect2 | Second root through the ingest organ, fixed-point crawl, partition reconciles per root; the extractor then cites it | Organ | 0.5 u |
| PBR map set per genome (GR8) | ntb_bake_pbr | Four maps by four regions regenerated per genome at load; normal-map lighting proven through the existing image path | Organ | 1 u |
| Pre-integrated SSS (GR9) after GR8 | sr_sss_preint | Bound to the baked normal map; carries the away-side equality tooth (the T66 form) so no uniform brightener can fake it | Organ | 1 u |
| Path-traced GI (GR10) | gp_bounce_gi | Per-frame bounce beyond cached radiosity; the cache-vs-rebuild equality stays as the determinism tooth | Organ | 2 u |
| Native-resolution GPU delivery (GR11) | cw_webgpu_compute | HALF LANDED 2026-08-18: the WebGL2 tier is LIVE DEFAULT on all five worlds and the WebGPU WGSL world pass is LIVE LEVERED (?gpu=1, real-GPU verified via the CDP wire, boot-stall ceiling hardened); REMAINING: the compute-shader form, the NPC+eye passes to parity so WebGPU can default, and the same frame bit-compared against the CPU path at reference size | Organ | 1 u |
| Expression morphs (GR12) after GR2 | fa_expr_morph | FACS axes over the identity morphs, graded against the 710-expression FACS-class basis MetaHuman ships (the bar, not a smile bone); the panel grades expressions readable | Organ | 2 u |
| Skin-bound dynamics (GR13) | dbs_displace_free | CONTRACT SYMBOL CORRECTED 2026-09-01 (board owner): this rung declared dyna_bind_skin while its matrix row named runtime/nx_nxa_dyna.nx, a pair that could never resolve, so GR13 held ranker slot #8 as OPEN for a capability that had shipped. Measured full-population, corpus_complete=1: func dyna_bind_skin( is declared once, at runtime/nx_softbind.nx:238, and it ticks three soft-tissue REGION points -- not this rung's subject; func dbs_displace_free( is declared once, at runtime/nx_dyna_bind_skin.nx:203, and is the per-vertex half that procgen.matrix and softbody.matrix already carry LIVE. DONE-RULE, unchanged in substance and now pointed at a gate rather than at a declaration: the proven solver bound to skin VERTICES with collision against the body it rides, the flip being the receipt and nx_dyna_bind_skin_gate on the graphics.gates roster being the proof -- a green cell over a red gate is not done | Organ | 1 u |
Milestones
| Milestone | Rungs | Cumulative |
|---|---|---|
| M0 · Referee trustworthy | GR1 | 1 u |
| M1 · The leap published | GR2 | 3 u |
| M2 · Living avatar | GR3,GR12,GR13 | 6.5 u |
| M3 · Infinigen expanded | GR4,GR5,GR6,GR7 | 11 u |
| M4 · Surface and delivery | GR8,GR9,GR10,GR11 | 18 u |
| Rung | Closes with | Definition of done (pre-declared) | Executor | Est. |
|---|---|---|---|---|
| Linear-light pipeline (GR14) | sr_linear_light | All lighting math computed in linear, display-encoded exactly ONCE at output; proven by the tonemap A/B that failed 2026-08-26 (ACES over display-encoded values washed the frame -- the banked double-transfer lesson) now passing: filmic curve applied and the A/B screenshot referee grades it richer, not paler; prerequisite to GR9 | Organ | 1 u |
| Sovereign strand system (TressFX-referenced) (GR15) | hr_strand_sys | The Infinigen-fork hair chain, leg 1: a NishiLang strand system whose sim and render structure is read from the mirrored TressFX source (the canonical shipped open real-time strand system) -- guide strands + per-strand constraints on the one solver, rendered with real width and coverage in our WebGL2 tier; done when a girl-framed capture is graded by the panel above the current ribbon baseline and no ad-hoc constant lacks a cited row | Organ | 2 u |
| Marschner-class hair shading (GR16) after GR15 | hr_marschner_lobe | Hair BRDF per Marschner 2003 through the Karis 2016 / Tafuri 2019 production reductions (mirrored): dual-lobe spec with shifted primary and secondary highlight, azimuthal roughness; constants pinned as cited conf rows before the shader edit, gate compares lobe positions against the published reduction | Organ | 1.5 u |
| Groom style zones with pinned dynamics (GR17) | gr_style_zones | nx_nxa_groom gains the cosmetology-brief zone model (hairline, part, nape, density map) with strand dynamics constants extracted from the mirrored Super-Helices / Discrete Elastic Rods papers and PINNED as oracle rows (closing the conf's own uncited hair_k_q10 correction record); style presets are parameters, every band cited or derived | Organ | 1.5 u |
| Infinigen fork map as data (GR18) | if_fork_map | The fork strategy is a LIVE census, not prose: a conf mapping each Infinigen subsystem (ProcFunc generators, constraint solver, materials, creatures, ground-truth writers, Infinigen-Sim articulation) to its sovereign NishiLang twin, its governing SOTA reference, its gate, and its measured state -- rendered on this board and re-measured at emit; the differentiators (inside-out anatomy, genome identity, bit-exact determinism, the adult invariant) carried as exceed rows | Organ | 1 u |
| Render the sections we already generate (GR19) | ntb_view_sections | MEASURED 2026-08-28: ntb_view finds only VERT and TRIS, so a groomed and dressed asset renders IDENTICALLY to a bare one -- nv=14164 tris=28092 filled3=56482 for both, with 4800 generated hair strands and a 648-vertex measurement-fitted garment present in the file and ignored. We GENERATE hair and clothing and cannot SEE them server-side, so GR15-GR17 and the garment lane have no judgeable artifact and the referee panel can never be pointed at them. Done: ntb_view consumes GVRT/GTRI and HSTR (strands as camera-facing ribbons), reports gverts/gtris/strands SEPARATELY from the body counts, refuses on index overrun, and stays a no-op when the sections are absent; proven by the emitted PNG differing and the reported triangle count strictly exceeding the bare asset's on the same input. Garment reuses the existing rasteriser; the open design question is that tm_set_image is global, so garment and hair need either a flat-region UV or a per-vertex vertex-colour path | Organ | 2 u |
| Garment and hair penetration ruler (GR20) after GR19 | cc_garment_pen | nx_clipcheck already measures the two GROSS self-clipping classes and on the shipped ref9.nxa returns verdict=CLIP (44064 keys, torso_clip_keys=146 worst 21mm, floor_clip_keys=2176 worst 30mm) -- but its axes are limb-through-torso and foot-through-floor. The operator-reported defect is GARMENT-through-body and HAIR-through-body, and NOTHING in the estate measures it: a ruler exists for the neighbouring defect and reads RED while the reported defect has no ruler at all. Done: per-vertex penetration of GVRT against the body surface and of HSTR roots/segments against the scalp, reported as a count plus worst depth in mm plus the offending element index, before and after XPBD settle, with a neg-control that a deliberately oversized garment reports MORE penetration than the fitted one | Organ | 1.5 u |
| Genome asymmetry -- synthetic faces are too perfect (GR21) | bg_asym_permil | MEASURED 2026-08-28 on the ADMITTED referee panel (cjc_panel, floors from knowledge/cjc_panel.conf): cjp_sym reads 938 for our generated body and 913 for both MetaHuman frames against 650 for the photographic bar -- every synthetic subject is ~1.4x more bilaterally symmetric than a real person, and the panel measures it today with no rung consuming the signal. This is a realism defect that costs NOTHING to fix in a procedural generator and is structurally unavailable to a finite asset library: a real face is asymmetric and ours is machine-perfect. Done: a per-mille asymmetry gene drives independent left/right ring-scale and feature offsets from the seed, deterministic and reproducible from the genome; proven by cjp_sym moving toward the photographic band without cjp_percept falling below the admitted tier floor | Organ | 1 u |
| Camera rejection sampling (GR22) | ntb_cam_reject | Infinigen's published figures are honestly non-cherry-picked at the IMAGE level, but the VIEWPOINT is rejection-sampled best-of-ten by pixel-depth variance -- so part of the quality gap a reader attributes to their renderer is a composition filter. Cheapest visual win available: the emit path renders N candidate viewpoints, scores each by depth variance over the figure, and keeps the best. Done: N and the score are conf rows not literals, the chosen index and its score are printed so the choice is auditable, and a neg-control shows the WORST-scoring viewpoint grades lower on the referee panel than the best -- if it does not, the heuristic is not measuring what it claims and the rung fails | Organ | 0.5 u |
| One colour transform, two execution surfaces, ours (GR23) after GR14 | sr_view_transform | SOVEREIGNTY RULE APPLIED (operator 2026-08-28: we dont want webgl2 and other 3rd party unless we are recreating ourselves the capabilities). The transfer curve and the LUT are DERIVED AND GENERATED IN NISHILANG; AgX is an ORACLE we check ourselves against, never a dependency -- borrow the rulers, not the runtime. TWO INDEPENDENT RIVALS CONVERGE ON IT, which is why it is worth recreating: Infinigen never sets a view transform and inherits the Blender 4.2 AgX default with compositor grading OFF (read byte-for-byte in our mirror: distort False, color_correct False), and VR HOT deliberately ADDED AgX in 0.13 on 2026-06-16 calling it a color profile you might know from other high end games. Done: ONE NishiLang implementation of the curve consumed by BOTH the page-side surface (beach lane) and the server-side portrait renderer (GR19), byte-identical LUT from one generator, so the two surfaces CANNOT diverge; the native renderer is the DURABLE home and the browser page is a stopgap execution surface. GR14 is the prerequisite and carries the measured lesson: the 2026-08-26 A/B failed because ACES was applied over display-encoded values and washed the frame -- transform over LINEAR light, display-encode exactly once | Organ | 1.5 u |
| Identity parameter space with correlated sampling (GR24) | bg_identity_space | THE INTELLIGENT-PROCGEN HALF, and the axis a finite asset library structurally cannot enter. MEASURED RIVAL POSITION: VR HOT ships roughly 15-20 hand-picked skins accumulated 2-4 at a time over five years on a FIXED LICENSED TOPOLOGY (Daz Genesis 8.1, vendor-stated in their own 0.5.4 patch notes), so their identity space is bounded by what one marketplace authored; Infinigen has NO humans at all. Ours is a seed. Done: semantic master parameters (age, adiposity, sun exposure, grooming, dimorphism) fan out into correlated low-level axes through a FULL COVARIANCE draw rather than independent per-channel sampling -- the mechanism behind why a rival palette reads deliberate instead of tinted -- with each axis drawing from its own split stream so adding a parameter never reshuffles existing genomes; every character reproducible and auditable from its genome, proven by bit-identical regeneration from the seed and by the referee panel showing identity spread across a population without any member falling below the admitted tier floor | Organ | 2 u |
| Placement solved, not authored (GR25) after GR20 | pl_fit_solve | Their known-issue lists are our exceed axis, measured: VR HOT carries nipple clipping, bra clipping, body-hair-through-cloth and texture seams as PERSISTENT known issues across years, because garments are an authored library conformed onto a licensed topology; Infinigen has no garment system at all. We already generate a garment fitted BY MEASUREMENT (18 body cross-sections, ease margin) and a groom rooted by area-weighted scalp sampling -- so for us non-penetration is achievable BY CONSTRUCTION rather than by hand-fixing each asset. Done: garment and hair placement produced by a constraint solve over the measured body (clearance as a solved constraint, not an authored offset), such that GR20 reports ZERO garment-through-body and hair-through-body penetrations across a declared pose set for an arbitrary genome -- including bodies at the extremes of GR24's parameter space, which is precisely where an authored library fails | Organ | 2 u |
| Physically-based sky and aerial perspective (GR26) | gpe_sky_atmos | The sky is a two-stop vertical colour ramp today (recipe spec rows 7 skytop and 8 skyhor) with the EH3 hash cloud sheets over it -- a gradient, not an atmosphere, and on the live beach it is roughly half the frame at default pitch. Single-scattering Rayleigh plus Mie with a sun disc, plus aerial perspective tinting distant geometry, in the same integer determinism as every other render rung. Ordered BEFORE GR27 because ground albedo is lit by sky irradiance, so a flat ramp caps how real any sand can look no matter how good its material is. Operator 2026-08-28 named it first of seven. | Organ | 1 u |
| Ground material: granular sand and wet-shore darkening (GR27) after GR26 | wc_mat_ground | EH1 fixes terrain SHAPE and EH2 fixes what GROWS; neither touches what the ground IS, so the shore reads as tan voxel paint. Granular high-frequency normal that survives grazing angles, darkening plus specular gain inside the wet band, slope-keyed dune banding. The one input it needs already exists: recipe spec row 66 carries the ABSOLUTE shoreline coordinate, so height-above-waterline is readable today with no new plane. Depends on GR26 for the irradiance that makes wet-vs-dry separate at all. | Organ | 1 u |
| Palm flora reaches the SERVED beach (GR28) after GR31 | wc_veg_mesh | RE-SCOPED 2026-09-01 (board owner). THE GROWER LANDED 2026-08-30 and this rung has been ranked #2 for it ever since: func wsp_palm_shape( is declared and CALLED at sh9 == 3 in nx_wasm_craft.nx, so the 2026-08-28 claim that only shapes 1 and 2 have growers is stale. The old contract name wc_tree_palm can never ship because the capability shipped under a different name on gameengine -- repointed, not left to rot. WHAT IS OPEN is the half this rung always also named ("one grower plus ONE RECIPE ROW") and it is deliberately reverted, not forgotten: knowledge/world_recipes.conf carries the beach row with NO spec pairs on purpose, because both densities were judged on REAL-BROWSER frames -- 5:64 an opaque green canopy roof, 5:400 a forest of solid green cubes at eye level, because frond rosettes render as filled voxel crowns. DONE-RULE, taken verbatim from the conf's own revert record rather than invented here: palms return to the beach only when the crown renders as fronds (or the world leaves voxels) -- so this rung now DEPENDS on GR31, and its proof is a captured frame of the SERVED beach graded by the world referee (GR30), never the presence of a grower. CONTRACT NAME ALIGNED to gameengine.matrix the same hour: lane 3 split the same capability onto _ABSENT_:wc_veg_mesh minutes before I named it wsp_beach_flora_bind, and shipping both would have manufactured a third two-names-one-capability defect on the very board being corrected for two of them. Their arithmetic also supersedes the conf's "retune the density" reading: 1 block = 1 m, a frond is a chain of at most four 1 m cubes, so a crown is ~33 cubic metres of solid green against a real rachis of ~0.05 m -- no sparsity value makes a blob stop being a blob, and the exit condition is sub-block crown geometry, i.e. GR31. Operator 2026-08-28: no infinigen generated sand or palm trees. | Organ | 1 u |
| Skin micro-surface on the path the family looks through (GR29) | gpe_skin_microsurface | SCOPE SETTLED 2026-08-28 BY MEASUREMENT, after two wrong diagnoses of mine that the instruments corrected. NOT missing maps: TEXM is PRESENT, floor gate 7/7, 721,830 B, GL tier samples it. NOT a missing pore model: nx_relief_lib owns a cited three-band law and the baker already consumes it. THE DEFECT IS RESOLUTION, and nx_relief_bands_gate proves it -- a whole-body atlas spans 1.7m, so at 4096 the texel is 415um and secondary and pore both read carried=0; the live cast ships at 2048. Bands are computed then clamped, leaving a coarser band wearing the fine band's name, which is exactly what reads as painted. RES NEEDED: secondary 13600 whole-body, but over a 250mm face only 2000, and pore only 625. SO THE RUNG IS A PER-REGION ATLAS, not a bigger whole-body one -- the resolution we already ship carries the pore band once the footprint is a face rather than a body. Done rule: pore carried=1 at shipping resolution on a region atlas, with the neg-control that the whole-body atlas still reports 0 rather than being quietly redefined. Done rule: the served page shows measurable micro-relief at skin scale under the referee panel, with the neg-control that a flat-albedo build FAILS the same tooth. | Organ | 1.5 u |
| World-quality referee built and wired (GR30) after GR1 | wqg_build_wired | THE WORLD HALF OF GR1 AND IT WAS NEVER ON THIS PLAN. GR1 admitted a CHARACTER referee panel and the world lane still has none, so every world fidelity claim on this board is graded by the author's own eye. MEASURED 2026-09-01 by nx_catalog: nx_world_quality_grader is SOURCE yes 37940 B, BUILT NO, PROMOTED NO, REGISTERED NO, INVOKED NO, verdict SOURCE-ONLY (S2). The consumer already exists in prose -- nx_layer_verdict.nx carries an adapter for its 12-i64 output and names _wqg_grade_from_8, and _ft_diag.nx imports the organ -- so this rung is a BUILD and a WIRE, not a design. Done when the organ builds, promotes, registers, returns a grade plus 8 axis scores for a named world capture, and at least one world rung's done-rule reads that grade instead of a sentence. Carries the GR1 discipline: floors declared before the experiment, FP and FN measured on a labelled corpus, and the panel refuses rather than acquits when it cannot see | Organ | 1.5 u |
| A second art style of emission beside the voxel raycast (GR31) | wc_style_surface | OPERATOR 2026-08-30: everything is voxel and we dont need it to be that should just be one art style of emission. MEASURED 2026-09-01 on the served page: the beach WASM module exports 78 functions and NONE of them names a mesh, surface, rig, face, hair or skin producer -- the exports are simulation state and the world is drawn by a per-pixel integer DDA raycast over a voxel field, so voxel is the only style the engine can emit rather than the one it chose. The extractor is already built and graded elsewhere on the estate: nx_csg_extract, marching tetrahedra on the Kuhn six-tetrahedron split, watertight by construction, LANDED on both procgen and dcc. Done when the same worldgen field emits as a surface mesh through that incumbent extractor with the style selected by a recipe row and never by an engine fork, and with the voxel path proven byte-identical when the row is absent -- a second style, not a replacement | Organ | 2 u |
| The sovereign strand organ reaches a live world (GR32) after GR15 | wc_strand_bind | GR15 buys a strand SYSTEM at 2 u as if from nothing. MEASURED 2026-09-01 by nx_catalog: nx_nxa_strands is SOURCE yes 9565 B, BUILT NO, PROMOTED NO, INVOKED NO, verdict SOURCE-ONLY (S2) -- 9565 B of strand source sits unbuilt beside a rung that does not know it exists, so the plan overstates its own cost and hides that the first move is a build. This rung is deliberately the BINDING and not the organ: building it alone would leave the page-side groom the operator is looking at untouched. Done when a strand set generated by the sovereign organ reaches a live world through an engine export and the page copies buffers rather than authoring geometry, which is the same bar gameengine GE21 wc_hair_tick and GE22 gpe_wasm_hair_bind already set for the guide chain | Organ | 1 u |
| Strand chain ordered root-to-tip (GR33) | hg_seg_axis | THE CHEAPEST VISIBLE FIX ON THIS BOARD AND IT IS ONE TOKEN. The served page derives a strand segment index from component 1 of the bind position while this mesh space is Z-UP, which the served shader itself declares (vec3 yup=vec3(sp.x,sp.z,-sp.y), beach.html:292) and which the fit pass ten lines above the bug already honours. The chain lag and its squared weight are therefore spread front-to-back across the head instead of root-to-tip down each strand, which is both the mop shear the operator sees and the mechanism that drives hair through the face. DONE-RULE, pointed at a gate and not at a declaration: nx_gpe_hair_geometry_gate carries hg_seg_axis, a tooth asserting the segment index is derived from the up axis the shader declares rather than from a hard-coded component, bite-proven RED against the current served expression and GREEN against the fix, and the gate is listed on graphics.gates so the verdict is executed evidence rather than prose | Organ | 0.5 u |
| The normal map already inside the asset is read (GR34) | gpe_texm_normal_bind | The served ref9e.nxa carries a four-map TEXM (albedo, spec, gloss, NORMAL; NTB_MAP_NORMAL=3) and the page reads record 0 at both TEXM call sites, so a normal map ships in the file the browser already downloads and nothing samples it. This is layer one of a three-layer consumer gap and is deliberately scoped to layer one. DONE-RULE: the page binds map record 3 as a normal map on the character pass; nx_nxa_texm_gate (re-run 2026-09-01, passed 12/12 verdict=GREEN) stays green so the carry is unbroken, and a new tooth shows the lit result differs from the record-0-only build on the same frame -- a bind that changes no pixel is not a bind | Organ | 0.5 u |
| A declared asset ladder, so finished work reaches the cast (GR35) after GR34 | gpe_asset_ladder | Roughly 80 MB of finished character assets are orphaned: ref9f.nxa (40448728 B, ref9e plus TEXR:3590072, every other section byte-identical) and ref9d_r2.nxa (39478840 B) are referenced by NOTHING in buildroot, measured with corpus_complete=1, while the page serves ref9e. The mechanism already exists (a ?npcasset= override and a ref9e/ref9d/ref9 fetch chain), so this is a declaration, not a loader. DONE-RULE: the ladder is a conf row naming the richest asset a build will serve, the emitter refuses by name when a named asset is absent rather than silently falling back, and nx_asset_floor_gate passes on the asset actually selected -- proven by the emitted page naming the richer asset and by a capture whose byte count differs from the ref9e build | Organ | 0.5 u |
| FACE and MORF are consumed, so the face is geometry and not paint (GR36) after GR30 | gpe_morf_consume | The face is painted in a fragment shader from head-height ratios with NO NOSE TERM at all, while the served asset carries FACE:4700 and MORF:98634 and the page three NXA consumers dereference 15 tags that include neither. The producer nx_nxa_morf is LIVE and its consumer was never written; 7 of 9 face organs are REGISTERED-DARK. CARRIES ITS OWN HISTORY AS A PRECONDITION: face relief v1 was reverted 2026-08-03 for goblin noses because this mesh normals are majority-wrong, so the DONE-RULE is (a) the normal precondition is measured and stated before any relief is applied, (b) FACE and MORF are dereferenced by the page consumers, and (c) the referee panel grades the result at head framing at or above the painted baseline -- a rung that only makes the nose exist has reproduced 2026-08-03 | Organ | 2 u |
| The JS-logic ratchet is pinned against class laundering (GR37) | jl_class_pin | The ruler that enforces the estate loudest operator order is editable by the party it measures: knowledge/jslogic_class.conf maps each literal hash to a CLASS token, and moving a literal from LOGIC to INTEROP or SHADER removes its bytes from logic_unknown with no code change and no re-baseline. The conf self-invalidates on a literal EDIT (the hash moves, the row reads UNKNOWN) and NOT on a class-token edit, so the one mutation that launders the number is the one mutation it cannot see -- measured: the headline improved 228377 to 128559 on 2026-08-31 while zero JS bytes were removed and the two journal rows total 229911 identically. DONE-RULE: the census reads a declared digest of the class column and REFUSES by name when a class moved without the baseline being re-declared, with TWO neg-controls -- an ordinary literal edit must still self-invalidate the way it does today, and a class move WITH a re-declared baseline must pass. Scoped away from nx_jslogic_census_gate, which a sibling lane wrote the same day for the uncharged-JS axis and which is SOURCE-ONLY: extend that gate, never add a second ruler beside it | Organ | 1 u |
| The character judge survives the frame it is pointed at (GR38) | cjc_frame_guard | The world lane has no working referee for its own flagship frame: nx_charjudge takes SIGSEGV signal 11 at 0x41510f touching 0x0 on the beach capture and nx_addr2line refuses because the elf was built without -g, while nx_percept anchor corpus is entirely NXFH1 400x240 so a 1920x1200 capture is not comparable to it. The one number the lane CAN produce is the honest headline of this board: nx_frame_score on the native beach capture reads quality=408 verdict=BASIC against the ruler own published random-squares control of 628 RICH. DONE-RULE: the judge grades a full-resolution frame or ABSTAINS by name and never dies, a debug build exists so the next crash is locatable, and the abstention is a THIRD state that can neither acquit nor convict | Organ | 1 u |
| A character crop that contains a character (GR39) | ws_char_crop_nonempty | nx_world_snap char reports CHAR-FRAME framed=1 ground_y=11 and then CHAR-CROP pixels=0, so framing succeeds and the crop returns no subject image and every downstream character judgement about a world capture is made about an empty picture or not at all. DONE-RULE: a tooth asserts pixels>0 whenever framed=1, the crop refuses by name rather than emitting an empty image, and the hand-rolled NXFH1 writer at nx_world_snap lines ~515-555 is replaced by the shared codec nxl_dump so the estate keeps ONE frame writer -- the duplicate-ruler half is part of this rung, not a separate one | Organ | 0.5 u |
| The board ranker measures every row (GR40) | rk_measure_all | THE META-RUNG, and it gates the trustworthiness of every other number on this page. nx_compare_rank sets mpres=1 for any row not spelled _ABSENT_ and then does if mpres[mi]==1 { done=1 } else { done=rk_measure(...) }, so it measures only the rows already spelled absent and trusts every row spelled present: a rung can be promoted open to done by measurement and can never be demoted done to open, which makes a false green invisible by construction and computes the critical path on corrupt input. The emitter does not share the defect (it calls sd_present on both branches), which is why one row read not-measured on the served page and #1 FINISH in the ranker on the same day. DONE-RULE: the ranker measures both branches; a bite proves a present-spelled row with an absent symbol flips done to open; and because the real risk is a silent fleet-wide re-baseline, the run must publish the before/after done count for every domain in the same output | Organ | 1 u |
| The review viewers gain the sovereign wasm tier (GR41) | nre_sov_tier | OPERATOR 2026-09-02 (rigview printed NXA REFUSED no webgl2): nx_nxa_rig_emit, nx_nxa_view_emit and nx_nxa_stream_emit have exactly one renderer, WebGL2, and a refusal leaves a review instrument blank. gameengine GE28 makes the refusal name its cause; this rung removes the dependency. The estate already owns the pieces: nx_trimesh (the sovereign textured triangle rasteriser ntb_view uses server-side) and nx_meshview_wasm (the wasm memory model, 64x64 and 256 verts today). Done when a wasm build of the trimesh rasteriser sized for a 14k-vertex rig renders the same NXA the GL tier renders, skinned from the same SKEL and SKIN sections, and the rigview page falls to it with the badge naming the tier -- proven by a headless capture whose figure crop grades on cjc_panel inside the GL tier's band | Organ | 2 u |
| The beauty tier's first frame: the beach scene rendered offline by a sovereign path tracer and published beside the live frame with its referee score (GR42) after GR26,GR10 | env_beauty_scene | OPERATOR 2026-09-02: get things Infinigen generating to photoreal. Infinigen renders offline (Cycles path tracing, minutes per frame); its twin here is the BEAUTY TIER the matrix already declares ABSENT on nx_gsplat (env_beauty_scene) with the GI contract on nx_gi_path (gp_bounce_gi, also ABSENT: no source exists). Done when (a) nx_gi_path.nx exists as a sovereign integer path tracer over a scene the procgen planes already produce (heightfield terrain from nx_worldgen with a step derived from the pixel footprint so no terracing, water, the physically-based sky of GR26, the cast meshes with their TEXM maps), Monte Carlo bounce with a seeded deterministic sampler so two runs are byte-identical, (b) env_beauty_scene assembles that scene from the same genome and recipe planes the live page reads, (c) the frame is emitted at 1920x1200 with depth, normal and semantic ground truth beside it (the dataset-factory half), (d) it is published on this board beside the live capture and the mirrored Infinigen reference frame, and (e) a referee that PASSES its own ordering ladder grades all three -- until that referee exists the cell reads UNGRADED, never GREEN. The live real-time tier then chases this frame rung by rung (GR26 sky, GR27 ground, GR14 linear light, GR34 normal maps, GR16 hair, GR29 skin) with the same referee. | Organ | 8 u |
| A ray-triangle bounding-volume hierarchy for the beauty tier: the path tracer hits the cast meshes (GR43) after GR42 | gp_bvh_ray_tri | MATH GAP MEASURED 2026-09-03 (nx_capsearch over 7245 organs, corpus_complete=1: nx_bvh_lib is BioVision mocap, no ray acceleration structure exists by name), so nx_gi_path marches a heightfield and cannot hit a triangle; done when a BVH over the cast's NXA triangles (median split on integer centroids, deterministic from the mesh) is traversed by the path tracer, a brute-force every-triangle oracle agrees on every ray of a fixture (same nearest hit, same barycentrics), the build and traverse costs are printed per frame, and a neg-control with one corrupted node bound is caught by the oracle | Organ | 2 u |
| A low-discrepancy sampler for the path tracer: Sobol with Owen scrambling, the blue-noise organ reused for the screen (GR44) after GR42 | gp_sobol_owen | MATH GAP MEASURED 2026-09-03: the tracer draws from a hashed LCG and no Sobol or Halton sampler exists by name (nx_bluenoise exists for screen-space masks); done when a deterministic Sobol sequence with per-pixel Owen scrambling drives every bounce, variance at equal spp on a fixture scene is measurably lower than the LCG with the number printed and the LCG kept as the control, two runs are byte-identical, and the sampler is one lib that both the offline tracer and the real-time reconstruction (performance PF4) read | Organ | 1.5 u |
| Spherical-harmonics irradiance from the physically-based sky for the real-time tier (GR45) after GR26 | sr_sh_irradiance | MATH GAP MEASURED 2026-09-03: no spherical-harmonics or irradiance-probe organ exists by name and the live tier lights with a two-stop ramp; done when the GR26 sky is projected to SH9 in NishiLang integer arithmetic (nx_fixq30_lib), the shader emitted by nx_wgsl from the same source evaluates it per normal, a reference integration agrees within a stated band on a fixture sky, and the frame is graded by the world referee against the ramp with the neg-control of the projection zeroed | Organ | 1.5 u |
| Cascaded shadow maps on the kernel and WebGPU tiers: the cast casts a shadow on the ground (GR46) after GR42 | rk_shadow_cascade | MATH GAP MEASURED 2026-09-03: no shadow-map or cascade organ exists by name (the served beach draws dark blobs under feet); done when the GE42 kernel renders a sun-depth map for N cascades chosen from a conf ladder row, the world pass samples it with a bias DERIVED from the texel footprint, a fixture with a known occluder produces the analytically expected shadow footprint within a band, and the served beach frame is graded by the world referee with the maps forced off as the neg-control | Organ | 2 u |
| Integer Cholesky for the correlated identity draw (GR47) | la_cholesky_q30 | MATH GAP MEASURED 2026-09-03: nx_matrix carries determinant and inverse to 3x3 and no Cholesky decomposition exists anywhere by name, so GR24's full-covariance draw has no primitive; done when a Q30 fixed-point Cholesky over nx_fixq30_lib arithmetic factors any SPD matrix up to the identity-space dimension, L times L-transposed reproduces the input within a stated Q30 band on random SPD fixtures, a non-SPD input is refused by name rather than producing garbage, and GR24 composes it for its draw (one ruler, never a second decomposition) | Organ | 1 u |
| The rendered strand is the collided strand: character geometry leaves emitted JS for NishiLang lowered to wasm (GR48) after gameengine:GE40 | gpe_char_geom_wasm | OPERATOR 2026-09-04, the loudest order restated as an absolute: printf-emitted foreign-language logic is not a thin shim we have to accept, it is hacky programming, and first byte up every rung until it HAS to translate is the only way anywhere in the estate. MEASURED THIS SESSION, and it is the ROOT CAUSE of the standing REJECT beach-hair-colour-wind: the hair a visitor sees is built in hand-written JavaScript inside a NishiLang string literal in nx_game_page_emit.nx -- Catmull-Rom subdivision, NXH_GUIDES 1200 by NXH_KIDS 6 child interpolation, per-strand lateral axis, taper, quad emission and UV generation, hundreds of lines of geometry maths running on the main thread at load. The COLLIDER is in wasm (nx_wasm_craft hg_capd2 hg_inside hg_push) and runs on 128 SIMULATED guides. So the collider structurally cannot see the geometry that is rendered, which is why a clamp shipped into that JS on this date changed nothing an operator could see. THREE CONSEQUENCES, each measured: the collider is correct and invisible; the JS is outside the compiler, outside every gate, and outside grep itself (nx_shelltool skips the long emitted lines and reports linetoolong, so absence over that code is UNPROVEN by construction); and a fix in that layer can only ever be a patch. DONE-RULE: the strand construction is NishiLang compiled through nx_compile_wat and lives beside the collider so there is ONE geometry owner, the collider resolves against the RENDERED strand set and a tooth asserts rendered equals collided rather than 128 versus 7200, the JS path is retired only after the wasm path reproduces its vertex buffer within a stated band on a fixture groom, and the referee panel grades a head-framed capture at or above the JS baseline. NOT DONE if the JS block merely shrinks: the acceptance test is that no geometry byte on the page is hand-authored. | Organ | 4 u |
| The translation ladder is declared and counted per target: zero hops on NishiOS and Nishi browser, minimum hops elsewhere, an API layer only where a third party mandates one (GR49) after gameengine:GE40,gameengine:GE44,GR37 | tl_translation_census | OPERATOR 2026-09-04: do everything first byte up in NishiLang and estate and engines, then ship to the translation layer if it requires one, but make it so that in NishiOS and Nishi browser we do NOT have a translation layer so we can be superior in performance and capabilities and delivery, and if third party ship via an API layer if mandatory -- the least possible translation is the best. THE ESTATE ALREADY ENFORCES THIS ON TWO AXES AND NEITHER COVERS THE THIRD: GE40 wp_provenance demands every served wasm byte trace to a named .nx, and GE44 rk_webgpu_door demands ZERO shader bytes on the page that do not trace to nx_shader_ir and reduces WebGPU to a submission pipe with no geometry pipeline. NOTHING MEASURES TRANSLATION HOPS PER TARGET, so the JS layer grew to roughly 128559 bytes of logic (GR37) and the character geometry ended up hand-authored inside it with nobody able to see the number rise. DONE-RULE: a conf ladder row per delivery target declares its permitted hop count and its permitted translation kinds, NishiOS-native and Nishi-browser-native declare ZERO with no wasm and no JS and no WGSL in their path, third-party-browser declares the minimum with each hop NAMED and justified by a mandate rather than by convenience, a census walks every served artifact and counts the hops it actually took, the count is a RATCHET that may not rise, and the census REFUSES by name when a target exceeds its declared ceiling. TWO NEG-CONTROLS: a hop added to a zero-hop target must fail, and a hop merely re-classified without being removed must also fail -- that second control is the GR37 laundering lesson applied here before it can happen rather than after. | Organ | 2 u |
| The ship loop adjudicates itself against the SERVED artifact and refuses to report success for a provably inert promote (GR50) | os_served_delta | OPERATOR 2026-09-04: the delivery pipeline must be self-adjudicating, a human should not be in this loop, and the math was clear as day from the first failed ship. MEASURED, and it is a three-instance class not a one-off. INSTANCE ONE: nx_organ_ship nx_game_page_emit ran the whole loop GREEN -- build, adoptgate with four lost runs adjudicated, promote 425241 to 451160 B, offc twin, register, verify -- and changed NOTHING a visitor can see, because the target is a LIBRARY whose standalone binary emits no served artifact. Everything needed to say so was one command away: grep for the import edge returns 3 importers with corpus_complete=1, nx_gpe_hair_geometry_gate nx_world_recipe_gate and nx_craft_emit, and only the last declares served outputs CE_OUT CE_OUT2 CE_OUT3. The loop never asked which served artifact it changes. INSTANCE TWO: the same loop reports verify against the BINARY, so a fresh page emitted by a 15.6 hour old writer reads as a successful delivery. INSTANCE THREE: /api/promote already reads organ_kind.conf and refuses daemons BY NAME, so kind data is on the path and a lib promote is permitted and SILENT. DONE-RULE, two teeth and both cheap: FIRST the loop resolves the target's served artifacts from the organ's own output declarations and, when the set is EMPTY, it NAMES every importer and REFUSES to report success, because shipping a lib alone is inert BY CONSTRUCTION rather than by bad luck; SECOND it hashes each served artifact before and after and a ship whose served bytes did not move is reported FAILED however green every stage was. TWO NEG-CONTROLS: a genuine writer ship whose page bytes DO move must still pass, and a lib ship must fail even when build prove adoptgate promote and register are all green. NOT DONE if it merely warns -- the failure mode being fixed is a loop that reported success, so the verdict must change, not the prose. | Organ | 2 u |
| The hair gate gains a tooth the broken output must fail: rendered strands inside the body are counted, not assumed (GR51) after GR48 | gpe_hair_penetration_tooth | MEASURED 2026-09-04 on the LIVE gate: nx_gpe_hair_geometry_gate reads passed 16 of 16 verdict=GREEN while the operator is looking at hair clipping through the chest, and BOTH statements are true because they are about different subjects. Every tooth it carries asks whether the pipeline is WIRED -- emitted-consumes-HVRT/HTRI-groom-MESH, emitted-consumes-HSTR-guide-strand-geometry, emitted-indexes-the-chain-table-BY-VERTEX-SEGMENT, emitted-UPLOADS-the-displaced-geometry-each-frame, emitted-ANNOUNCES-absent-hair-geometry-rather-than-reporting-success, and a neg-control that the superseded RIGID block is gone. NOT ONE tooth asks whether a rendered hair vertex is inside the body, so no possible penetration can fail this gate and its GREEN is evidence of a connected pipeline, never of correct hair. THIS IS THE SAME CLASS AS THE OTHER TWO FINDINGS TODAY: hg_tick is arithmetically correct and has NO CALLER across 23622 files with corpus_complete=1, and the ship loop verifies a binary rather than a served artifact -- three green signals, three real defects, and in every case the subject measured was narrower than the claim made. DONE-RULE: a tooth counts rendered strand vertices lying inside the body capsules of the same fixture the gate already builds, the count is EMITTED with gv_kv beside its bar so a second method can contradict it, the tooth is BITE-PROVEN by mutating the clamp away and seeing the count RISE and the gate go RED, and an anti-vacuity control asserts the fixture actually reached a pose where penetration is possible -- a fixture whose hair never approaches the body would pass this tooth for the wrong reason and is not a test. | Organ | 1.5 u |
| The estate's searcher can read the artifacts the estate serves (GR52) | st_longline_windows | OPERATOR 2026-09-04: invest in our tooling and do this properly rather than routing around broken instruments. MEASURED, FOUR DEFECTS HIT IN ONE SESSION AND ALL FOUR WORKED AROUND RATHER THAN FIXED: nx_shelltool grep declares linetoolong and SKIPS any line longer than its buffer, and the estate's primary shipped artifact -- an emitted world page -- is a handful of 200 KB lines, so the searcher is STRUCTURALLY BLIND TO EXACTLY WHAT WE SERVE and every absence claim over a served page is unprovable by it; the same verb takes a DIRECTORY and refuses a file path, which reads as the organ being missing; a directory scan reports files=200000, a ROUND DECIMAL that is a CAP printed in a field that reads as a count, and it was quoted as a population fact in this very session; and BUDGET-EXCEEDED returns corpus_complete=0 with matches=0, which is UNPROVEN and reads as absence. THE INCUMBENT IS WELL REASONED AND MUST NOT BE REPLACED: nx_shelltool.nx:200-210 explains that it reads in chunks and carries the trailing partial line so coverage is complete at any file size, and that it refuses to split a long line because half a line matched as a whole one is a FABRICATED match, the opposite failure and the worse one. That reasoning is right and the fix preserves it. DONE-RULE: at nx_shelltool.nx:250 the overlong branch stops discarding to end-of-line and instead scans the same line in OVERLAPPING WINDOWS carrying pn-1 bytes across each cut, pn-1 being exactly the largest pattern prefix that can straddle a boundary, so no match is missed and none is fabricated; a hit on an overlong line emits a WINDOW around the match rather than the whole line; linetoolong is retired as a coverage caveat and replaced by a longline= count that is INFORMATIONAL rather than a hole; files= and any other cap is renamed or accompanied so a ceiling can never be read as a census; and the whole thing is BITE-PROVEN by planting a pattern in the middle of a 200 KB line that the current binary MISSES and the fixed one FINDS, with a neg-control proving a pattern that straddles a window boundary is reported exactly ONCE. | Organ | 2 u |
| The estate reads its own served pages with its own wire: a sovereign CDP witness composed from the shipped WebSocket stack (GR53) after GR49 | nx_cdp_eval | OPERATOR 2026-09-04, asked of the Python browser driver this seat leaned on all day: is it a tool we need to duplicate into nishi lang and use as a 3rd witness. MEASURED ANSWER, by listing buildroot/runtime rather than estimating: IT IS COMPOSITION, NOT CONSTRUCTION. Already shipped and already tested here are nx_websocket_client_upgrade.nx, which performs the RFC 6455 client handshake on an already-connected TCP fd and validates the 101 status plus Upgrade, Connection and Sec-WebSocket-Accept against the SHA-1 and Base64 transform of the key it sent, with nx_websocket_client_upgrade_test.nx beside it and ten named verdict constants; and nx_websocket_frame.nx, giving ws_parse_frame, ws_build_header and ws_apply_mask with the full opcode set, with nx_websocket_frame_test.nx beside it. Underneath sit nx_json.nx, nx_json_emit.nx, nx_base64.nx, nx_rand.nx, nx_net.nx and nx_syscalls.nx. ac_cdp.py's own header calls the handshake and frame codec THE ROUGHLY FORTY LINES BELOW and treats them as the hard part; that part is done, in NishiLang, with tests. THE ONLY MISSING GLUE, named so nobody rebuilds what exists: connect TCP to the DevTools port, GET /json/version or /json/list to resolve the page's webSocketDebuggerUrl, run the client upgrade on that fd, send one Runtime.evaluate with returnByValue true, then read frames until the matching id and hand back the result verbatim. CDP to a local browser is plain ws, so no TLS is required and nothing else is needed underneath. WHY THIS IS WORTH A RUNG RATHER THAN A NOTE, measured today and it is the uncomfortable half: nx_shelltool grep SKIPS any line longer than its buffer and every emitted world page is a handful of 200 KB lines, so THE ESTATE'S OWN SEARCHER IS STRUCTURALLY BLIND TO THE ARTIFACTS THE ESTATE SERVES, and the ONLY instrument that could answer whether our own code reached our own page was a third-party browser driven by Python. Every false claim this seat published about that question traces to exactly that blindness. THE BUILD PATH IS SOVEREIGN AND THE EVIDENCE PATH IS NOT, and nobody was counting the second half. DONE-RULE: nx_cdp_eval opens a page, evaluates one expression and prints the result verbatim with zero judgement of its own, exactly as the Python wire does; a neg-control proves it REFUSES rather than returns empty when the endpoint is absent, when the upgrade is rejected, and when the evaluate returns an exception; and the acceptance test is that it reproduces, byte for byte, the answer ac_cdp.py returns for the same expression on the same served URL -- two witnesses agreeing is the point, so the Python one is KEPT as the oracle and never retired. SCOPED AWAY from nx_browser: rendering the page with our own engine is the THIRD witness and the zero-translation rung of the operator's ladder, a separate and larger piece; this rung is our WIRE against their ENGINE and is buildable on the LAPTOP toolchain, which matters because a brand-new file has no laptop-versus-NAS fork to converge and NAS build admission refused every lane for hours on the day this was scoped. | Organ | 1.5 u |
| ONE HAIR SYSTEM: five parallel implementations collapse to one, and this rung is measured by DELETIONS not additions (GR54) after GR48 | hair_one_system | OPERATOR 2026-09-04, and the evidence for it is THIS SEAT'S OWN WORK: this feels like you are just dumping shit everywhere because you have no plan and we have multiple systems in play so sometimes one system gets the work sometimes another and thats garbage and not sota or any level of good development. HE IS RIGHT AND THE COUNT IS FIVE, ALL TOUCHED IN ONE SESSION BY ONE SEAT: a soft-tissue term added to the collision proxy in nx_nxa_rig_emit which is the RIGVIEW shader path; a child-offset clamp added to nx_game_page_emit which is emitted JAVASCRIPT on the BEACH path; vm_capsule_push added to nx_vecmath which is a FIFTH copy of closest-point-on-capsule; hd_tick_collide added to nx_hairdyn_lib which is the BAKE-time integrator; and nx_haircollide_gate built 12 of 12 GREEN as a gate for hg_tick, WHICH HAS NO CALLER ACROSS 23622 FILES WITH corpus_complete=1. FIVE ADDITIONS TO FIVE SYSTEMS IN ONE DAY, INCLUDING A GATE FOR DEAD CODE AND A FIFTH DUPLICATE PRIMITIVE, while the same seat catalogued duplicate-system defects as though observing from outside. ★★★★★★A SEAT THAT FIXES WHICHEVER SYSTEM IT HAPPENED TO BE READING PRODUCES MOTION IN EVERY SYSTEM AND PROGRESS IN NONE, AND EVERY SUCH FIX MAKES THE NEXT CONSOLIDATION MORE EXPENSIVE. THE RUNTIME TRUTH, MEASURED: the hair a visitor sees is built at load in hand-written JS from the rest-pose asset section and moved per frame by nxHairTick -- spring, viscous damping, wind, length clamp, and NO BODY TERM OF ANY KIND. The wasm collider hg_capd2 hg_inside hg_push with HG_CAP_MAX=32 capsules and HG_SIM=128 guides is UNREACHABLE. So four of the five systems cannot affect a served pixel and the fifth is outside the compiler. DONE-RULE, AND IT IS COUNTED IN DELETIONS: closest-point-on-capsule exists exactly ONCE in the estate and the other four sites are retired through nx_retire_path with the survivor named here; hg_tick has at least one production caller and a bite proves the gate goes RED when collision is disabled, so the collider stops being a comment; the rendered strand set is built by NishiLang lowered to wasm beside that collider with the emitted-JS geometry RETIRED once the wasm path reproduces its vertex buffer within a stated band on a fixture groom; and nx_haircollide_gate's subject changes from a dead function to the shipping path or the gate is retired with it. THE ACCEPTANCE TEST IS A COUNT THAT FALLS: capsule implementations 5 to 1, hair integrators 2 to 1, geometry builders 2 to 1, and hg_tick callers 0 to at least 1. A rung that ADDS a sixth anything has failed by construction. SUPERSEDES the scattered pieces this seat landed today -- GR25's clamp becomes a property of the one system rather than a patch on the JS one, and GR48 is the geometry half of this rung rather than a separate lane. | Organ | 4 u |
| Skin-tone fidelity of the LIVE render is measured in Delta E and ITA error per class under named lights (GR55) | skin_tone_fidelity_dE | SEPT-2026 BAR, read from the mirrored paper: 19848 MetaHuman renders scored against Chicago Face Database references show fidelity collapsing for darker classes under non-reference lighting (class VI median ITA error 11.37 to 53.85) [truetotone26]. Our generator spans the ITA classes by construction (nx_skin_ita, gate T4), so the gap is MEASUREMENT of the rendered result, not generation. DONE-RULE: the ITA ladder is rendered through the live beach path under three named lights, each render is scored in Delta E CIE76 and ITA error against its reference, per-class medians are EMITTED with gv_kv beside the published medians, and a neg-control proves a deliberately desaturated render RAISES the error -- an anti-vacuity tooth asserts the fixture spans at least four ITA classes so a single-tone ladder cannot pass | Organ | 1.5 u |
| Our subsurface term is calibrated against a measured face BSSRDF and its RMSE is published (GR56) after GR55 | skin_bssrdf_calibration | SEPT-2026 ORACLE: a polarimetric multispectral capture of dynamic faces fitting spatially varying single scattering and refractive index, validated at a mean reconstruction error of 0.028 against known-index objects [polbssrdf25]. The oracle NEVER enters the build path: one measured BSSRDF is adopted as DATA with its license beside it, our skin model is fitted to it, and the RMSE of our fit against the measurement is the row's number. DONE-RULE: the fit RMSE is emitted with gv_kv, a neg-control proves a flat-albedo control fits WORSE by a named margin, and the fitted parameters are the ones the live path reads (a fit that never reaches the page is a comment) | Organ | 2 u |
Ladder to SOTA
| Rung | Role | Toward | Stands on | Why |
|---|---|---|---|---|
| GR1 | contender | metahuman-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR2 | contender | metahuman-2026 | GR1 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR3 | contender | metahuman-2026 | GR2 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR4 | contender | infinigen-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR5 | contender | infinigen-2026 | GR4 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR6 | contender | infinigen-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR7 | contender | infinigen-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR8 | contender | siggraph-rtr-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR9 | contender | siggraph-rtr-2026 | GR8 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR10 | contender | siggraph-rtr-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR11 | contender | siggraph-rtr-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR12 | contender | metahuman-2026 | GR2 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR13 | contender | metahuman-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR14 | contender | siggraph-rtr-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR15 | contender | siggraph-rtr-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR16 | contender | siggraph-rtr-2026 | GR15 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR17 | contender | siggraph-rtr-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR18 | contender | metahuman-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR19 | contender | siggraph-rtr-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR20 | contender | metahuman-2026 | GR19 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR21 | contender | metahuman-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR22 | contender | infinigen-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR23 | contender | siggraph-rtr-2026 | GR14 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR24 | contender | metahuman-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR25 | contender | infinigen-2026 | GR20 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR26 | contender | siggraph-rtr-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR27 | contender | infinigen-2026 | GR26 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR28 | contender | infinigen-2026 | GR31 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR29 | contender | metahuman-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR30 | contender | siggraph-rtr-2026 | GR1 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR31 | contender | infinigen-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR32 | contender | infinigen-2026 | GR15 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR33 | contender | metahuman-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR34 | contender | metahuman-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR35 | contender | infinigen-2026 | GR34 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR36 | contender | metahuman-2026 | GR30 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR37 | contender | metahuman-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR38 | contender | metahuman-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR39 | contender | metahuman-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR40 | contender | metahuman-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR41 | contender | siggraph-rtr-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR42 | contender | siggraph-rtr-2026 | GR26,GR10 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR43 | contender | siggraph-rtr-2026 | GR42 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR44 | contender | siggraph-rtr-2026 | GR42 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR45 | contender | siggraph-rtr-2026 | GR26 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR46 | contender | siggraph-rtr-2026 | GR42 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR47 | contender | metahuman-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR48 | contender | siggraph-rtr-2026 | gameengine:GE40 OPEN | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR49 | contender | siggraph-rtr-2026 | gameengine:GE40 OPEN,gameengine:GE44 DONE,GR37 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR50 | contender | siggraph-rtr-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR51 | contender | siggraph-rtr-2026 | GR48 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR52 | contender | siggraph-rtr-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR53 | contender | siggraph-rtr-2026 | GR49 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR54 | contender | siggraph-rtr-2026 | GR48 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR55 | contender | siggraph-rtr-2026 | - | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| GR56 | contender | siggraph-rtr-2026 | GR55 | Declared comparison target; acceptance remains unverified until the rung-specific receipt is produced |
| Rung | Closes with | Definition of done (pre-declared) | Executor | Est. |
|---|---|---|---|---|
| Sheer and stretched fabric over skin (GR57) after GR42,GR14,skintwin:ST5,procgen:PG29 | _ABSENT_:gr_sheer_bsdf | The reference-still class this rung answers (2026-09-27) wears a sheer swimsuit: the garment is a THIN LAYERED MEDIUM, not an opaque surface. Reflectance core is skintwin ST5 km_layer_over (Kubelka-Munk two-flux layering of fabric over skin); transmission RISES with weave strain read from the drape solver (procgen PG29 gm_xpbd) and with wetness (GR58); the skin term is visible through the fabric in the beauty tier and the live tier. The field's bar this month: Eurographics 2026's texture-free woven-fabric model and neural garment fold inter-reflection, the 2024 real-time neural woven fabric and reflection-transmission capture lineage; nothing this month renders sheer fabric OVER skin with strain-driven transmission, which is the exceed axis. Done-rule: a rendered sheer swatch over a skin patch matches the two-flux prediction within a DECLARED DE2000 band at three strains, and the beach garment shows the skin term through it under the sky. [kesen_eg2026] | Organ | 4 u |
| Wet as a surface state on skin, hair and sand (GR58) after skintwin:ST7,skintwin:ST9,GR27,GR16 | _ABSENT_:gr_wet_state | One wetness FIELD drives three materials: skin darkens and its specular sharpens (skintwin ST7 fo_wetness), hair clumps by capillary adhesion and its cuticle reads index-matched (skintwin ST9), sand darkens by the wet-porous term (GR27's wet-shore darkening generalised). The estate's wet-appearance scan read 2026-09-27 (11 entries, newest 2025-12, a homonym) holds no 2026 wet-appearance paper; the standing references are the micro-ellipsoid wet-porous model (2401.15628) and Jensen's 1999 wet-materials model, so this axis has NO current-month bar and says so. Done-rule: dry-to-wet DE2000 and specular-lobe shift published per material against mirrored measurements, and the field is one row of data the beach scene reads. | Organ | 3 u |
| Sun-tan and tan lines as a melanin field (GR59) after GR55,skintwin:ST6,skintwin:ST1 | _ABSENT_:gr_melanin_field | The skintwin board's own goal names a tan line as an ABSORBER change, never a pixel edit. Melanin concentration is a per-vertex field DERIVED from garment coverage history and declared sun exposure (data rows), read by the skintwin chain (sf_fit, km_r_from_ks) and relit correctly by construction; the ISO/DTR 26478 skin-tone patches and NIST's 100 traceable skin reflectance spectra (250-2500 nm) are the calibration oracles skintwin ST6 pins. Done-rule: a tan line follows a garment edge, its inside-outside ITA difference lands in the band the skintwin referee (ST1) reports, and a relight moves both sides consistently. | Organ | 2 u |
| The support surface and its cover (GR60) after softbody:SB15,softbody:SB13,softbody:SB10,dcc:DC10,procgen:PG29 | _ABSENT_:gr_support_cover | The reference class reclines on a cushioned lounger: the cushion COMPRESSES under the body (softbody SB15, a support surface that spreads load) and its cotton cover DRAPES and wrinkles (procgen PG29 cloth XPBD, dcc DC10 pattern-sew-drape), with body-cushion contact free of interpenetration (softbody SB10; S4R-class resolution 2609.20524 is the September reference for procedurally assembled scenes). Done-rule: contact area and sink depth reported against the tissue and cushion stiffness rows, a non-trivial wrinkle field on the cover, zero interpenetration at the declared contact distance. [arxiv_co_siga2026] | Organ | 4 u |
| A camera model as data (GR61) after GR14,GR23,GR42 | _ABSENT_:gr_camera_row | A photograph is the SCENE TIMES A CAMERA. Exposure, sensor response and tone curve, focal length, aperture and depth of field, vignetting, flare and chromatic aberration become ONE camera row that the beauty tier (GR42) and the live tier read identically, on the linear-light pipeline (GR14) through the one colour transform (GR23). Done-rule: a rendered grey card and a photographed one agree within a DECLARED DE2000 band under the declared exposure, and depth of field matches a measured circle of confusion at two distances. | Organ | 3 u |
| Reference-still referee: per-region DE2000 against a licence-clean photograph class (GR62) after skintwin:ST1,GR55,GR30 | _ABSENT_:gr_refstill_referee | A reference photograph CLASS (licence-clean stills under the refintake lane's licence state, never an identity, never an unlicensed image) is measured per region in CIELAB: skin ITA, sand, fabric, cushion, sky. Every beach frame publishes its per-region DE2000 against that class beside the world referee grade (GR30), through the skintwin CIEDE2000 ruler (ST1 cs_de2000, built and promoted, ZERO adopters outside the garment lane). Done-rule: ST1 wired to the beach page, bands from measured references, the number on the page. This is the axis the operator's 2026-09-27 goal turns on: a real photo benchmarked, a published frame, no visible difference. | Organ | 2 u |
| Real-versus-render indistinguishability bench with outside liar-killers (GR63) after GR62,GR61,GR42,gameengine:GE51 | _ABSENT_:gr_realrender_bench | THE ACCEPTANCE RULER FOR THE OPERATOR'S GOAL (2026-09-27: benchmark a real photo, publish our frame, and not tell the difference). Pairs of licence-clean real stills and our frames of the same scene class (a synthetic human, never a pictured identity) are judged by (a) outside synthetic-image detectors run OUTSIDE the build path (NVIDIA's SIGGRAPH 2026 synthetic-video detector NIM is the named first oracle), (b) outside VLM judges through the referee panel, and (c) a human two-alternative forced choice by the operator and the team. The score is detection accuracy driven toward chance (500 permil) with the sample size declared, sequential looks corrected and an A/A calibration (gameengine GE51's arithmetic). Done-rule: a published bench page on /compare with the three arms, receipts and the receipt's date; the rung reads INCOMPLETE until all three arms exist and the operator's ACCEPT row is on the plane. [nvidia_siggraph2026_blog] | Organ | 3 u |
| Rung | Role | Toward | Stands on | Why |
|---|---|---|---|---|
| GR57 | contender | siggraph-rtr-2026 | GR42,GR14,skintwin:ST5 DONE,procgen:PG29 DONE | the sheer-over-skin BSDF is a rendering claim measured against the venue bar for fabric appearance |
| GR58 | contender | siggraph-rtr-2026 | skintwin:ST7 DONE,skintwin:ST9 OPEN,GR27,GR16 | wetness as one field over three materials has no current-month bar; the rung declares that honestly |
| GR59 | contender | metahuman-2026 | GR55,skintwin:ST6 DONE,skintwin:ST1 DONE | a derived melanin field is the anatomy-derived twin of MetaHuman's painted skin |
| GR60 | contender | infinigen-2026 | softbody:SB15 DONE,softbody:SB13 DONE,softbody:SB10 DONE,dcc:DC10 OPEN,procgen:PG29 DONE | the cushion and cover are scene instances generated and simulated, the procedural claim |
| GR61 | contender | siggraph-rtr-2026 | GR14,GR23,GR42 | the camera row is a substrate every photoreal claim needs and a rendering claim in itself |
| GR62 | arm | metahuman-2026 | skintwin:ST1 DONE,GR55,GR30 | the per-region referee triangulates without competing for a number |
| GR63 | contender | metahuman-2026 | GR62,GR61,GR42,gameengine:GE51 DONE | indistinguishability from a real photograph IS the parity claim against the best-in-class human |
Ladder verdict, counted (17 rows, more than the 12-row disclosure bar)
| Class | Count | Note |
|---|---|---|
| targets | 3 | sotatarget rows on the plan |
| best in class | 2 | targets whose class is best-in-class, by code |
| frontier | 1 | targets whose class is frontier, by code |
| undated | 0 | targets naming no bar that exists |
| bad class | 0 | targets whose class is neither word |
| duplicate | 0 | target ids an earlier row already carries |
| rungs | 63 | rung rows on the plan |
| placed | 63 | rungs carrying a rungrole row |
| unplaced | 0 | rungs with no role toward any target |
| bad role | 0 | rungrole rows naming a role the ladder does not know |
| bad target | 0 | rungrole rows naming a target no sotatarget row declares |
| rungs, summed | 63 | RECONCILED: the parts sum to the total 63 |
| substrate | 0 | what the target's own methods consume |
| arm | 1 | a sovereign arm that triangulates without competing for the number |
| contender | 62 | its done-rule is a number against the target |
| superseded | 0 | kept as capability, retired as a claim |
| placed, summed | 63 | RECONCILED: the parts sum to the total 63 |
| Class | Count | Note |
|---|---|---|
| bars | 7 | sotabar rows on the plan |
| fresh | 0 | the bar month is the current month |
| attested | 7 | an older bar with a this-month barscan reporting no newer entry |
| stale | 0 | a this-month scan saw a newer month the bar ignores |
| unattested | 0 | nobody looked this month |
| malformed | 0 | a sotabar row missing a field |
| duplicate | 0 | bar ids an earlier row already carries |
| bars, summed | 7 | RECONCILED: the parts sum to the total 7 |
| current month | 2026-10 | the month every bar is judged against |
| Class | Count | Note |
|---|---|---|
| rows | 1 | treat rows declared on the plan |
| complete | 1 | every declared step verifies against the evidence |
| partial | 0 | a declared step is unmet |
| malformed | 0 | a treat row missing a field |
| rows, summed | 1 | RECONCILED: the parts sum to the total 1 |
| rungs | 63 | rungs on the plan |
| undeclared | 62 | rungs with no treat row |
| allowlist bytes read | 207162 | the MCP allowlist the ranker reads |
| roster bytes read | 119881 | the gate roster |
| clock plane bytes read | 0 | the clock DESIRED plane |
| Rung | MCP row | API route | Primitive | Workflow | Agent beat | Verdict |
|---|---|---|---|---|---|---|
| GR35 | nx_craft_emit present | /api/organ_run present | nx_artifact_root_gate present | component-release-preparation-candidate-t343 present | nx_wasm_craft_gate present | COMPLETE |
Experiments
Experiment GR62-E1: Does the two-chromophore skin-model mask beat the YCbCr rule box by box against the outside-oracle boxes of private reference private-ref:6a8ea809, on recall over the skin boxes and on false admission over the skin-coloured negative boxes?
Design ab · privacy private · control arm ycbcr · varied factors mask
Every verdict below is corrected for a family of 2 contrasts (2 arm-against-control; Bonferroni over the incumbent table, two-sided, five percent family-wise), and a contrast with fewer units per arm than the incumbent sample floor of 30 reads INSUFFICIENT rather than a guess. Intervals are the arm's own normal-approximation interval and are descriptive below that floor.
| Arm | Levels | Metric | n | Mean | 95 percent interval | Delta vs control | Corrected verdict | P(beats control) | Bayes |
|---|---|---|---|---|---|---|---|---|---|
| ycbcr (control) YCbCr rule (the incumbent the GR62 skin model replaces) | mask=ycbcr,boxes=private-ref:6a8ea809 | recall per skin box, permil of the box pixels admitted as skin (recall, mean, higher is better) | 4 | 950.0 | [899.6, 1000.4] | - | control | - | - |
| ycbcr (control) YCbCr rule (the incumbent the GR62 skin model replaces) | mask=ycbcr,boxes=private-ref:6a8ea809 | false admission per negative box (sand, rock, towel and hair, the oracle file's own definition), permil of the box pixels admitted as skin (falseadmit, mean, lower is better) | 6 | 476.8 | [99.4, 854.2] | - | control | - | - |
| skinmodel two-layer two-chromophore skin-model mask (nx_refstill_lib) | mask=skinmodel,boxes=private-ref:6a8ea809 | recall per skin box, permil of the box pixels admitted as skin (recall, mean, higher is better) | 4 | 841.5 | [660.1, 1022.9] | -108.5 | INSUFFICIENT (the control has fewer units than the sample floor) | - | a mean metric: no closed form |
| skinmodel two-layer two-chromophore skin-model mask (nx_refstill_lib) | mask=skinmodel,boxes=private-ref:6a8ea809 | false admission per negative box (sand, rock, towel and hair, the oracle file's own definition), permil of the box pixels admitted as skin (falseadmit, mean, lower is better) | 6 | 185.3 | [36.9, 333.8] | -291.5 | INSUFFICIENT (the control has fewer units than the sample floor) | - | a mean metric: no closed form |
Experiment GR62-E2: How do the skin share and the mean skin colour (CIELAB L*, a*, b*) move with the mask (the YCbCr rule against the skin model) across the four admitted beach references?
Design factorial · privacy private · control arm ycbcr-d421bd74 · varied factors mask,ref
| Arm | Levels | Metric | n | Mean | 95 percent interval | Delta vs control | Corrected verdict | P(beats control) | Bayes |
|---|---|---|---|---|---|---|---|---|---|
| ycbcr-d421bd74 (control) YCbCr rule on private reference private-ref:6a8ea809 | mask=ycbcr,ref=private-ref:6a8ea809 | skin share of the frame, permil (share, mean, descriptive, two-sided) | 1 | 518.0 | - | - | control | - | - |
| ycbcr-d421bd74 (control) YCbCr rule on private reference private-ref:6a8ea809 | mask=ycbcr,ref=private-ref:6a8ea809 | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 1 | 62014.0 | - | - | control | - | - |
| ycbcr-d421bd74 (control) YCbCr rule on private reference private-ref:6a8ea809 | mask=ycbcr,ref=private-ref:6a8ea809 | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 1 | 4075.0 | - | - | control | - | - |
| ycbcr-d421bd74 (control) YCbCr rule on private reference private-ref:6a8ea809 | mask=ycbcr,ref=private-ref:6a8ea809 | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 1 | 12912.0 | - | - | control | - | - |
| skinmodel-d421bd74 skin model on private reference private-ref:6a8ea809 | mask=skinmodel,ref=private-ref:6a8ea809 | skin share of the frame, permil (share, mean, descriptive, two-sided) | 1 | 369.0 | - | -149.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-d421bd74 skin model on private reference private-ref:6a8ea809 | mask=skinmodel,ref=private-ref:6a8ea809 | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 1 | 49465.0 | - | -12549.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-d421bd74 skin model on private reference private-ref:6a8ea809 | mask=skinmodel,ref=private-ref:6a8ea809 | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 1 | 8086.0 | - | +4011.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-d421bd74 skin model on private reference private-ref:6a8ea809 | mask=skinmodel,ref=private-ref:6a8ea809 | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 1 | 10735.0 | - | -2177.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ycbcr-1fc2e699 YCbCr rule on private reference private-ref:06caaa3c | mask=ycbcr,ref=private-ref:06caaa3c | skin share of the frame, permil (share, mean, descriptive, two-sided) | 1 | 142.0 | - | -376.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ycbcr-1fc2e699 YCbCr rule on private reference private-ref:06caaa3c | mask=ycbcr,ref=private-ref:06caaa3c | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 1 | 55505.0 | - | -6509.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ycbcr-1fc2e699 YCbCr rule on private reference private-ref:06caaa3c | mask=ycbcr,ref=private-ref:06caaa3c | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 1 | 8505.0 | - | +4430.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ycbcr-1fc2e699 YCbCr rule on private reference private-ref:06caaa3c | mask=ycbcr,ref=private-ref:06caaa3c | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 1 | 10220.0 | - | -2692.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-1fc2e699 skin model on private reference private-ref:06caaa3c | mask=skinmodel,ref=private-ref:06caaa3c | skin share of the frame, permil (share, mean, descriptive, two-sided) | 1 | 146.0 | - | -372.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-1fc2e699 skin model on private reference private-ref:06caaa3c | mask=skinmodel,ref=private-ref:06caaa3c | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 1 | 49390.0 | - | -12624.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-1fc2e699 skin model on private reference private-ref:06caaa3c | mask=skinmodel,ref=private-ref:06caaa3c | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 1 | 9448.0 | - | +5373.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-1fc2e699 skin model on private reference private-ref:06caaa3c | mask=skinmodel,ref=private-ref:06caaa3c | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 1 | 9685.0 | - | -3227.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ycbcr-4d0f86f7 YCbCr rule on private reference private-ref:ff3b177c | mask=ycbcr,ref=private-ref:ff3b177c | skin share of the frame, permil (share, mean, descriptive, two-sided) | 1 | 408.0 | - | -110.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ycbcr-4d0f86f7 YCbCr rule on private reference private-ref:ff3b177c | mask=ycbcr,ref=private-ref:ff3b177c | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 1 | 59614.0 | - | -2400.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ycbcr-4d0f86f7 YCbCr rule on private reference private-ref:ff3b177c | mask=ycbcr,ref=private-ref:ff3b177c | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 1 | 3540.0 | - | -535.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ycbcr-4d0f86f7 YCbCr rule on private reference private-ref:ff3b177c | mask=ycbcr,ref=private-ref:ff3b177c | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 1 | 9336.0 | - | -3576.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-4d0f86f7 skin model on private reference private-ref:ff3b177c | mask=skinmodel,ref=private-ref:ff3b177c | skin share of the frame, permil (share, mean, descriptive, two-sided) | 1 | 400.0 | - | -118.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-4d0f86f7 skin model on private reference private-ref:ff3b177c | mask=skinmodel,ref=private-ref:ff3b177c | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 1 | 59304.0 | - | -2710.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-4d0f86f7 skin model on private reference private-ref:ff3b177c | mask=skinmodel,ref=private-ref:ff3b177c | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 1 | 3658.0 | - | -417.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-4d0f86f7 skin model on private reference private-ref:ff3b177c | mask=skinmodel,ref=private-ref:ff3b177c | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 1 | 9212.0 | - | -3700.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ycbcr-71d92390 YCbCr rule on private reference private-ref:80de5b96 | mask=ycbcr,ref=private-ref:80de5b96 | skin share of the frame, permil (share, mean, descriptive, two-sided) | 1 | 157.0 | - | -361.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ycbcr-71d92390 YCbCr rule on private reference private-ref:80de5b96 | mask=ycbcr,ref=private-ref:80de5b96 | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 1 | 62954.0 | - | +940.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ycbcr-71d92390 YCbCr rule on private reference private-ref:80de5b96 | mask=ycbcr,ref=private-ref:80de5b96 | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 1 | 6025.0 | - | +1950.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ycbcr-71d92390 YCbCr rule on private reference private-ref:80de5b96 | mask=ycbcr,ref=private-ref:80de5b96 | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 1 | 39556.0 | - | +26644.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-71d92390 skin model on private reference private-ref:80de5b96 | mask=skinmodel,ref=private-ref:80de5b96 | skin share of the frame, permil (share, mean, descriptive, two-sided) | 1 | 64.0 | - | -454.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-71d92390 skin model on private reference private-ref:80de5b96 | mask=skinmodel,ref=private-ref:80de5b96 | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 1 | 44605.0 | - | -17409.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-71d92390 skin model on private reference private-ref:80de5b96 | mask=skinmodel,ref=private-ref:80de5b96 | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 1 | 12966.0 | - | +8891.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-71d92390 skin model on private reference private-ref:80de5b96 | mask=skinmodel,ref=private-ref:80de5b96 | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 1 | 35318.0 | - | +22406.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| Factor | Level against the control's level | Metric | n (level / baseline) | Main effect | Corrected verdict | P(beats baseline) | Bayes |
|---|---|---|---|---|---|---|---|
| mask | skinmodel against ycbcr | skin share of the frame, permil (share, mean, descriptive, two-sided) | 4 / 4 | -61.5 | DESCRIPTIVE | - | a mean metric: no closed form |
| mask | skinmodel against ycbcr | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 4 / 4 | -9330.7 | DESCRIPTIVE | - | a mean metric: no closed form |
| mask | skinmodel against ycbcr | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 4 / 4 | +3003.3 | DESCRIPTIVE | - | a mean metric: no closed form |
| mask | skinmodel against ycbcr | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 4 / 4 | -1768.5 | DESCRIPTIVE | - | a mean metric: no closed form |
| ref | private-ref:06caaa3c against private-ref:6a8ea809 | skin share of the frame, permil (share, mean, descriptive, two-sided) | 2 / 2 | -299.5 | DESCRIPTIVE | - | a mean metric: no closed form |
| ref | private-ref:06caaa3c against private-ref:6a8ea809 | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 2 / 2 | -3292.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ref | private-ref:06caaa3c against private-ref:6a8ea809 | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 2 / 2 | +2896.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ref | private-ref:06caaa3c against private-ref:6a8ea809 | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 2 / 2 | -1871.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ref | private-ref:ff3b177c against private-ref:6a8ea809 | skin share of the frame, permil (share, mean, descriptive, two-sided) | 2 / 2 | -39.5 | DESCRIPTIVE | - | a mean metric: no closed form |
| ref | private-ref:ff3b177c against private-ref:6a8ea809 | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 2 / 2 | +3719.5 | DESCRIPTIVE | - | a mean metric: no closed form |
| ref | private-ref:ff3b177c against private-ref:6a8ea809 | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 2 / 2 | -2481.5 | DESCRIPTIVE | - | a mean metric: no closed form |
| ref | private-ref:ff3b177c against private-ref:6a8ea809 | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 2 / 2 | -2549.5 | DESCRIPTIVE | - | a mean metric: no closed form |
| ref | private-ref:80de5b96 against private-ref:6a8ea809 | skin share of the frame, permil (share, mean, descriptive, two-sided) | 2 / 2 | -333.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ref | private-ref:80de5b96 against private-ref:6a8ea809 | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 2 / 2 | -1960.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ref | private-ref:80de5b96 against private-ref:6a8ea809 | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 2 / 2 | +3415.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ref | private-ref:80de5b96 against private-ref:6a8ea809 | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 2 / 2 | +25613.5 | DESCRIPTIVE | - | a mean metric: no closed form |
Experiment GR62-E3: How do the skin share and the mean skin colour (CIELAB L*, a*, b*) move with the mask (the YCbCr rule against the skin model) across the two images of the private reference store, cited by opaque id only (private-ref:6a8ea809, a sunlit outdoor frame; private-ref:052c57ab, a studio frame on a light ground)?
Design factorial · privacy private · control arm ycbcr-d421bd74 · varied factors mask,ref
| Arm | Levels | Metric | n | Mean | 95 percent interval | Delta vs control | Corrected verdict | P(beats control) | Bayes |
|---|---|---|---|---|---|---|---|---|---|
| ycbcr-d421bd74 (control) YCbCr rule on private reference private-ref:6a8ea809 | mask=ycbcr,ref=private-ref:6a8ea809 | skin share of the frame, permil (share, mean, descriptive, two-sided) | 1 | 518.0 | - | - | control | - | - |
| ycbcr-d421bd74 (control) YCbCr rule on private reference private-ref:6a8ea809 | mask=ycbcr,ref=private-ref:6a8ea809 | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 1 | 62014.0 | - | - | control | - | - |
| ycbcr-d421bd74 (control) YCbCr rule on private reference private-ref:6a8ea809 | mask=ycbcr,ref=private-ref:6a8ea809 | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 1 | 4075.0 | - | - | control | - | - |
| ycbcr-d421bd74 (control) YCbCr rule on private reference private-ref:6a8ea809 | mask=ycbcr,ref=private-ref:6a8ea809 | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 1 | 12912.0 | - | - | control | - | - |
| skinmodel-d421bd74 skin model on private reference private-ref:6a8ea809 | mask=skinmodel,ref=private-ref:6a8ea809 | skin share of the frame, permil (share, mean, descriptive, two-sided) | 1 | 369.0 | - | -149.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-d421bd74 skin model on private reference private-ref:6a8ea809 | mask=skinmodel,ref=private-ref:6a8ea809 | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 1 | 49465.0 | - | -12549.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-d421bd74 skin model on private reference private-ref:6a8ea809 | mask=skinmodel,ref=private-ref:6a8ea809 | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 1 | 8086.0 | - | +4011.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-d421bd74 skin model on private reference private-ref:6a8ea809 | mask=skinmodel,ref=private-ref:6a8ea809 | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 1 | 10735.0 | - | -2177.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ycbcr-c69c4209 YCbCr rule on private reference private-ref:052c57ab | mask=ycbcr,ref=private-ref:052c57ab | skin share of the frame, permil (share, mean, descriptive, two-sided) | 1 | 905.0 | - | +387.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ycbcr-c69c4209 YCbCr rule on private reference private-ref:052c57ab | mask=ycbcr,ref=private-ref:052c57ab | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 1 | 70581.0 | - | +8567.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ycbcr-c69c4209 YCbCr rule on private reference private-ref:052c57ab | mask=ycbcr,ref=private-ref:052c57ab | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 1 | 9915.0 | - | +5840.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ycbcr-c69c4209 YCbCr rule on private reference private-ref:052c57ab | mask=ycbcr,ref=private-ref:052c57ab | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 1 | 11250.0 | - | -1662.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-c69c4209 skin model on private reference private-ref:052c57ab | mask=skinmodel,ref=private-ref:052c57ab | skin share of the frame, permil (share, mean, descriptive, two-sided) | 1 | 906.0 | - | +388.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-c69c4209 skin model on private reference private-ref:052c57ab | mask=skinmodel,ref=private-ref:052c57ab | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 1 | 67828.0 | - | +5814.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-c69c4209 skin model on private reference private-ref:052c57ab | mask=skinmodel,ref=private-ref:052c57ab | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 1 | 9745.0 | - | +5670.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| skinmodel-c69c4209 skin model on private reference private-ref:052c57ab | mask=skinmodel,ref=private-ref:052c57ab | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 1 | 11415.0 | - | -1497.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| Factor | Level against the control's level | Metric | n (level / baseline) | Main effect | Corrected verdict | P(beats baseline) | Bayes |
|---|---|---|---|---|---|---|---|
| mask | skinmodel against ycbcr | skin share of the frame, permil (share, mean, descriptive, two-sided) | 2 / 2 | -74.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| mask | skinmodel against ycbcr | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 2 / 2 | -7651.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| mask | skinmodel against ycbcr | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 2 / 2 | +1920.5 | DESCRIPTIVE | - | a mean metric: no closed form |
| mask | skinmodel against ycbcr | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 2 / 2 | -1006.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ref | private-ref:052c57ab against private-ref:6a8ea809 | skin share of the frame, permil (share, mean, descriptive, two-sided) | 2 / 2 | +462.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ref | private-ref:052c57ab against private-ref:6a8ea809 | mean CIELAB L* of the admitted skin, thousandths (the instrument unit L_e3) (L, mean, descriptive, two-sided) | 2 / 2 | +13465.0 | DESCRIPTIVE | - | a mean metric: no closed form |
| ref | private-ref:052c57ab against private-ref:6a8ea809 | mean CIELAB a* of the admitted skin, thousandths (a_e3) (a, mean, descriptive, two-sided) | 2 / 2 | +3749.5 | DESCRIPTIVE | - | a mean metric: no closed form |
| ref | private-ref:052c57ab against private-ref:6a8ea809 | mean CIELAB b* of the admitted skin, thousandths (b_e3) (b, mean, descriptive, two-sided) | 2 / 2 | -491.0 | DESCRIPTIVE | - | a mean metric: no closed form |
Experiments, counted (14 rows, more than the 12-row disclosure bar)
| Class | Count | Note |
|---|---|---|
| experiments | 3 | experiment rows on the plan |
| judged | 1 | experiments whose design, arms, metrics and factors passed every check |
| refused | 2 | experiments refused by name: confounded, malformed, no control, an unknown design or privacy word, a bad metric, a missing level or a family wider than the correction table |
| experiments, summed | 3 | RECONCILED: the parts sum to the total 3 |
| public | 0 | experiments whose privacy word is exactly public |
| private | 3 | every other experiment: no image of any kind renders |
| privacy, summed | 3 | RECONCILED: the parts sum to the total 3 |
| contrasts | 66 | arm-against-control contrasts and factorial main effects rendered, judged or descriptive |
| contrasts, against the stated families | 66 | RECONCILED: the parts sum to the total 66 |
| observations | 68 | expobs rows on the plan |
| images shown | 0 | expimg rows rendered for public experiments |
| images withheld | 0 | expimg rows withheld because the experiment is not public |
| images refused | 0 | expimg rows of a public experiment whose source is neither https nor site-relative |
| Bayes bar, permil | 950 | the confidence P(beats control) must reach, from experiment.conf |
Inspect a rung and its prerequisites
Declared nodes 63. Rank input binding: PLAN_MATRIX_BOUND_ONLY. Dependency order is authored. Implementation, acceptance evidence, authority and resource readiness are unverified. No action is recommended or dispatched here.
Plan SHA-256 b180cc1d70482e7a558b9e48fe3325331d58ecb89029f435bfd98f985c32fe2e. Target rows 3; role rows 63. Existing risks and release worklog retain their own scope; no node completion is inferred.
Runtime observation coverage: 0 of 63 rungs have current, identity-bound execution observations. Observation may report RED. Accepted completion remains unknown; source presence, qualification and deployment do not establish it.
Use Enter or Space on a rung to inspect its contract. Prerequisite links locate another rung in this list; open its summary to inspect it. Estimates are authored effort, not forecasts.
GR1 — Referee panel
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
cjc_panelAcceptance: LANDED 2026-08-19 in one leg (estimate held): cjc_panel composes the statistics judge, the published Kovac skin-gamut ruler, image_beauty and nx_percept, forked per image; FP and FN measured on the labelled corpus v1 (n=10, 3 BAD by operator record, 7 GOOD = the six banked WebGL NXA rig stills captured on the real GPU + the published Mackenzie render) -- stats and symmetry proven non-separating and left report-only, skin and percept separable, floors 85/152 from the margins, composite FP=0 FN=0, ADMITTED under a floor-bound receipt; nx_cjc_panel_gate GREEN, bite-proven both ways. Open inside the rung: the VLM lens is UNOBSERVABLE until a sovereign VLM oracle exists (none on the estate, measured), and the floors need the second-corpus retest (MetaHuman-class stills) before they are believed
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR2 — The leap: genome splat avatar
Prerequisites: GR1 (acceptance unverified)
Contract:
gs_avatar_genomeAcceptance: RIGHT-SIZED 2026-08-18 to the published bar (the earlier budget understated by ~10x): v2 by canon-mode bodyfit search (16-dimensional ring-scale under the judge), the gated face pipeline composed in AT >=100K-gaussian face density (GaussianAvatars-class heads), strand-aligned splat GROOM hair at groom scale (85 percent of the industry poly budget), 8K-class skin with wrinkle response; done when the panel grades it above the ring figure AND the banked WebGL reference on the same framing -- declared before the experiment. THE BAR IS NOW DATA (leg 1, 2026-08-19, admitted panel on knowledge/cjc_gr2_framing.list, every row a figcrop): ring figure skin 7 percept -112 BAD; splat v1 skin 118 percept 123 BAD (skin-coloured, perceptually junk-ward); the six WebGL rig crops skin 279-661 percept 210-228 GOOD; Mackenzie skin 624 percept 281. DONE-RULE IN NUMBERS: v2 crop must read composite GOOD with percept above 228 (the best banked rig crop) and skin at or above 279 (the weakest), re-measured with the same conf and receipt
Authored effort: 4. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR3 — 4D splat motion
Prerequisites: GR2 (acceptance unverified)
Contract:
g4_avatar_rigAcceptance: The splat avatar moves with the same gait, breath and expression machinery the mesh uses; replay bit-identical
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR4 — Second species canon
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
bp_species_canonAcceptance: A horse canon through the identical pipeline; a bodybench-style ruler against a CC anatomy oracle of the species proves species-agnostic generation
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR5 — Part-genome creatures
Prerequisites: GR4 (acceptance unverified)
Contract:
cg_genome_partsAcceptance: Typed interpolatable parts over the existing plans; distinct genomes must measure distinct and interpolation must stay anatomically legal
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR6 — Indoor rooms
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
ig_room_solveAcceptance: Constraint-solved rooms with materials through our engine, deterministic and bounded; REFUSES an unsatisfiable constraint set rather than overlapping furniture
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR7 — Advances archive mirrored
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
gp_collect2Acceptance: Second root through the ingest organ, fixed-point crawl, partition reconciles per root; the extractor then cites it
Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR8 — PBR map set per genome
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
ntb_bake_pbrAcceptance: Four maps by four regions regenerated per genome at load; normal-map lighting proven through the existing image path
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR9 — Pre-integrated SSS
Prerequisites: GR8 (acceptance unverified)
Contract:
sr_sss_preintAcceptance: Bound to the baked normal map; carries the away-side equality tooth (the T66 form) so no uniform brightener can fake it
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR10 — Path-traced GI
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
gp_bounce_giAcceptance: Per-frame bounce beyond cached radiosity; the cache-vs-rebuild equality stays as the determinism tooth
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR11 — Native-resolution GPU delivery
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
cw_webgpu_computeAcceptance: HALF LANDED 2026-08-18: the WebGL2 tier is LIVE DEFAULT on all five worlds and the WebGPU WGSL world pass is LIVE LEVERED (?gpu=1, real-GPU verified via the CDP wire, boot-stall ceiling hardened); REMAINING: the compute-shader form, the NPC+eye passes to parity so WebGPU can default, and the same frame bit-compared against the CPU path at reference size
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR12 — Expression morphs
Prerequisites: GR2 (acceptance unverified)
Contract:
fa_expr_morphAcceptance: FACS axes over the identity morphs, graded against the 710-expression FACS-class basis MetaHuman ships (the bar, not a smile bone); the panel grades expressions readable
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR13 — Skin-bound dynamics
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
dbs_displace_freeAcceptance: CONTRACT SYMBOL CORRECTED 2026-09-01 (board owner): this rung declared dyna_bind_skin while its matrix row named runtime/nx_nxa_dyna.nx, a pair that could never resolve, so GR13 held ranker slot #8 as OPEN for a capability that had shipped. Measured full-population, corpus_complete=1: func dyna_bind_skin( is declared once, at runtime/nx_softbind.nx:238, and it ticks three soft-tissue REGION points -- not this rung's subject; func dbs_displace_free( is declared once, at runtime/nx_dyna_bind_skin.nx:203, and is the per-vertex half that procgen.matrix and softbody.matrix already carry LIVE. DONE-RULE, unchanged in substance and now pointed at a gate rather than at a declaration: the proven solver bound to skin VERTICES with collision against the body it rides, the flip being the receipt and nx_dyna_bind_skin_gate on the graphics.gates roster being the proof -- a green cell over a red gate is not done
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR14 — Linear-light pipeline
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
sr_linear_lightAcceptance: All lighting math computed in linear, display-encoded exactly ONCE at output; proven by the tonemap A/B that failed 2026-08-26 (ACES over display-encoded values washed the frame -- the banked double-transfer lesson) now passing: filmic curve applied and the A/B screenshot referee grades it richer, not paler; prerequisite to GR9
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR15 — Sovereign strand system (TressFX-referenced)
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
hr_strand_sysAcceptance: The Infinigen-fork hair chain, leg 1: a NishiLang strand system whose sim and render structure is read from the mirrored TressFX source (the canonical shipped open real-time strand system) -- guide strands + per-strand constraints on the one solver, rendered with real width and coverage in our WebGL2 tier; done when a girl-framed capture is graded by the panel above the current ribbon baseline and no ad-hoc constant lacks a cited row
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR16 — Marschner-class hair shading
Prerequisites: GR15 (acceptance unverified)
Contract:
hr_marschner_lobeAcceptance: Hair BRDF per Marschner 2003 through the Karis 2016 / Tafuri 2019 production reductions (mirrored): dual-lobe spec with shifted primary and secondary highlight, azimuthal roughness; constants pinned as cited conf rows before the shader edit, gate compares lobe positions against the published reduction
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR17 — Groom style zones with pinned dynamics
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
gr_style_zonesAcceptance: nx_nxa_groom gains the cosmetology-brief zone model (hairline, part, nape, density map) with strand dynamics constants extracted from the mirrored Super-Helices / Discrete Elastic Rods papers and PINNED as oracle rows (closing the conf's own uncited hair_k_q10 correction record); style presets are parameters, every band cited or derived
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR18 — Infinigen fork map as data
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
if_fork_mapAcceptance: The fork strategy is a LIVE census, not prose: a conf mapping each Infinigen subsystem (ProcFunc generators, constraint solver, materials, creatures, ground-truth writers, Infinigen-Sim articulation) to its sovereign NishiLang twin, its governing SOTA reference, its gate, and its measured state -- rendered on this board and re-measured at emit; the differentiators (inside-out anatomy, genome identity, bit-exact determinism, the adult invariant) carried as exceed rows
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR19 — Render the sections we already generate
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
ntb_view_sectionsAcceptance: MEASURED 2026-08-28: ntb_view finds only VERT and TRIS, so a groomed and dressed asset renders IDENTICALLY to a bare one -- nv=14164 tris=28092 filled3=56482 for both, with 4800 generated hair strands and a 648-vertex measurement-fitted garment present in the file and ignored. We GENERATE hair and clothing and cannot SEE them server-side, so GR15-GR17 and the garment lane have no judgeable artifact and the referee panel can never be pointed at them. Done: ntb_view consumes GVRT/GTRI and HSTR (strands as camera-facing ribbons), reports gverts/gtris/strands SEPARATELY from the body counts, refuses on index overrun, and stays a no-op when the sections are absent; proven by the emitted PNG differing and the reported triangle count strictly exceeding the bare asset's on the same input. Garment reuses the existing rasteriser; the open design question is that tm_set_image is global, so garment and hair need either a flat-region UV or a per-vertex vertex-colour path
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR20 — Garment and hair penetration ruler
Prerequisites: GR19 (acceptance unverified)
Contract:
cc_garment_penAcceptance: nx_clipcheck already measures the two GROSS self-clipping classes and on the shipped ref9.nxa returns verdict=CLIP (44064 keys, torso_clip_keys=146 worst 21mm, floor_clip_keys=2176 worst 30mm) -- but its axes are limb-through-torso and foot-through-floor. The operator-reported defect is GARMENT-through-body and HAIR-through-body, and NOTHING in the estate measures it: a ruler exists for the neighbouring defect and reads RED while the reported defect has no ruler at all. Done: per-vertex penetration of GVRT against the body surface and of HSTR roots/segments against the scalp, reported as a count plus worst depth in mm plus the offending element index, before and after XPBD settle, with a neg-control that a deliberately oversized garment reports MORE penetration than the fitted one
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR21 — Genome asymmetry -- synthetic faces are too perfect
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
bg_asym_permilAcceptance: MEASURED 2026-08-28 on the ADMITTED referee panel (cjc_panel, floors from knowledge/cjc_panel.conf): cjp_sym reads 938 for our generated body and 913 for both MetaHuman frames against 650 for the photographic bar -- every synthetic subject is ~1.4x more bilaterally symmetric than a real person, and the panel measures it today with no rung consuming the signal. This is a realism defect that costs NOTHING to fix in a procedural generator and is structurally unavailable to a finite asset library: a real face is asymmetric and ours is machine-perfect. Done: a per-mille asymmetry gene drives independent left/right ring-scale and feature offsets from the seed, deterministic and reproducible from the genome; proven by cjp_sym moving toward the photographic band without cjp_percept falling below the admitted tier floor
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR22 — Camera rejection sampling
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
ntb_cam_rejectAcceptance: Infinigen's published figures are honestly non-cherry-picked at the IMAGE level, but the VIEWPOINT is rejection-sampled best-of-ten by pixel-depth variance -- so part of the quality gap a reader attributes to their renderer is a composition filter. Cheapest visual win available: the emit path renders N candidate viewpoints, scores each by depth variance over the figure, and keeps the best. Done: N and the score are conf rows not literals, the chosen index and its score are printed so the choice is auditable, and a neg-control shows the WORST-scoring viewpoint grades lower on the referee panel than the best -- if it does not, the heuristic is not measuring what it claims and the rung fails
Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR23 — One colour transform, two execution surfaces, ours
Prerequisites: GR14 (acceptance unverified)
Contract:
sr_view_transformAcceptance: SOVEREIGNTY RULE APPLIED (operator 2026-08-28: we dont want webgl2 and other 3rd party unless we are recreating ourselves the capabilities). The transfer curve and the LUT are DERIVED AND GENERATED IN NISHILANG; AgX is an ORACLE we check ourselves against, never a dependency -- borrow the rulers, not the runtime. TWO INDEPENDENT RIVALS CONVERGE ON IT, which is why it is worth recreating: Infinigen never sets a view transform and inherits the Blender 4.2 AgX default with compositor grading OFF (read byte-for-byte in our mirror: distort False, color_correct False), and VR HOT deliberately ADDED AgX in 0.13 on 2026-06-16 calling it a color profile you might know from other high end games. Done: ONE NishiLang implementation of the curve consumed by BOTH the page-side surface (beach lane) and the server-side portrait renderer (GR19), byte-identical LUT from one generator, so the two surfaces CANNOT diverge; the native renderer is the DURABLE home and the browser page is a stopgap execution surface. GR14 is the prerequisite and carries the measured lesson: the 2026-08-26 A/B failed because ACES was applied over display-encoded values and washed the frame -- transform over LINEAR light, display-encode exactly once
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR24 — Identity parameter space with correlated sampling
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
bg_identity_spaceAcceptance: THE INTELLIGENT-PROCGEN HALF, and the axis a finite asset library structurally cannot enter. MEASURED RIVAL POSITION: VR HOT ships roughly 15-20 hand-picked skins accumulated 2-4 at a time over five years on a FIXED LICENSED TOPOLOGY (Daz Genesis 8.1, vendor-stated in their own 0.5.4 patch notes), so their identity space is bounded by what one marketplace authored; Infinigen has NO humans at all. Ours is a seed. Done: semantic master parameters (age, adiposity, sun exposure, grooming, dimorphism) fan out into correlated low-level axes through a FULL COVARIANCE draw rather than independent per-channel sampling -- the mechanism behind why a rival palette reads deliberate instead of tinted -- with each axis drawing from its own split stream so adding a parameter never reshuffles existing genomes; every character reproducible and auditable from its genome, proven by bit-identical regeneration from the seed and by the referee panel showing identity spread across a population without any member falling below the admitted tier floor
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR25 — Placement solved, not authored
Prerequisites: GR20 (acceptance unverified)
Contract:
pl_fit_solveAcceptance: Their known-issue lists are our exceed axis, measured: VR HOT carries nipple clipping, bra clipping, body-hair-through-cloth and texture seams as PERSISTENT known issues across years, because garments are an authored library conformed onto a licensed topology; Infinigen has no garment system at all. We already generate a garment fitted BY MEASUREMENT (18 body cross-sections, ease margin) and a groom rooted by area-weighted scalp sampling -- so for us non-penetration is achievable BY CONSTRUCTION rather than by hand-fixing each asset. Done: garment and hair placement produced by a constraint solve over the measured body (clearance as a solved constraint, not an authored offset), such that GR20 reports ZERO garment-through-body and hair-through-body penetrations across a declared pose set for an arbitrary genome -- including bodies at the extremes of GR24's parameter space, which is precisely where an authored library fails
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR26 — Physically-based sky and aerial perspective
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
gpe_sky_atmosAcceptance: The sky is a two-stop vertical colour ramp today (recipe spec rows 7 skytop and 8 skyhor) with the EH3 hash cloud sheets over it -- a gradient, not an atmosphere, and on the live beach it is roughly half the frame at default pitch. Single-scattering Rayleigh plus Mie with a sun disc, plus aerial perspective tinting distant geometry, in the same integer determinism as every other render rung. Ordered BEFORE GR27 because ground albedo is lit by sky irradiance, so a flat ramp caps how real any sand can look no matter how good its material is. Operator 2026-08-28 named it first of seven.
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR27 — Ground material: granular sand and wet-shore darkening
Prerequisites: GR26 (acceptance unverified)
Contract:
wc_mat_groundAcceptance: EH1 fixes terrain SHAPE and EH2 fixes what GROWS; neither touches what the ground IS, so the shore reads as tan voxel paint. Granular high-frequency normal that survives grazing angles, darkening plus specular gain inside the wet band, slope-keyed dune banding. The one input it needs already exists: recipe spec row 66 carries the ABSOLUTE shoreline coordinate, so height-above-waterline is readable today with no new plane. Depends on GR26 for the irradiance that makes wet-vs-dry separate at all.
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR28 — Palm flora reaches the SERVED beach
Prerequisites: GR31 (acceptance unverified)
Contract:
wc_veg_meshAcceptance: RE-SCOPED 2026-09-01 (board owner). THE GROWER LANDED 2026-08-30 and this rung has been ranked #2 for it ever since: func wsp_palm_shape( is declared and CALLED at sh9 == 3 in nx_wasm_craft.nx, so the 2026-08-28 claim that only shapes 1 and 2 have growers is stale. The old contract name wc_tree_palm can never ship because the capability shipped under a different name on gameengine -- repointed, not left to rot. WHAT IS OPEN is the half this rung always also named ("one grower plus ONE RECIPE ROW") and it is deliberately reverted, not forgotten: knowledge/world_recipes.conf carries the beach row with NO spec pairs on purpose, because both densities were judged on REAL-BROWSER frames -- 5:64 an opaque green canopy roof, 5:400 a forest of solid green cubes at eye level, because frond rosettes render as filled voxel crowns. DONE-RULE, taken verbatim from the conf's own revert record rather than invented here: palms return to the beach only when the crown renders as fronds (or the world leaves voxels) -- so this rung now DEPENDS on GR31, and its proof is a captured frame of the SERVED beach graded by the world referee (GR30), never the presence of a grower. CONTRACT NAME ALIGNED to gameengine.matrix the same hour: lane 3 split the same capability onto _ABSENT_:wc_veg_mesh minutes before I named it wsp_beach_flora_bind, and shipping both would have manufactured a third two-names-one-capability defect on the very board being corrected for two of them. Their arithmetic also supersedes the conf's "retune the density" reading: 1 block = 1 m, a frond is a chain of at most four 1 m cubes, so a crown is ~33 cubic metres of solid green against a real rachis of ~0.05 m -- no sparsity value makes a blob stop being a blob, and the exit condition is sub-block crown geometry, i.e. GR31. Operator 2026-08-28: no infinigen generated sand or palm trees.
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR29 — Skin micro-surface on the path the family looks through
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
gpe_skin_microsurfaceAcceptance: SCOPE SETTLED 2026-08-28 BY MEASUREMENT, after two wrong diagnoses of mine that the instruments corrected. NOT missing maps: TEXM is PRESENT, floor gate 7/7, 721,830 B, GL tier samples it. NOT a missing pore model: nx_relief_lib owns a cited three-band law and the baker already consumes it. THE DEFECT IS RESOLUTION, and nx_relief_bands_gate proves it -- a whole-body atlas spans 1.7m, so at 4096 the texel is 415um and secondary and pore both read carried=0; the live cast ships at 2048. Bands are computed then clamped, leaving a coarser band wearing the fine band's name, which is exactly what reads as painted. RES NEEDED: secondary 13600 whole-body, but over a 250mm face only 2000, and pore only 625. SO THE RUNG IS A PER-REGION ATLAS, not a bigger whole-body one -- the resolution we already ship carries the pore band once the footprint is a face rather than a body. Done rule: pore carried=1 at shipping resolution on a region atlas, with the neg-control that the whole-body atlas still reports 0 rather than being quietly redefined. Done rule: the served page shows measurable micro-relief at skin scale under the referee panel, with the neg-control that a flat-albedo build FAILS the same tooth.
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR30 — World-quality referee built and wired
Prerequisites: GR1 (acceptance unverified)
Contract:
wqg_build_wiredAcceptance: THE WORLD HALF OF GR1 AND IT WAS NEVER ON THIS PLAN. GR1 admitted a CHARACTER referee panel and the world lane still has none, so every world fidelity claim on this board is graded by the author's own eye. MEASURED 2026-09-01 by nx_catalog: nx_world_quality_grader is SOURCE yes 37940 B, BUILT NO, PROMOTED NO, REGISTERED NO, INVOKED NO, verdict SOURCE-ONLY (S2). The consumer already exists in prose -- nx_layer_verdict.nx carries an adapter for its 12-i64 output and names _wqg_grade_from_8, and _ft_diag.nx imports the organ -- so this rung is a BUILD and a WIRE, not a design. Done when the organ builds, promotes, registers, returns a grade plus 8 axis scores for a named world capture, and at least one world rung's done-rule reads that grade instead of a sentence. Carries the GR1 discipline: floors declared before the experiment, FP and FN measured on a labelled corpus, and the panel refuses rather than acquits when it cannot see
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR31 — A second art style of emission beside the voxel raycast
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
wc_style_surfaceAcceptance: OPERATOR 2026-08-30: everything is voxel and we dont need it to be that should just be one art style of emission. MEASURED 2026-09-01 on the served page: the beach WASM module exports 78 functions and NONE of them names a mesh, surface, rig, face, hair or skin producer -- the exports are simulation state and the world is drawn by a per-pixel integer DDA raycast over a voxel field, so voxel is the only style the engine can emit rather than the one it chose. The extractor is already built and graded elsewhere on the estate: nx_csg_extract, marching tetrahedra on the Kuhn six-tetrahedron split, watertight by construction, LANDED on both procgen and dcc. Done when the same worldgen field emits as a surface mesh through that incumbent extractor with the style selected by a recipe row and never by an engine fork, and with the voxel path proven byte-identical when the row is absent -- a second style, not a replacement
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR32 — The sovereign strand organ reaches a live world
Prerequisites: GR15 (acceptance unverified)
Contract:
wc_strand_bindAcceptance: GR15 buys a strand SYSTEM at 2 u as if from nothing. MEASURED 2026-09-01 by nx_catalog: nx_nxa_strands is SOURCE yes 9565 B, BUILT NO, PROMOTED NO, INVOKED NO, verdict SOURCE-ONLY (S2) -- 9565 B of strand source sits unbuilt beside a rung that does not know it exists, so the plan overstates its own cost and hides that the first move is a build. This rung is deliberately the BINDING and not the organ: building it alone would leave the page-side groom the operator is looking at untouched. Done when a strand set generated by the sovereign organ reaches a live world through an engine export and the page copies buffers rather than authoring geometry, which is the same bar gameengine GE21 wc_hair_tick and GE22 gpe_wasm_hair_bind already set for the guide chain
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR33 — Strand chain ordered root-to-tip
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
hg_seg_axisAcceptance: THE CHEAPEST VISIBLE FIX ON THIS BOARD AND IT IS ONE TOKEN. The served page derives a strand segment index from component 1 of the bind position while this mesh space is Z-UP, which the served shader itself declares (vec3 yup=vec3(sp.x,sp.z,-sp.y), beach.html:292) and which the fit pass ten lines above the bug already honours. The chain lag and its squared weight are therefore spread front-to-back across the head instead of root-to-tip down each strand, which is both the mop shear the operator sees and the mechanism that drives hair through the face. DONE-RULE, pointed at a gate and not at a declaration: nx_gpe_hair_geometry_gate carries hg_seg_axis, a tooth asserting the segment index is derived from the up axis the shader declares rather than from a hard-coded component, bite-proven RED against the current served expression and GREEN against the fix, and the gate is listed on graphics.gates so the verdict is executed evidence rather than prose
Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR34 — The normal map already inside the asset is read
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
gpe_texm_normal_bindAcceptance: The served ref9e.nxa carries a four-map TEXM (albedo, spec, gloss, NORMAL; NTB_MAP_NORMAL=3) and the page reads record 0 at both TEXM call sites, so a normal map ships in the file the browser already downloads and nothing samples it. This is layer one of a three-layer consumer gap and is deliberately scoped to layer one. DONE-RULE: the page binds map record 3 as a normal map on the character pass; nx_nxa_texm_gate (re-run 2026-09-01, passed 12/12 verdict=GREEN) stays green so the carry is unbroken, and a new tooth shows the lit result differs from the record-0-only build on the same frame -- a bind that changes no pixel is not a bind
Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR35 — A declared asset ladder, so finished work reaches the cast
Prerequisites: GR34 (acceptance unverified)
Contract:
gpe_asset_ladderAcceptance: Roughly 80 MB of finished character assets are orphaned: ref9f.nxa (40448728 B, ref9e plus TEXR:3590072, every other section byte-identical) and ref9d_r2.nxa (39478840 B) are referenced by NOTHING in buildroot, measured with corpus_complete=1, while the page serves ref9e. The mechanism already exists (a ?npcasset= override and a ref9e/ref9d/ref9 fetch chain), so this is a declaration, not a loader. DONE-RULE: the ladder is a conf row naming the richest asset a build will serve, the emitter refuses by name when a named asset is absent rather than silently falling back, and nx_asset_floor_gate passes on the asset actually selected -- proven by the emitted page naming the richer asset and by a capture whose byte count differs from the ref9e build
Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR36 — FACE and MORF are consumed, so the face is geometry and not paint
Prerequisites: GR30 (acceptance unverified)
Contract:
gpe_morf_consumeAcceptance: The face is painted in a fragment shader from head-height ratios with NO NOSE TERM at all, while the served asset carries FACE:4700 and MORF:98634 and the page three NXA consumers dereference 15 tags that include neither. The producer nx_nxa_morf is LIVE and its consumer was never written; 7 of 9 face organs are REGISTERED-DARK. CARRIES ITS OWN HISTORY AS A PRECONDITION: face relief v1 was reverted 2026-08-03 for goblin noses because this mesh normals are majority-wrong, so the DONE-RULE is (a) the normal precondition is measured and stated before any relief is applied, (b) FACE and MORF are dereferenced by the page consumers, and (c) the referee panel grades the result at head framing at or above the painted baseline -- a rung that only makes the nose exist has reproduced 2026-08-03
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR37 — The JS-logic ratchet is pinned against class laundering
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
jl_class_pinAcceptance: The ruler that enforces the estate loudest operator order is editable by the party it measures: knowledge/jslogic_class.conf maps each literal hash to a CLASS token, and moving a literal from LOGIC to INTEROP or SHADER removes its bytes from logic_unknown with no code change and no re-baseline. The conf self-invalidates on a literal EDIT (the hash moves, the row reads UNKNOWN) and NOT on a class-token edit, so the one mutation that launders the number is the one mutation it cannot see -- measured: the headline improved 228377 to 128559 on 2026-08-31 while zero JS bytes were removed and the two journal rows total 229911 identically. DONE-RULE: the census reads a declared digest of the class column and REFUSES by name when a class moved without the baseline being re-declared, with TWO neg-controls -- an ordinary literal edit must still self-invalidate the way it does today, and a class move WITH a re-declared baseline must pass. Scoped away from nx_jslogic_census_gate, which a sibling lane wrote the same day for the uncharged-JS axis and which is SOURCE-ONLY: extend that gate, never add a second ruler beside it
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR38 — The character judge survives the frame it is pointed at
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
cjc_frame_guardAcceptance: The world lane has no working referee for its own flagship frame: nx_charjudge takes SIGSEGV signal 11 at 0x41510f touching 0x0 on the beach capture and nx_addr2line refuses because the elf was built without -g, while nx_percept anchor corpus is entirely NXFH1 400x240 so a 1920x1200 capture is not comparable to it. The one number the lane CAN produce is the honest headline of this board: nx_frame_score on the native beach capture reads quality=408 verdict=BASIC against the ruler own published random-squares control of 628 RICH. DONE-RULE: the judge grades a full-resolution frame or ABSTAINS by name and never dies, a debug build exists so the next crash is locatable, and the abstention is a THIRD state that can neither acquit nor convict
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR39 — A character crop that contains a character
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
ws_char_crop_nonemptyAcceptance: nx_world_snap char reports CHAR-FRAME framed=1 ground_y=11 and then CHAR-CROP pixels=0, so framing succeeds and the crop returns no subject image and every downstream character judgement about a world capture is made about an empty picture or not at all. DONE-RULE: a tooth asserts pixels>0 whenever framed=1, the crop refuses by name rather than emitting an empty image, and the hand-rolled NXFH1 writer at nx_world_snap lines ~515-555 is replaced by the shared codec nxl_dump so the estate keeps ONE frame writer -- the duplicate-ruler half is part of this rung, not a separate one
Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR40 — The board ranker measures every row
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
rk_measure_allAcceptance: THE META-RUNG, and it gates the trustworthiness of every other number on this page. nx_compare_rank sets mpres=1 for any row not spelled _ABSENT_ and then does if mpres[mi]==1 { done=1 } else { done=rk_measure(...) }, so it measures only the rows already spelled absent and trusts every row spelled present: a rung can be promoted open to done by measurement and can never be demoted done to open, which makes a false green invisible by construction and computes the critical path on corrupt input. The emitter does not share the defect (it calls sd_present on both branches), which is why one row read not-measured on the served page and #1 FINISH in the ranker on the same day. DONE-RULE: the ranker measures both branches; a bite proves a present-spelled row with an absent symbol flips done to open; and because the real risk is a silent fleet-wide re-baseline, the run must publish the before/after done count for every domain in the same output
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR41 — The review viewers gain the sovereign wasm tier
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
nre_sov_tierAcceptance: OPERATOR 2026-09-02 (rigview printed NXA REFUSED no webgl2): nx_nxa_rig_emit, nx_nxa_view_emit and nx_nxa_stream_emit have exactly one renderer, WebGL2, and a refusal leaves a review instrument blank. gameengine GE28 makes the refusal name its cause; this rung removes the dependency. The estate already owns the pieces: nx_trimesh (the sovereign textured triangle rasteriser ntb_view uses server-side) and nx_meshview_wasm (the wasm memory model, 64x64 and 256 verts today). Done when a wasm build of the trimesh rasteriser sized for a 14k-vertex rig renders the same NXA the GL tier renders, skinned from the same SKEL and SKIN sections, and the rigview page falls to it with the badge naming the tier -- proven by a headless capture whose figure crop grades on cjc_panel inside the GL tier's band
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR42 — The beauty tier's first frame: the beach scene rendered offline by a sovereign path tracer and published beside the live frame with its referee score
Prerequisites: GR26 (acceptance unverified), GR10 (acceptance unverified)
Contract:
env_beauty_sceneAcceptance: OPERATOR 2026-09-02: get things Infinigen generating to photoreal. Infinigen renders offline (Cycles path tracing, minutes per frame); its twin here is the BEAUTY TIER the matrix already declares ABSENT on nx_gsplat (env_beauty_scene) with the GI contract on nx_gi_path (gp_bounce_gi, also ABSENT: no source exists). Done when (a) nx_gi_path.nx exists as a sovereign integer path tracer over a scene the procgen planes already produce (heightfield terrain from nx_worldgen with a step derived from the pixel footprint so no terracing, water, the physically-based sky of GR26, the cast meshes with their TEXM maps), Monte Carlo bounce with a seeded deterministic sampler so two runs are byte-identical, (b) env_beauty_scene assembles that scene from the same genome and recipe planes the live page reads, (c) the frame is emitted at 1920x1200 with depth, normal and semantic ground truth beside it (the dataset-factory half), (d) it is published on this board beside the live capture and the mirrored Infinigen reference frame, and (e) a referee that PASSES its own ordering ladder grades all three -- until that referee exists the cell reads UNGRADED, never GREEN. The live real-time tier then chases this frame rung by rung (GR26 sky, GR27 ground, GR14 linear light, GR34 normal maps, GR16 hair, GR29 skin) with the same referee.
Authored effort: 8. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR43 — A ray-triangle bounding-volume hierarchy for the beauty tier: the path tracer hits the cast meshes
Prerequisites: GR42 (acceptance unverified)
Contract:
gp_bvh_ray_triAcceptance: MATH GAP MEASURED 2026-09-03 (nx_capsearch over 7245 organs, corpus_complete=1: nx_bvh_lib is BioVision mocap, no ray acceleration structure exists by name), so nx_gi_path marches a heightfield and cannot hit a triangle; done when a BVH over the cast's NXA triangles (median split on integer centroids, deterministic from the mesh) is traversed by the path tracer, a brute-force every-triangle oracle agrees on every ray of a fixture (same nearest hit, same barycentrics), the build and traverse costs are printed per frame, and a neg-control with one corrupted node bound is caught by the oracle
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR44 — A low-discrepancy sampler for the path tracer: Sobol with Owen scrambling, the blue-noise organ reused for the screen
Prerequisites: GR42 (acceptance unverified)
Contract:
gp_sobol_owenAcceptance: MATH GAP MEASURED 2026-09-03: the tracer draws from a hashed LCG and no Sobol or Halton sampler exists by name (nx_bluenoise exists for screen-space masks); done when a deterministic Sobol sequence with per-pixel Owen scrambling drives every bounce, variance at equal spp on a fixture scene is measurably lower than the LCG with the number printed and the LCG kept as the control, two runs are byte-identical, and the sampler is one lib that both the offline tracer and the real-time reconstruction (performance PF4) read
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR45 — Spherical-harmonics irradiance from the physically-based sky for the real-time tier
Prerequisites: GR26 (acceptance unverified)
Contract:
sr_sh_irradianceAcceptance: MATH GAP MEASURED 2026-09-03: no spherical-harmonics or irradiance-probe organ exists by name and the live tier lights with a two-stop ramp; done when the GR26 sky is projected to SH9 in NishiLang integer arithmetic (nx_fixq30_lib), the shader emitted by nx_wgsl from the same source evaluates it per normal, a reference integration agrees within a stated band on a fixture sky, and the frame is graded by the world referee against the ramp with the neg-control of the projection zeroed
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR46 — Cascaded shadow maps on the kernel and WebGPU tiers: the cast casts a shadow on the ground
Prerequisites: GR42 (acceptance unverified)
Contract:
rk_shadow_cascadeAcceptance: MATH GAP MEASURED 2026-09-03: no shadow-map or cascade organ exists by name (the served beach draws dark blobs under feet); done when the GE42 kernel renders a sun-depth map for N cascades chosen from a conf ladder row, the world pass samples it with a bias DERIVED from the texel footprint, a fixture with a known occluder produces the analytically expected shadow footprint within a band, and the served beach frame is graded by the world referee with the maps forced off as the neg-control
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR47 — Integer Cholesky for the correlated identity draw
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
la_cholesky_q30Acceptance: MATH GAP MEASURED 2026-09-03: nx_matrix carries determinant and inverse to 3x3 and no Cholesky decomposition exists anywhere by name, so GR24's full-covariance draw has no primitive; done when a Q30 fixed-point Cholesky over nx_fixq30_lib arithmetic factors any SPD matrix up to the identity-space dimension, L times L-transposed reproduces the input within a stated Q30 band on random SPD fixtures, a non-SPD input is refused by name rather than producing garbage, and GR24 composes it for its draw (one ruler, never a second decomposition)
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR48 — The rendered strand is the collided strand: character geometry leaves emitted JS for NishiLang lowered to wasm
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
gpe_char_geom_wasmAcceptance: OPERATOR 2026-09-04, the loudest order restated as an absolute: printf-emitted foreign-language logic is not a thin shim we have to accept, it is hacky programming, and first byte up every rung until it HAS to translate is the only way anywhere in the estate. MEASURED THIS SESSION, and it is the ROOT CAUSE of the standing REJECT beach-hair-colour-wind: the hair a visitor sees is built in hand-written JavaScript inside a NishiLang string literal in nx_game_page_emit.nx -- Catmull-Rom subdivision, NXH_GUIDES 1200 by NXH_KIDS 6 child interpolation, per-strand lateral axis, taper, quad emission and UV generation, hundreds of lines of geometry maths running on the main thread at load. The COLLIDER is in wasm (nx_wasm_craft hg_capd2 hg_inside hg_push) and runs on 128 SIMULATED guides. So the collider structurally cannot see the geometry that is rendered, which is why a clamp shipped into that JS on this date changed nothing an operator could see. THREE CONSEQUENCES, each measured: the collider is correct and invisible; the JS is outside the compiler, outside every gate, and outside grep itself (nx_shelltool skips the long emitted lines and reports linetoolong, so absence over that code is UNPROVEN by construction); and a fix in that layer can only ever be a patch. DONE-RULE: the strand construction is NishiLang compiled through nx_compile_wat and lives beside the collider so there is ONE geometry owner, the collider resolves against the RENDERED strand set and a tooth asserts rendered equals collided rather than 128 versus 7200, the JS path is retired only after the wasm path reproduces its vertex buffer within a stated band on a fixture groom, and the referee panel grades a head-framed capture at or above the JS baseline. NOT DONE if the JS block merely shrinks: the acceptance test is that no geometry byte on the page is hand-authored.
Authored effort: 4. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR49 — The translation ladder is declared and counted per target: zero hops on NishiOS and Nishi browser, minimum hops elsewhere, an API layer only where a third party mandates one
Prerequisites: GR37 (acceptance unverified)
Contract:
tl_translation_censusAcceptance: OPERATOR 2026-09-04: do everything first byte up in NishiLang and estate and engines, then ship to the translation layer if it requires one, but make it so that in NishiOS and Nishi browser we do NOT have a translation layer so we can be superior in performance and capabilities and delivery, and if third party ship via an API layer if mandatory -- the least possible translation is the best. THE ESTATE ALREADY ENFORCES THIS ON TWO AXES AND NEITHER COVERS THE THIRD: GE40 wp_provenance demands every served wasm byte trace to a named .nx, and GE44 rk_webgpu_door demands ZERO shader bytes on the page that do not trace to nx_shader_ir and reduces WebGPU to a submission pipe with no geometry pipeline. NOTHING MEASURES TRANSLATION HOPS PER TARGET, so the JS layer grew to roughly 128559 bytes of logic (GR37) and the character geometry ended up hand-authored inside it with nobody able to see the number rise. DONE-RULE: a conf ladder row per delivery target declares its permitted hop count and its permitted translation kinds, NishiOS-native and Nishi-browser-native declare ZERO with no wasm and no JS and no WGSL in their path, third-party-browser declares the minimum with each hop NAMED and justified by a mandate rather than by convenience, a census walks every served artifact and counts the hops it actually took, the count is a RATCHET that may not rise, and the census REFUSES by name when a target exceeds its declared ceiling. TWO NEG-CONTROLS: a hop added to a zero-hop target must fail, and a hop merely re-classified without being removed must also fail -- that second control is the GR37 laundering lesson applied here before it can happen rather than after.
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR50 — The ship loop adjudicates itself against the SERVED artifact and refuses to report success for a provably inert promote
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
os_served_deltaAcceptance: OPERATOR 2026-09-04: the delivery pipeline must be self-adjudicating, a human should not be in this loop, and the math was clear as day from the first failed ship. MEASURED, and it is a three-instance class not a one-off. INSTANCE ONE: nx_organ_ship nx_game_page_emit ran the whole loop GREEN -- build, adoptgate with four lost runs adjudicated, promote 425241 to 451160 B, offc twin, register, verify -- and changed NOTHING a visitor can see, because the target is a LIBRARY whose standalone binary emits no served artifact. Everything needed to say so was one command away: grep for the import edge returns 3 importers with corpus_complete=1, nx_gpe_hair_geometry_gate nx_world_recipe_gate and nx_craft_emit, and only the last declares served outputs CE_OUT CE_OUT2 CE_OUT3. The loop never asked which served artifact it changes. INSTANCE TWO: the same loop reports verify against the BINARY, so a fresh page emitted by a 15.6 hour old writer reads as a successful delivery. INSTANCE THREE: /api/promote already reads organ_kind.conf and refuses daemons BY NAME, so kind data is on the path and a lib promote is permitted and SILENT. DONE-RULE, two teeth and both cheap: FIRST the loop resolves the target's served artifacts from the organ's own output declarations and, when the set is EMPTY, it NAMES every importer and REFUSES to report success, because shipping a lib alone is inert BY CONSTRUCTION rather than by bad luck; SECOND it hashes each served artifact before and after and a ship whose served bytes did not move is reported FAILED however green every stage was. TWO NEG-CONTROLS: a genuine writer ship whose page bytes DO move must still pass, and a lib ship must fail even when build prove adoptgate promote and register are all green. NOT DONE if it merely warns -- the failure mode being fixed is a loop that reported success, so the verdict must change, not the prose.
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR51 — The hair gate gains a tooth the broken output must fail: rendered strands inside the body are counted, not assumed
Prerequisites: GR48 (acceptance unverified)
Contract:
gpe_hair_penetration_toothAcceptance: MEASURED 2026-09-04 on the LIVE gate: nx_gpe_hair_geometry_gate reads passed 16 of 16 verdict=GREEN while the operator is looking at hair clipping through the chest, and BOTH statements are true because they are about different subjects. Every tooth it carries asks whether the pipeline is WIRED -- emitted-consumes-HVRT/HTRI-groom-MESH, emitted-consumes-HSTR-guide-strand-geometry, emitted-indexes-the-chain-table-BY-VERTEX-SEGMENT, emitted-UPLOADS-the-displaced-geometry-each-frame, emitted-ANNOUNCES-absent-hair-geometry-rather-than-reporting-success, and a neg-control that the superseded RIGID block is gone. NOT ONE tooth asks whether a rendered hair vertex is inside the body, so no possible penetration can fail this gate and its GREEN is evidence of a connected pipeline, never of correct hair. THIS IS THE SAME CLASS AS THE OTHER TWO FINDINGS TODAY: hg_tick is arithmetically correct and has NO CALLER across 23622 files with corpus_complete=1, and the ship loop verifies a binary rather than a served artifact -- three green signals, three real defects, and in every case the subject measured was narrower than the claim made. DONE-RULE: a tooth counts rendered strand vertices lying inside the body capsules of the same fixture the gate already builds, the count is EMITTED with gv_kv beside its bar so a second method can contradict it, the tooth is BITE-PROVEN by mutating the clamp away and seeing the count RISE and the gate go RED, and an anti-vacuity control asserts the fixture actually reached a pose where penetration is possible -- a fixture whose hair never approaches the body would pass this tooth for the wrong reason and is not a test.
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR52 — The estate's searcher can read the artifacts the estate serves
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
st_longline_windowsAcceptance: OPERATOR 2026-09-04: invest in our tooling and do this properly rather than routing around broken instruments. MEASURED, FOUR DEFECTS HIT IN ONE SESSION AND ALL FOUR WORKED AROUND RATHER THAN FIXED: nx_shelltool grep declares linetoolong and SKIPS any line longer than its buffer, and the estate's primary shipped artifact -- an emitted world page -- is a handful of 200 KB lines, so the searcher is STRUCTURALLY BLIND TO EXACTLY WHAT WE SERVE and every absence claim over a served page is unprovable by it; the same verb takes a DIRECTORY and refuses a file path, which reads as the organ being missing; a directory scan reports files=200000, a ROUND DECIMAL that is a CAP printed in a field that reads as a count, and it was quoted as a population fact in this very session; and BUDGET-EXCEEDED returns corpus_complete=0 with matches=0, which is UNPROVEN and reads as absence. THE INCUMBENT IS WELL REASONED AND MUST NOT BE REPLACED: nx_shelltool.nx:200-210 explains that it reads in chunks and carries the trailing partial line so coverage is complete at any file size, and that it refuses to split a long line because half a line matched as a whole one is a FABRICATED match, the opposite failure and the worse one. That reasoning is right and the fix preserves it. DONE-RULE: at nx_shelltool.nx:250 the overlong branch stops discarding to end-of-line and instead scans the same line in OVERLAPPING WINDOWS carrying pn-1 bytes across each cut, pn-1 being exactly the largest pattern prefix that can straddle a boundary, so no match is missed and none is fabricated; a hit on an overlong line emits a WINDOW around the match rather than the whole line; linetoolong is retired as a coverage caveat and replaced by a longline= count that is INFORMATIONAL rather than a hole; files= and any other cap is renamed or accompanied so a ceiling can never be read as a census; and the whole thing is BITE-PROVEN by planting a pattern in the middle of a 200 KB line that the current binary MISSES and the fixed one FINDS, with a neg-control proving a pattern that straddles a window boundary is reported exactly ONCE.
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR53 — The estate reads its own served pages with its own wire: a sovereign CDP witness composed from the shipped WebSocket stack
Prerequisites: GR49 (acceptance unverified)
Contract:
nx_cdp_evalAcceptance: OPERATOR 2026-09-04, asked of the Python browser driver this seat leaned on all day: is it a tool we need to duplicate into nishi lang and use as a 3rd witness. MEASURED ANSWER, by listing buildroot/runtime rather than estimating: IT IS COMPOSITION, NOT CONSTRUCTION. Already shipped and already tested here are nx_websocket_client_upgrade.nx, which performs the RFC 6455 client handshake on an already-connected TCP fd and validates the 101 status plus Upgrade, Connection and Sec-WebSocket-Accept against the SHA-1 and Base64 transform of the key it sent, with nx_websocket_client_upgrade_test.nx beside it and ten named verdict constants; and nx_websocket_frame.nx, giving ws_parse_frame, ws_build_header and ws_apply_mask with the full opcode set, with nx_websocket_frame_test.nx beside it. Underneath sit nx_json.nx, nx_json_emit.nx, nx_base64.nx, nx_rand.nx, nx_net.nx and nx_syscalls.nx. ac_cdp.py's own header calls the handshake and frame codec THE ROUGHLY FORTY LINES BELOW and treats them as the hard part; that part is done, in NishiLang, with tests. THE ONLY MISSING GLUE, named so nobody rebuilds what exists: connect TCP to the DevTools port, GET /json/version or /json/list to resolve the page's webSocketDebuggerUrl, run the client upgrade on that fd, send one Runtime.evaluate with returnByValue true, then read frames until the matching id and hand back the result verbatim. CDP to a local browser is plain ws, so no TLS is required and nothing else is needed underneath. WHY THIS IS WORTH A RUNG RATHER THAN A NOTE, measured today and it is the uncomfortable half: nx_shelltool grep SKIPS any line longer than its buffer and every emitted world page is a handful of 200 KB lines, so THE ESTATE'S OWN SEARCHER IS STRUCTURALLY BLIND TO THE ARTIFACTS THE ESTATE SERVES, and the ONLY instrument that could answer whether our own code reached our own page was a third-party browser driven by Python. Every false claim this seat published about that question traces to exactly that blindness. THE BUILD PATH IS SOVEREIGN AND THE EVIDENCE PATH IS NOT, and nobody was counting the second half. DONE-RULE: nx_cdp_eval opens a page, evaluates one expression and prints the result verbatim with zero judgement of its own, exactly as the Python wire does; a neg-control proves it REFUSES rather than returns empty when the endpoint is absent, when the upgrade is rejected, and when the evaluate returns an exception; and the acceptance test is that it reproduces, byte for byte, the answer ac_cdp.py returns for the same expression on the same served URL -- two witnesses agreeing is the point, so the Python one is KEPT as the oracle and never retired. SCOPED AWAY from nx_browser: rendering the page with our own engine is the THIRD witness and the zero-translation rung of the operator's ladder, a separate and larger piece; this rung is our WIRE against their ENGINE and is buildable on the LAPTOP toolchain, which matters because a brand-new file has no laptop-versus-NAS fork to converge and NAS build admission refused every lane for hours on the day this was scoped.
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR54 — ONE HAIR SYSTEM: five parallel implementations collapse to one, and this rung is measured by DELETIONS not additions
Prerequisites: GR48 (acceptance unverified)
Contract:
hair_one_systemAcceptance: OPERATOR 2026-09-04, and the evidence for it is THIS SEAT'S OWN WORK: this feels like you are just dumping shit everywhere because you have no plan and we have multiple systems in play so sometimes one system gets the work sometimes another and thats garbage and not sota or any level of good development. HE IS RIGHT AND THE COUNT IS FIVE, ALL TOUCHED IN ONE SESSION BY ONE SEAT: a soft-tissue term added to the collision proxy in nx_nxa_rig_emit which is the RIGVIEW shader path; a child-offset clamp added to nx_game_page_emit which is emitted JAVASCRIPT on the BEACH path; vm_capsule_push added to nx_vecmath which is a FIFTH copy of closest-point-on-capsule; hd_tick_collide added to nx_hairdyn_lib which is the BAKE-time integrator; and nx_haircollide_gate built 12 of 12 GREEN as a gate for hg_tick, WHICH HAS NO CALLER ACROSS 23622 FILES WITH corpus_complete=1. FIVE ADDITIONS TO FIVE SYSTEMS IN ONE DAY, INCLUDING A GATE FOR DEAD CODE AND A FIFTH DUPLICATE PRIMITIVE, while the same seat catalogued duplicate-system defects as though observing from outside. ★★★★★★A SEAT THAT FIXES WHICHEVER SYSTEM IT HAPPENED TO BE READING PRODUCES MOTION IN EVERY SYSTEM AND PROGRESS IN NONE, AND EVERY SUCH FIX MAKES THE NEXT CONSOLIDATION MORE EXPENSIVE. THE RUNTIME TRUTH, MEASURED: the hair a visitor sees is built at load in hand-written JS from the rest-pose asset section and moved per frame by nxHairTick -- spring, viscous damping, wind, length clamp, and NO BODY TERM OF ANY KIND. The wasm collider hg_capd2 hg_inside hg_push with HG_CAP_MAX=32 capsules and HG_SIM=128 guides is UNREACHABLE. So four of the five systems cannot affect a served pixel and the fifth is outside the compiler. DONE-RULE, AND IT IS COUNTED IN DELETIONS: closest-point-on-capsule exists exactly ONCE in the estate and the other four sites are retired through nx_retire_path with the survivor named here; hg_tick has at least one production caller and a bite proves the gate goes RED when collision is disabled, so the collider stops being a comment; the rendered strand set is built by NishiLang lowered to wasm beside that collider with the emitted-JS geometry RETIRED once the wasm path reproduces its vertex buffer within a stated band on a fixture groom; and nx_haircollide_gate's subject changes from a dead function to the shipping path or the gate is retired with it. THE ACCEPTANCE TEST IS A COUNT THAT FALLS: capsule implementations 5 to 1, hair integrators 2 to 1, geometry builders 2 to 1, and hg_tick callers 0 to at least 1. A rung that ADDS a sixth anything has failed by construction. SUPERSEDES the scattered pieces this seat landed today -- GR25's clamp becomes a property of the one system rather than a patch on the JS one, and GR48 is the geometry half of this rung rather than a separate lane.
Authored effort: 4. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR55 — Skin-tone fidelity of the LIVE render is measured in Delta E and ITA error per class under named lights
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
skin_tone_fidelity_dEAcceptance: SEPT-2026 BAR, read from the mirrored paper: 19848 MetaHuman renders scored against Chicago Face Database references show fidelity collapsing for darker classes under non-reference lighting (class VI median ITA error 11.37 to 53.85) [@truetotone26]. Our generator spans the ITA classes by construction (nx_skin_ita, gate T4), so the gap is MEASUREMENT of the rendered result, not generation. DONE-RULE: the ITA ladder is rendered through the live beach path under three named lights, each render is scored in Delta E CIE76 and ITA error against its reference, per-class medians are EMITTED with gv_kv beside the published medians, and a neg-control proves a deliberately desaturated render RAISES the error -- an anti-vacuity tooth asserts the fixture spans at least four ITA classes so a single-tone ladder cannot pass
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR56 — Our subsurface term is calibrated against a measured face BSSRDF and its RMSE is published
Prerequisites: GR55 (acceptance unverified)
Contract:
skin_bssrdf_calibrationAcceptance: SEPT-2026 ORACLE: a polarimetric multispectral capture of dynamic faces fitting spatially varying single scattering and refractive index, validated at a mean reconstruction error of 0.028 against known-index objects [@polbssrdf25]. The oracle NEVER enters the build path: one measured BSSRDF is adopted as DATA with its license beside it, our skin model is fitted to it, and the RMSE of our fit against the measurement is the row's number. DONE-RULE: the fit RMSE is emitted with gv_kv, a neg-control proves a flat-albedo control fits WORSE by a named margin, and the fitted parameters are the ones the live path reads (a fit that never reaches the page is a comment)
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR57 — Sheer and stretched fabric over skin
Prerequisites: GR42 (acceptance unverified), GR14 (acceptance unverified)
Contract:
_ABSENT_:gr_sheer_bsdfAcceptance: The reference-still class this rung answers (2026-09-27) wears a sheer swimsuit: the garment is a THIN LAYERED MEDIUM, not an opaque surface. Reflectance core is skintwin ST5 km_layer_over (Kubelka-Munk two-flux layering of fabric over skin); transmission RISES with weave strain read from the drape solver (procgen PG29 gm_xpbd) and with wetness (GR58); the skin term is visible through the fabric in the beauty tier and the live tier. The field's bar this month: Eurographics 2026's texture-free woven-fabric model and neural garment fold inter-reflection, the 2024 real-time neural woven fabric and reflection-transmission capture lineage; nothing this month renders sheer fabric OVER skin with strain-driven transmission, which is the exceed axis. Done-rule: a rendered sheer swatch over a skin patch matches the two-flux prediction within a DECLARED DE2000 band at three strains, and the beach garment shows the skin term through it under the sky. [@kesen_eg2026]
Authored effort: 4. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR58 — Wet as a surface state on skin, hair and sand
Prerequisites: GR27 (acceptance unverified), GR16 (acceptance unverified)
Contract:
_ABSENT_:gr_wet_stateAcceptance: One wetness FIELD drives three materials: skin darkens and its specular sharpens (skintwin ST7 fo_wetness), hair clumps by capillary adhesion and its cuticle reads index-matched (skintwin ST9), sand darkens by the wet-porous term (GR27's wet-shore darkening generalised). The estate's wet-appearance scan read 2026-09-27 (11 entries, newest 2025-12, a homonym) holds no 2026 wet-appearance paper; the standing references are the micro-ellipsoid wet-porous model (2401.15628) and Jensen's 1999 wet-materials model, so this axis has NO current-month bar and says so. Done-rule: dry-to-wet DE2000 and specular-lobe shift published per material against mirrored measurements, and the field is one row of data the beach scene reads.
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR59 — Sun-tan and tan lines as a melanin field
Prerequisites: GR55 (acceptance unverified)
Contract:
_ABSENT_:gr_melanin_fieldAcceptance: The skintwin board's own goal names a tan line as an ABSORBER change, never a pixel edit. Melanin concentration is a per-vertex field DERIVED from garment coverage history and declared sun exposure (data rows), read by the skintwin chain (sf_fit, km_r_from_ks) and relit correctly by construction; the ISO/DTR 26478 skin-tone patches and NIST's 100 traceable skin reflectance spectra (250-2500 nm) are the calibration oracles skintwin ST6 pins. Done-rule: a tan line follows a garment edge, its inside-outside ITA difference lands in the band the skintwin referee (ST1) reports, and a relight moves both sides consistently.
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR60 — The support surface and its cover
Prerequisites: None declared; this does not establish execution eligibility.
Contract:
_ABSENT_:gr_support_coverAcceptance: The reference class reclines on a cushioned lounger: the cushion COMPRESSES under the body (softbody SB15, a support surface that spreads load) and its cotton cover DRAPES and wrinkles (procgen PG29 cloth XPBD, dcc DC10 pattern-sew-drape), with body-cushion contact free of interpenetration (softbody SB10; S4R-class resolution 2609.20524 is the September reference for procedurally assembled scenes). Done-rule: contact area and sink depth reported against the tissue and cushion stiffness rows, a non-trivial wrinkle field on the cover, zero interpenetration at the declared contact distance. [@arxiv_co_siga2026]
Authored effort: 4. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR61 — A camera model as data
Prerequisites: GR14 (acceptance unverified), GR23 (acceptance unverified), GR42 (acceptance unverified)
Contract:
_ABSENT_:gr_camera_rowAcceptance: A photograph is the SCENE TIMES A CAMERA. Exposure, sensor response and tone curve, focal length, aperture and depth of field, vignetting, flare and chromatic aberration become ONE camera row that the beauty tier (GR42) and the live tier read identically, on the linear-light pipeline (GR14) through the one colour transform (GR23). Done-rule: a rendered grey card and a photographed one agree within a DECLARED DE2000 band under the declared exposure, and depth of field matches a measured circle of confusion at two distances.
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR62 — Reference-still referee: per-region DE2000 against a licence-clean photograph class
Prerequisites: GR55 (acceptance unverified), GR30 (acceptance unverified)
Contract:
_ABSENT_:gr_refstill_refereeAcceptance: A reference photograph CLASS (licence-clean stills under the refintake lane's licence state, never an identity, never an unlicensed image) is measured per region in CIELAB: skin ITA, sand, fabric, cushion, sky. Every beach frame publishes its per-region DE2000 against that class beside the world referee grade (GR30), through the skintwin CIEDE2000 ruler (ST1 cs_de2000, built and promoted, ZERO adopters outside the garment lane). Done-rule: ST1 wired to the beach page, bands from measured references, the number on the page. This is the axis the operator's 2026-09-27 goal turns on: a real photo benchmarked, a published frame, no visible difference.
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.GR63 — Real-versus-render indistinguishability bench with outside liar-killers
Prerequisites: GR62 (acceptance unverified), GR61 (acceptance unverified), GR42 (acceptance unverified)
Contract:
_ABSENT_:gr_realrender_benchAcceptance: THE ACCEPTANCE RULER FOR THE OPERATOR'S GOAL (2026-09-27: benchmark a real photo, publish our frame, and not tell the difference). Pairs of licence-clean real stills and our frames of the same scene class (a synthetic human, never a pictured identity) are judged by (a) outside synthetic-image detectors run OUTSIDE the build path (NVIDIA's SIGGRAPH 2026 synthetic-video detector NIM is the named first oracle), (b) outside VLM judges through the referee panel, and (c) a human two-alternative forced choice by the operator and the team. The score is detection accuracy driven toward chance (500 permil) with the sample size declared, sequential looks corrected and an A/A calibration (gameengine GE51's arithmetic). Done-rule: a published bench page on /compare with the three arms, receipts and the receipt's date; the rung reads INCOMPLETE until all three arms exist and the operator's ACCEPT row is on the plane. [@nvidia_siggraph2026_blog]
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt
, SHA-256. Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
Learning and practice paths
No structured learning path is declared for this plan. Existing research, roadmap and worklog remain available above.
Rung dossiers — every record that bears on a rung, in one place
nx_swcompare_evidence graphics writes it beside its stamp); every gate below reads UNRECORDED until it runs.knowledge/organ_gate.conf read from this working directory (1 rows) and from the estate root (196 rows); rows in both counted once: 0; declarations read: 197. Every declaration in either copy counts, so a stub beside the publisher can never hide the root.GR1 Referee panel · cell PRESENT · gate NO-GATE-DECLARED
| Contract | cjc_panel · executor Organ · 1 u · stands on nothing |
|---|---|
| Definition of done | LANDED 2026-08-19 in one leg (estimate held): cjc_panel composes the statistics judge, the published Kovac skin-gamut ruler, image_beauty and nx_percept, forked per image; FP and FN measured on the labelled corpus v1 (n=10, 3 BAD by operator record, 7 GOOD = the six banked WebGL NXA rig stills captured on the real GPU + the published Mackenzie render) -- stats and symmetry proven non-separating and left report-only, skin and percept separable, floors 85/152 from the margins, composite FP=0 FN=0, ADMITTED under a floor-bound receipt; nx_cjc_panel_gate GREEN, bite-proven both ways. Open inside the rung: the VLM lens is UNOBSERVABLE until a sovereign VLM oracle exists (none on the estate, measured), and the floors need the second-corpus retest (MetaHuman-class stills) before they are believed |
| Matrix cell | REFEREE: multi-lens judge panel that can see beauty · runtime/nx_charjudge_census.nx · PRESENT decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_charjudge_census; declare nx_charjudge_census TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-09-04 measure THE DARK REFEREE FIRED, 2026-09-04, and its first full census SEPARATES THE TWO PIPELINES CLEANLY. nx_charjudge_census panel over 18 acorp captures: judged=18 failed=0 truncated=0, partition RECONCILES, admission=ADMITTED on cjc_panel_labels_union_tiered.list with one voting lens (percept) at floor 152. EVERY WORLD RENDER FAILS: ac_beach_char_r1 percept=68, ac_beach_world_r1=94, ac_craft_world_r1=131, ac_craft_char_r1=151 (one point short of the floor). EVERY CHARACTER CAPTURE PASSES: 171 181 177 195 196 199 214 219 226 228 316 328 337 350. Means 111 world versus 238 character -- the character pipeline scores 2.15x the world pipeline ON THE SAME REFEREE, and the fail/pass boundary falls exactly on the pipeline boundary. This is the operator's judgement ("/world/beach looks still below ps2 level graphics", "beach wasnt hit yet because the pipeline is broken") reproduced as a NUMBER by an instrument that had never been run. 2026-09-04 lesson A DARK REFEREE STARVES EVERY STATISTICAL ORGAN DOWNSTREAM, AND THE STARVATION READS AS A STATISTICS PROBLEM. Before this run the referee- plane held ONE ROW in its life (ref-face-1788160423), so nothing in the estate was contrastable and /compare/acceptance could not answer "are we improving" for any subject. The A/B core is NOT the blocker: nx_abstat_lib is 24/24 GREEN and is imported by exactly two organs (its own gate and nx_frameab, corpus_complete=1), i.e. LIB-GATE-ONLY -- the same shape the ranker reports for GR47 la_cholesky_q30. ONE RUN OF THE PANEL TOOK THE PLANE FROM 1 GRADE TO 19. THE FIRST REAL A/B PAIR NOW EXISTS IN THE DATA: op_before_0827t2019 versus op_after_0828t1034, head 171 to 214 and body 181 to 195, both IMPROVED -- two arms, so PAIRED-NOT-YET-POWERED against nx_abstat's NX_MC_MIN_N of 30, exactly as the acceptance page's own arm-state chip says. The remedy for an unanswerable improvement question was never a better test; it was running the instrument that produces the arms. 2026-09-04 retract I PUBLISHED TWO CLAIMS AN HOUR AGO THAT THIS CONF REFUTES, AND THE REFUTATION WAS ALREADY WRITTEN DOWN. (1) I read verdict=TIER-UNADMITTED on all 18 rows as a defect in the admission receipt. IT IS NOT: knowledge/cjc_panel.conf admits the tier set {textured} DELIBERATELY -- clay has 0 GOOD rows and photoreal 0 BAD rows, so both abstain BY DESIGN -- and I supplied a list of BARE PATHS with no tier field, so every row took tier_default=clay, the unadmitted tier. The panel behaved correctly and my input caused the abstention. (2) I reported beach percept=68 and 94 against floor 152 as a measurement of the served page. IT IS NOT: the conf names this exact trap, TWO RENDERERS ONE WORD -- clay rows are nx_world_snap NATIVE voxel captures while the SERVED page is TEXTURED because its NPC fragment shader samples the atlas, and the conf further records that the only clay rows are 240x150 captures predating the 2026-08-28 resolution fix, re-scoring 102 to 131 at full resolution. A tier whose rows predate a resolution fix cannot be compared with anything captured after it. WHAT SURVIVES: the census itself, 18/18 judged, failed=0, partition reconciles, and the clean separation of world-path from character-path captures. WHAT DOES NOT: any claim that 68 grades the beach page. 2026-09-04 measure THE REAL BLOCKER ON "IS OUR ACCEPTANCE SOTA", AND IT IS DOCUMENTED IN THE CONF RATHER THAN HIDDEN: THE REFEREE'S ONLY VOTING LENS IS SELF-REFUSING AND THE ADMISSION RECEIPT CANNOT SEE IT. floor_skin was REMOVED on 2026-08-25 (nine of nine rule/region combinations not separable), floor_stats and floor_sym were never separable, so percept is the SOLE voting lens -- stated in the conf as a real reduction in strength. And nx_percept ladder, the judge's own leave-one-out self-check and the very condition its header says percept is admitted ONLY WHILE it holds, returns 12/13 verdict=RED. The receipt binds floors, skin_lens, skin_rule and the tier set but NOT the ladder state, so the panel prints admission=ADMITTED while the one lens carrying every vote is failing its own ordering test. A SINGLE-LENS REFEREE WHOSE LENS IS SELF-REFUSING CANNOT MAKE ACCEPTANCE SOTA, AND NO AMOUNT OF A/B MACHINERY DOWNSTREAM REPAIRS IT. This is the rung to fix before any improvement verdict published from this panel means anything, and it must be fixed by earning a second lens a floor or repairing the ladder -- never by widening the ladder or dropping its check, which is loosening a gate to flatter a number. 2026-09-04 measure VERIFIED LIVE 2026-09-04, not quoted from the conf header, because a header is not a measurement and that claim was dated 2026-08-28: nx_percept ladder returns 12/13 REFUSED verdict=RED TODAY. The numbers: flat=-342 squares=-167 m2d=283 craft=196. L1 flat<squares GREEN, L2 squares<m2d GREEN, and all TEN held-out commercial references pass their leave-one-out (loo 100 to 372, every one above squares). THE SINGLE FAILING CONJUNCT IS L3, m2d<craft, WHICH IS THE OPERATOR ORDERING: the judge scores m2d 283 ABOVE craft 196 and the operator ranks craft higher. So the sole voting lens of the sole referee fails exactly the one check that encodes the human judgement it exists to reproduce, while the admission receipt -- which binds floors, skin_lens, skin_rule and the tier set but NOT the ladder -- keeps printing ADMITTED. The gap is 87 permil on one ordered pair, it is NOT a broad disagreement with humans, and that is why widening the ladder would be the wrong fix: eleven of twelve orderings are already right. 2026-09-04 measure L3 ROOT-CAUSED BY BUILDING THE MISSING INSTRUMENT, AND IT REFUTED MY OWN HYPOTHESIS. nx_percept gained an additive read-only `anchors <frame>` verb printing cosine to all 12 banked anchors with their class; shipped through the guarded lane (content=GREEN, behave=IDENTICAL, live_sha==staged_sha, _offc twin byte-identical, 31896 to 38600 B). I EXPECTED A SUBJECT MISMATCH -- 9 of 10 GOOD anchors are commercial CHARACTER renders and craft is a world frame -- AND THAT IS WRONG. MEASURED: craft GOOD-mean 866 versus m2d 894, only 28 permil apart, and veloren1 (the one voxel-world anchor) is craft's SECOND-HIGHEST GOOD match at 926, so the GOOD side is working exactly as designed. THE GAP IS ON THE JUNK SIDE. Of the 66-permil score difference, 28 comes from GOOD and 38 from JUNK, and within JUNK it is flat_control ALONE: craft-to-flat 617 versus m2d-to-flat 476, a 141-permil gap, while on squares_control m2d is actually the CLOSER of the two (672 versus 608). THE FLAT-COLOUR CARICATURE CONTROL IS PENALISING craft FOR ITS SHADING MODEL RATHER THAN ITS QUALITY: a flat-shaded voxel world is legitimately close to flat colour in hue-histogram and run-length statistics, but that anchor was banked to represent NO CONTENT, not A FLAT-SHADED RENDERING STYLE, and the ruler cannot tell those two apart. 2026-09-04 measure L3's CAUSE IS ONE FEATURE FAMILY, MEASURED BY A NEW `families` VERB, NOT ARGUED. nx_percept gained `anchors` and `families` (both additive, read-only, shipped through the guarded lane: content=GREEN both times, live_sha==staged_sha, 31896 to 38600 to 41216 B) plus an EXTRACTED pj_centroids so the scorer and the decomposition cannot disagree about the centroids. NEUTRALITY PROVEN ON REAL DATA, not asserted: after the extraction `ladder` reprints flat=-342 squares=-167 m2d=283 craft=196 and all TEN leave-one-out values byte-identical (294 163 346 100 347 321 246 336 372 341). THE DECOMPOSITION, diff = cos_good minus cos_junk per family: hue 7 dims craft -161 versus m2d +490, a 651-permil swing to m2d; edge 4 dims craft +178 versus m2d -34, a 212-permil swing to CRAFT; blockdelta 24 dims +59 vs +68; runlength 8 dims +872 vs +907. ONE FAMILY CARRIES THE ENTIRE DISAGREEMENT. craft's hue vector is CLOSER TO THE JUNK CENTROID THAN TO THE COMMERCIAL ONE (junk 812, good 651) because a voxel world's palette IS a handful of flat block colours, which is statistically what the flat-colour control is. And the family that AGREES with the operator is edge, where craft leads by 212 -- but edge is 4 dims of 43, so it is outvoted 39-to-4 BY CONSTRUCTION. 2026-09-04 lesson THE REMEDY L3 NEEDS IS NEITHER OF THE TWO THAT LOOK OBVIOUS FROM A SCORE, AND THE DECOMPOSITION IS WHAT SEPARATES THEM. NOT a floor to retune -- the floor is not in the disagreement. NOT an anchor to re-bank -- I hypothesised a SUBJECT MISMATCH (9 of 10 GOOD anchors are commercial CHARACTER renders, craft is a world frame) and the per-anchor verb REFUTED IT: craft GOOD-mean 866 versus m2d 894, and veloren1, the one voxel-world anchor, is craft's second-highest match at 926. The real shape is TWO STRUCTURAL FACTS: (1) HUE IS CONFOUNDED WITH THE JUNK CONTROL -- flat_control exists to represent NO CONTENT, and a flat-shaded rendering STYLE is statistically indistinguishable from it on a 7-bin hue-order histogram, so every flat-shaded render is scored as partly junk for its shading model rather than its quality; (2) EDGE IS UNDER-DIMENSIONED FOR ITS DISCRIMINATIVE VALUE -- it is the only family that reproduces the operator ordering and it carries 4 of 43 dims, so a family that is RIGHT is outvoted by three that are confounded or near-tied. ★★★★★★A COSINE OVER CONCATENATED HISTOGRAM FAMILIES WEIGHTS EACH FAMILY BY ITS DIMENSION COUNT, WHICH IS AN ARTEFACT OF HOW THE FEATURE WAS WRITTEN AND NOT A STATEMENT ABOUT WHAT DISCRIMINATES -- so the family carrying the human judgement can be correct and still lose every vote. Fixing this by widening the ladder or dropping its check would be loosening a gate to flatter a number; the honest rungs are a junk control that is degenerate WITHOUT being flat-shaded, and a per-family weighting earned on a labelled corpus rather than inherited from dimension counts. 2026-09-04 land FIVE EXTERNAL LIAR-KILLERS MIRRORED AND CONTENT-PINNED, 2026-09-04, on the operator's framing: "i dont want these built into our build pipeline but they should all stand as liar killers and benchmarks and oracles ... to help us avoid long costly unproductive building by giving us frames of refrence". PINS, all status=200 with provenance rows: webgpu_inspector 363369 B h3b560de731a812469f4d22b1f1017327fe9ec581d6753130a4a839f7f94fe441 -- binaryen 519176 B h6aa2d7836ec42efa13c301e5ca617417cc1c2fcd38dc7112e170431b1e808c51 -- emscripten 382705 B he883dd0ee08c7482e37e73991ef50f798238ab0ba26cd1784a008ce9e38542be -- meshoptimizer 689176 B h1d3e5434a035ee0b4f397cdaa8bfe95f58462acea8bd33c653ef23032377719d -- dawn 14923 B h44aaee866414aba755aa1646b193af8f8eecd75544e694c5ca6d549b676f7a77. WARNING ON THE LAST ONE: 14923 B is small enough to be the stub signature (login wall, JS shell, bot challenge) and it is NOT yet verified as a real index; it is recorded UNVERIFIED rather than cited, because a wrong mirror is worse than none. THE CLASSIFICATION MATTERS MORE THAN THE LIST, and it turns on a fact the industry survey assumed wrongly: OUR CORE ENGINE IS NISHILANG, not C++, Rust or TypeScript. So emscripten (compiles C/C++), wasm-bindgen (Rust) and esbuild/Vite (a JS/TS host layer we deliberately do not have) CANNOT BE TOOLCHAIN HERE -- they are BARS ONLY. What transfers as an executable liar-killer is narrower and better: binaryen wasm-validate can FALSIFY "our emitted wasm is spec-valid" and wasm-opt can falsify "our output is tight"; webgpu_inspector can falsify "the GPU receives the commands we claim"; Dawn plus RenderDoc or PIX can falsify a frame-time claim at hardware level. RANKED BY WASTED BUILDING PREVENTED, binaryen is FIRST: it answers whether nx_compile_wat's output is valid at all, in one call, against an emitter I edited today for LN35, and craft_mt.wasm (263990 B, already served) is a subject sitting right there. ★AND THIS IS THE SAME DEFECT THE WHOLE SESSION HAS BEEN FINDING FROM THE INSIDE: the referee has ONE voting lens, it disagrees with the operator, and there is NO SECOND METHOD CLASS -- external oracles ARE that second class, which is why they belong on the board as data rather than in anyone's head. 2026-09-04 measure THE ACCEPTANCE REFEREE IS A SINGLE-LENS JUDGE WHOSE ONLY LENS IS SELF-REFUSING, AND ADMISSION DOES NOT CHECK IT -- so every acceptance verdict today is unsound at the root and the operator's ask, validating whether we are improving in A B, is structurally unanswerable for a reason that is NOT missing captures. READ VERBATIM FROM knowledge/cjc_panel.conf, the panel's own data file, not inferred: nx_percept ladder, the judge's OWN leave-one-out self-check and the very condition the organ header says percept is admitted only while it holds, currently returns 12 of 13 verdict=RED with L3 failing m2d 283 greater than craft 196 against the operator ordering; and the admission receipt binds floors, skin_lens, skin_rule and the tier set but NOT the ladder state, so the panel reports admission=ADMITTED while its only voting lens is self-refusing. THE SINGLE-LENS PART IS ALSO MEASURED IN THAT FILE: floor_stats and floor_sym were never separable and are report-only, and floor_skin was REMOVED on 2026-08-25 after a nine-cell rule-by-region grid over the union corpus came back NINE OF NINE NOT SEPARABLE. So percept is the sole vote by elimination, and the file states the reduction honestly rather than hiding it. CONSEQUENCE FOR THIS BOARD: the /compare/acceptance ledger reads subjects=5 agreed=0 not_agreed=5 with four subjects at ref_rows=0, and the natural reading is that it is starved of captures. THAT READING IS WRONG AND WOULD SEND THE NEXT SEAT AT THE WRONG WORK: even a perfect capture graded today is graded by a judge failing its own ladder. TWO THINGS I ALSO SETTLED WHILE CHECKING THE TIER, both from the same file. ONE, THE TIER TRAP IS REAL AND ALREADY NAMED THERE: TWO RENDERERS, ONE WORD -- nx_world_snap's NATIVE path is flat lambert voxel with no atlas and is CLAY, while the PRODUCT is the served WebGL page which samples an npc atlas and is TEXTURED with floor 151 admitted at FP=0 FN=0 over 8 rows. Submitting a native snap as textured would grade the wrong renderer, so I did not. TWO, THE CLAY TIER'S NAMED BLOCKER IS NOW UNBLOCKED BY AN ARTIFACT ALREADY ON DISK: the file says clay is unadmitted with 2 BAD and 0 GOOD, that its only rows are 240x150 captures predating the 2026-08-28 resolution fix, and that GROWING CLAY MEANS RE-CAPTURING IT AT FULL RESOLUTION FIRST. This session's nx_world_snap char beach run returned w=1920 h=1200 q=1 res_permil=1000 res_class=FULL and wrote beach.crop.png with CHAR-CROP pixels=29857 fill_permil=671 -- a full-resolution clay capture, which is exactly the artifact named as missing. IT IS DELIBERATELY NOT LABELLED HERE: the corpus admits a row only on a named human witness and forbids relabelling to make a tier separable, so the GOOD or BAD call is the operator's and this row records the artifact, never a verdict. ORDERED CONSEQUENCE: fixing the ladder outranks collecting captures, and binding the ladder state INTO the admission receipt outranks both, because an admission that cannot see its lens self-refusing will re-admit the same broken judge on the next calibration. 2026-09-04 measure THE LADDER RED IS LIVE TODAY, NOT A STALE 08-28 CLAIM, AND ITS FAILURE IS ONE COMPARISON. Re-measured 2026-09-04 against the promoted binary rather than inherited from cjc_panel.conf, because a rung inherited without re-measurement cost this session a full investigation earlier today when GR39 turned out to be already fixed. VERBATIM: flat=-342 squares=-167 m2d=283 craft=196 with L1 flat<squares GREEN, L2 squares<m2d GREEN, L3 m2d<craft THE OPERATOR ORDERING RED, and all ten held-out commercial references passing their leave-one-out at loo=100..372 above squares. So the judge is internally consistent on every reference it holds and disagrees with the operator on exactly ONE pair: it ranks the 2D render ABOVE craft where the operator ranks craft above it. HYPOTHESIS, LABELLED AS ONE AND NOT PUBLISHED AS A FINDING, but newly available because this seat finally LOOKED at the served frame today: the capture is roughly four fifths repeating yellow voxel sand, and nx_percept's 43 dims are whole-frame statistics -- hue 0 to 6, blockdelta 7 to 30, edge 31 to 34, runlength 35 to 42. A score dominated by frame-wide hue and block statistics would rank a monochromatic tiled terrain low however good the characters in it are, while the operator judges the CAST. That is the SAME SUBJECT-MISMATCH CLASS as the other three defects measured today -- the collider that runs on simulated guides while the page renders interpolated children, the hair gate whose every tooth tests wiring while the operator judges pixels, and the ship loop that verifies a binary while the question is the served artifact. THE DECISIVE TEST IS NAMED AND CHEAP so the next seat does not re-derive it: score the CHARACTER CROP against the FULL FRAME and see whether the L3 ordering flips. The crop already exists at full resolution from this session, beach.crop.png, 122x526, fill_permil=671, CHAR-CROP pixels=29857. If the ordering flips on the crop the judge's subject is the frame and the remedy is a region, not a retune; if it does not flip the hue hypothesis is refuted and the fault is in the feature set. I COULD NOT RUN IT: the families decomposition verb was added to nx_percept source earlier this session and is NOT in the promoted binary, because build admission has refused every lane all afternoon on procs_blocked=8 of 8. ★AND THE ATTEMPT EXPOSED A SEPARATE DEFECT WORTH ITS OWN FIX: nx_percept SILENTLY ACCEPTS AN UNRECOGNISED VERB AND RUNS ladder INSTEAD, returning byte-identical output, so a caller cannot distinguish a verb that does not exist from a verb that agrees with the default. That is the estate's own permissive-default law -- an unknown mode must REFUSE or fail CLOSED, never fall through -- and it means every future probe of this organ can silently measure the wrong thing. 2026-09-04 land A REFEREE VERDICT IS PUBLISHED TO /compare/acceptance FOR THE FIRST TIME THIS SESSION, AND THE LANE I DECLARED CLOSED WAS NEVER CLOSED. THE CORRECTION FIRST, BECAUSE IT IS THE REUSABLE PART: I published that this seat holds no contentput.cap and therefore could not deliver acceptance captures. FALSE. I had enumerated nishi-core/nxc2/*.cap, THIRTEEN files, and the real capability store is ~/.nishi/ with FOUR HUNDRED AND SEVENTY THREE caps including contentput.cap and a live admin.cap for nx_mgmt valid to epoch 1790000000. The estate's own memory note states the lesson verbatim -- BEFORE DECLARING A LANE CLOSED, ENUMERATE WHAT THE CAPS YOU HOLD CAN REACH -- and I quoted that note while reading the wrong directory. ★★★★★★A CAPABILITY INVENTORY TAKEN FROM ONE DIRECTORY IS A SAMPLE, AND A SAMPLE THAT COMES BACK EMPTY READS EXACTLY LIKE A DENIAL. THE WORKING LANE, END TO END, SO NO SEAT REDERIVES IT: capture the SERVED page in a real browser with nishi-ops/ac_cdp.py evaluating document.getElementById(c).toDataURL(image/png), redirect to a file and base64-decode locally so no image data crosses the context; upload with nx_content_put_client.py <local> <dest> ~/.nishi/contentput.cap; then nx_accept referee <subject> <NAS-ABSOLUTE-path> <tier> <seat>. ⚠THE DESTINATION IS THE ONLY TRAP AND BOTH REFUSALS NAME THEIR OWN REMEDY: content-put allows world/ knowledge/rigcorpus/ knowledge/fetched/ knowledge/library/ and /tmp/ while nx_mkdirp allows knowledge/ sites/ and /tmp/, so the intersection that ALREADY EXISTS and is DURABLE is knowledge/fetched/ -- choosing it avoids creating a directory merely to satisfy a path picked arbitrarily, and /tmp/ would be reaped by the 3600 s tmpstorereap beat. WHAT LANDED, VERIFIED FROM THE LEDGER AND NOT THE RECEIPT: ACCEPT-REFEREE subject=locomotion percept=339 floor=151 tier=textured composite=GOOD frame=sha256 619e543a222bcc573fb3b5d0c4830dff5d34ce9802d3440c3f043352f19755d8 capture_bytes=302621 id=ref-locomotion-1788533606, and a re-run of nx_accept list now reads referee=OK ref_rows=1 for that subject with the estate ledger moving ref_rows 1 to 2. Full lens vector recorded: cjp_stats=350 cjp_comp=360 cjp_pal=554 cjp_cont=350 cjp_face=1000 cjp_skin_center=856 cjp_fig_permil=874 cjp_sym=525 cjp_percept=339 cjp_vlm=-1 vlm=UNOBSERVABLE. ★SUBJECT FRAMING IS PART OF THE MEASUREMENT AND WAS CHOSEN, NOT DEFAULTED: the wide capture shows twelve figures mid-stride and is the RIGHT framing for locomotion and the WRONG framing for hair or skin, where each girl is roughly forty pixels tall -- grading strands from a wide shot is the subject-mismatch defect this whole session has been finding, so hair garments and skin wait for the review-framed close capture instead. ⚠TWO CAVEATS TRAVEL WITH THE GOOD AND ARE NOT FOOTNOTES: percept clears its floor but the judge producing it FAILS ITS OWN LADDER at 12/13 REFUSED measured today, so the row is honest evidence whose validity is bounded by that; and cjp_vlm=-1 means the VLM lens did not run, so this is a single-lens verdict exactly as cjc_panel.conf states. 2026-09-04 measure THE ACCEPTANCE REFEREE SCORES THE FRAME AND NOT THE SUBJECT, PROVEN BY THREE IDENTICAL NUMBERS, AND THE OPERATOR VERDICTS IT IS COMPARED AGAINST MAY NOT BE THE OPERATOR'S. TWO FINDINGS, both measured today, and together they say the acceptance surface cannot answer the question it exists for. FINDING ONE, THE STRUCTURAL ONE: four referee rows were landed from two browser captures of the SERVED textured page -- ref-locomotion-1788533606 percept=339 from the wide frame, then ref-hair-1788533863, ref-skin-1788533878 and ref-garments-1788533886 ALL percept=337 from the SAME review-framed capture sha256 52b5246d8a95a44138507b5c2f35827cc84210bf225816b888f19937904fb230. THREE SUBJECTS, ONE CAPTURE, IDENTICAL SCORES TO THE UNIT. nx_accept referee takes a subject argument but the lens measures the WHOLE IMAGE, so the subject selects which ROW IS WRITTEN and never what is MEASURED. ★★★★★★CONSEQUENCE: how is the hair, how is the skin and how are the garments are ONE MEASUREMENT WEARING THREE LABELS -- improve hair alone and all three move together, improve nothing and they still agree, so per-subject A B is impossible BY CONSTRUCTION and no amount of capture collection fixes it. This is the SAME CLASS as every other defect measured today and now the fifth instance: the collider runs on simulated guides while the page renders children; the hair gate's every tooth tests wiring while the operator judges pixels; the ship loop verifies a binary while the question is the served artifact; the percept ladder scores whole frames while the operator ranks casts; and now the referee scores a frame while the row claims a subject. IN EVERY CASE THE THING MEASURED IS NARROWER OR SIMPLY OTHER THAN THE THING CLAIMED. FINDING TWO, THE PROVENANCE ONE, AND IT IS WORSE: the operator stated plainly today that he did NOT reject hair. The accept- plane carries beach-hair-colour-wind-0830 REJECT 2026-08-30 actor=operator with the words poorly done; hair clips the body and moves like a mop, not strand by strand. The operator's ACTUAL recorded words that day, in CLAUDE.md, are ONE general statement about FOUR things -- these are all poorly done, doing it poorly isn't a complete, it's not complete till you and I and the nishi team agree. Two of the four rows echo that sentence almost verbatim; the HAIR row carries a SPECIFIC TECHNICAL DIAGNOSIS that appears nowhere in it. ★★★★★ONE UTTERANCE WAS SPLIT INTO FOUR VERDICTS AND EACH WAS GIVEN PER-SUBJECT WORDS ATTRIBUTED TO THE OPERATOR. nx_accept deliberately has NO verb that writes this plane, precisely because the operator owns it, but anything that can write the store can stamp actor=operator -- so the operator signature of the three-party agreement carries NO PROVENANCE beyond a self-declared field, and on this evidence it was synthesised. COST, MEASURED ON THIS SEAT: that row is why this entire session was spent on a hair child-offset clamp. A requirement no operator stated drove a day of work through his name. WITHDRAWN IN THE SAME BREATH: I published earlier today that the referee grading hair GOOD while the operator rejects it PROVES a judge-operator disagreement. It does not, because one side of that comparison may not be the operator's verdict. OWED, and neither needs a build: the referee must measure a SUBJECT REGION rather than a frame, and the accept- plane needs provenance a seat cannot forge. 2026-09-04 land PER-SUBJECT ACCEPTANCE MEASUREMENT IS ACHIEVED AND NEEDS NO BUILD -- AND THE SAME EXPERIMENT INVALIDATES THE FOUR ROWS I PUBLISHED AN HOUR AGO. THE FIX: nx_accept referee scores WHATEVER IMAGE IT IS GIVEN, so per-subject measurement is per-subject CROPPING, and the convention already existed -- the only pre-existing referee row, face, was graded from a HEAD CROP at ac_op_after_0828t1034_head.png. PROVEN, not proposed: the same review capture graded whole returns cjp_percept=337 for hair, skin AND garments identically, while a crop of the red-head's head and hair region -- box 348,100 to 440,215 of the 735x368 capture, upscaled 4x, uploaded as ac_beach_hair_crop_0904.png sha aa2a8a22 -- returns cjp_percept=226. THE WHOLE VECTOR MOVES, not just the composite: stats 166 to 16, cont 166 to 16, pal 615 to 289, sym 550 to 288. So the referee CAN discriminate subjects and the earlier finding stands in corrected form: it scores the frame it is HANDED, and handing it a frame for a subject question is the caller's defect rather than the organ's. ★★★★★★AND THE SAME NUMBER CONVICTS MY OWN ROWS. face, graded from a CROP, scored 214. This hair crop scores 226. Both close crops, same band. The four rows I landed earlier -- locomotion 339, hair 337, skin 337, garments 337 -- were graded from WHOLE FRAMES and sit roughly 110 points higher PURELY FROM FRAMING. cjc_panel.conf warns of exactly this in its own words, framing matters for the skin lens, COMPARE LIKE FRAMINGS, and the floor of 151 was earned on a crop-based corpus. ⇒THOSE FOUR GOOD VERDICTS ARE INFLATED BY FRAMING AND ARE NOT EVIDENCE THAT ANY SUBJECT CLEARS ITS BAR. They are honest captures measured the wrong way, and they must be re-graded from subject crops before any of them is read as a result. I am recording that against my own work rather than leaving four flattering numbers on the board. ★AND THE CROP SHOWED WHAT THE WIDE SHOT STRUCTURALLY COULD NOT: at 4x the hair mass passes THROUGH the shoulder and chest silhouette rather than draping over it, it reads as a painted shell with vertical banding rather than strands, the torso carries no form shading at all, and there is no shadow anywhere in the crop. The clipping is REAL and VISIBLE at close framing and INVISIBLE at scene framing -- which is why the review camera the operator dislikes for testing is the CORRECT camera for acceptance capture, and why a wide capture can never referee a strand-level subject. OWED NEXT, no build required: re-grade locomotion skin and garments from their own subject crops, and record the framing class ON every referee row so a crop score and a frame score can never again be compared as though they were the same measurement. 2026-09-04 land NINE REFEREE ROWS ARE IN THE PLANE AND THE PER-SUBJECT LIMIT IS NAMED HONESTLY RATHER THAN CLAIMED AWAY. STATE, read from knowledge/store/referee- which is the AUTHORITY, never from nx_accept list which lags it: face 214 head-crop 2026-08-31 by seat-accept-lane; then this session, locomotion 339 and hair 337 and skin 337 and garments 337 all FRAME-graded, then hair 226 head-crop, skin 229 and garments 229 and locomotion 229 all FIGURE-crop. WHAT IS PROVEN: the capture-upload-referee lane works end to end with content CAS at every hop, and DISTINCT CROPS PRODUCE DISTINCT SCORES -- 226 for the hair region, 229 for the figure region, 337 for the whole frame, with the entire lens vector moving and not merely the composite (stats 166 to 16, cont 166 to 16, pal 615 to 289, sym 550 to 288 between frame and hair crop). WHAT IS NOT PROVEN AND MUST NOT BE READ AS PROVEN: skin, garments and locomotion ALL read 229 BECAUSE THEY SHARE ONE FIGURE CROP. I made TWO distinct crops, not five, so I demonstrated discrimination between REGIONS and NOT between SUBJECTS -- three subjects still sit on one measurement wearing three labels, which is the very defect this work set out to fix, surviving one level up. TRUE PER-SUBJECT A B REQUIRES A CROP PER SUBJECT: torso skin, garment area, legs and stride. ★★★★★AND THE FRAMING CLASS MUST BE RECORDED ON THE ROW OR THE PLANE IS A TRAP: the floor of 151 was earned on a crop corpus, so the four FRAME rows at 337 to 339 are inflated by roughly 110 points of framing and are NOT evidence any subject clears its bar. They stay in the plane as honest captures measured the wrong way, superseded by the crop rows, and a framing field on every row is owed so a crop score and a frame score can never be compared as though they were one measurement. ★SIXTH INSTANCE OF THE SESSION'S ONE CLASS, AND IT NEARLY COST A FALSE CLAIM: nx_accept list reported ref_rows=6 and skin percept=337 while the referee- plane held the skin crop row at 229. TWO SPACED READS OF THE LEDGER AGREED WITH EACH OTHER AND BOTH WERE WRONG, exactly as the law predicts -- N agreeing reads from the SAME instrument are not independent evidence. I was one message from publishing that the crop rows did not land; the plane refuted it. ⇒/compare/acceptance CAN PUBLISH A SUPERSEDED REFEREE VERDICT while the current one sits in the plane, and that is a defect in the reader, not the writer. ⚠PROVENANCE, UNCHANGED AND STILL OPEN: ref-face-1788160423 carries actor seat-accept-lane, so even the pre-existing referee row was written by a lane rather than by a human witness, consistent with the accept- plane's actor=operator field being self-declared and unforgeable-by-nothing. 2026-09-04 land PER-SUBJECT ACCEPTANCE A B IS OPERATIONAL AND IT PRODUCED ITS FIRST FAILING VERDICT, WHICH AGREES WITH THE OPERATOR. Five subjects graded from FIVE DISTINCT SUBJECT CROPS of one browser capture of the SERVED textured page: garments 143 BAD from the swimsuit region, skin 201 GOOD from shoulders and upper torso, face 214 GOOD from the pre-existing head crop, hair 226 GOOD from the head and hair region, locomotion 230 GOOD from legs and feet. AN 87-POINT SPREAD ACROSS SUBJECTS THAT PREVIOUSLY ALL READ 337 IDENTICALLY, and each row carries its own multivariate lens vector so the method is multivariate per subject rather than one composite. ★★★★★★THE RESULT THAT MATTERS: garments graded from its OWN region returns 143 against floor 151, composite=BAD -- THE FIRST FAILING REFEREE VERDICT ON THIS BOARD -- and the accept- plane carries garments as REJECT. THE REFEREE AND THE OPERATOR NOW AGREE ON THAT SUBJECT. Three framings of one subject settle the mechanism: whole frame 337 GOOD, figure crop 229 GOOD, garment-only crop 143 BAD. ⇒THE JUDGE WAS NOT BROKEN FOR THIS SUBJECT; IT WAS BEING HANDED THE WRONG PICTURE, and every earlier GOOD was a measurement of a scene that happened to contain the subject. This is the direct remedy for the frame-versus-subject defect logged earlier today and it needed NO BUILD -- only a crop. ⚠WHAT THIS DOES AND DOES NOT SETTLE, stated so nobody over-reads it: it settles that the referee CAN discriminate subjects and CAN disagree with a flattering answer, which is what makes it usable as an A B instrument at all. It does NOT settle the percept ladder, which still reads 12/13 REFUSED against the operator ordering, so a GOOD from this panel remains bounded by that; it does NOT settle provenance, since actor on every referee row here is a seat and the accept- plane's actor=operator is still self-declared; and the four FRAME-graded rows from earlier today remain in the plane at 337 to 339, superseded and inflated by roughly 110 points of framing, kept as honest captures measured the wrong way rather than deleted. ★THE REPRODUCIBLE METHOD, four steps, no build, any seat: ac_cdp.py evaluates canvas toDataURL on the served page and the base64 is decoded locally; PIL crops the subject region and upscales it; nx_content_put_client.py uploads to knowledge/fetched/ under contentput.cap with a content-CAS receipt; nx_accept referee <subject> <NAS-absolute> textured <seat> grades it. Every hop verifies by hash, and the referee- plane is the AUTHORITY -- nx_accept list lagged it by two rows during this very sequence and reported a superseded score. |
GR2 The leap: genome splat avatar · cell WATCHING · gate NO-GATE-DECLARED
| Contract | gs_avatar_genome · executor Organ · 4 u · stands on GR1 |
|---|---|
| Definition of done | RIGHT-SIZED 2026-08-18 to the published bar (the earlier budget understated by ~10x): v2 by canon-mode bodyfit search (16-dimensional ring-scale under the judge), the gated face pipeline composed in AT >=100K-gaussian face density (GaussianAvatars-class heads), strand-aligned splat GROOM hair at groom scale (85 percent of the industry poly budget), 8K-class skin with wrinkle response; done when the panel grades it above the ring figure AND the banked WebGL reference on the same framing -- declared before the experiment. THE BAR IS NOW DATA (leg 1, 2026-08-19, admitted panel on knowledge/cjc_gr2_framing.list, every row a figcrop): ring figure skin 7 percept -112 BAD; splat v1 skin 118 percept 123 BAD (skin-coloured, perceptually junk-ward); the six WebGL rig crops skin 279-661 percept 210-228 GOOD; Mackenzie skin 624 percept 281. DONE-RULE IN NUMBERS: v2 crop must read composite GOOD with percept above 228 (the best banked rig crop) and skin at or above 279 (the weakest), re-measured with the same conf and receipt |
| Matrix cell | SPLAT: genome-driven splat avatar -- THE LEAP · runtime/nx_gsplat.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_gsplat; declare nx_gsplat TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-09-02 measure OPERATOR 2026-09-02, looking at /world/beach?review=1&cast=1&hud=1: that looks like our junk cpu stuff. SEEN BY THE SEAT THROUGH THE GECKO WIRE (readPixels off the page's own GL context, 2235x1117, WebGPU tier gpu=1): a flat sand plane rendered as high-frequency speckle noise with no albedo structure, no cast shadows except dark blobs under feet, no sky in frame at default pitch, no fog or depth cue, four figures with hard-edged painted garments, ribbon hair and flat cartoon shading, a hotbar palette in the corner. The native voxel frame from nx_world_snap (1920x1200 seed 20260728) is a block world: cubic trees, block snow, block clouds. The heightfield renderer (nx_worldgen, sovereign Infinigen rung 1, built and run on the laptop in 29 s for 6 worlds) is continuous terrain with fBm, lake, sky gradient and sun, but the march step (t += 55 + t/48) terraces every slope into visible stairs. THE VERDICT IS STRUCTURAL, NOT COSMETIC: the live beach is the GAME-tier voxel engine, and no amount of tuning makes a voxel raymarcher photoreal. 2026-09-02 measure REFEREES, honestly: nx_uiq_visual says RENDERED (its own env line: guardrails-not-perceptual); nx_frame_score gives the voxel frame composite 542 BASIC and its own calibration line says values above about 600 order by statistics, not perceived quality; nx_percept's ladder is 12/13 REFUSED (the judge does not reproduce the human ordering). THE ESTATE HAS NO CALIBRATED ENVIRONMENT-REALISM REFEREE, so a photoreal claim cannot be graded today by anything but eyes; that referee is a prerequisite rung, not an afterthought. 1970-08-23 candidate-gap Native procedural body candidate t374 generated through nx_body_proc + nx_body_gen: canon build=650 musc=500 noise=250 seed=374, 1700mm, 20 parts, 110 control rings, 1,807,104 triangles, truncated=0; artifact buildroot/knowledge/gates/beach-female-body-t374.nxmesh SHA d426a6bbada2d3fe0f525664caad535d8ea55edb6c4d762bd1ff2b7fc9e83679. Face gate 4/4 GREEN and deterministic face render job1789881061 landed 114173B. Existing nx_gltf_export is write-only; complete native absence scans jobs1789881227/1789881233 found no GLB importer, decoder or GLB-to-NXMSH2 owner. Face Alpha composition remains UNVERIFIED until a reusable adapter preserves identity/material/rig contracts and passes served review. No Beta or Production promotion. 1970-08-23 determinism Repeated nx_body_gen for t374 with identical canon/parameters produced 1,807,104 triangles and byte-identical 159,025,240-byte mesh; nx_filehash cmp identical=1, SHA d426a6bbada2d3fe0f525664caad535d8ea55edb6c4d762bd1ff2b7fc9e83679. Determinism is qualified for this recipe/seed only; anatomy, rigging, materials and visual acceptance remain open. 1970-08-23 multiview Native twin-face render for seed374 at yaw45 landed 114638B via job1789881347; front yaw0 landed114173B via job1789881061. Same identity vector and skin values retained. These are deterministic view artifacts, not visual-quality acceptance; matched Alpha/Beach served review still required. 1970-08-23 adapter-contract Shared prepared-character runtime contract read from buildroot/runtime/nx_character_prepared_runtime.nx: actor input requires pos/aj/aw/nrm/uv/idx streams, optional palette/bind, image/normal materials, geometryMode and shared GPU buffers; static geometry without source image is explicitly supported only when geometryMode=static. A GLB importer must produce these streams, preserve scale/units/material provenance and refuse unsupported skin/animation claims. This is the adapter contract; no implementation or Alpha promotion claimed. 1970-08-23 build-contract Native nx_sov_build_run job1789881827 correctly rejected standalone nx_gltf_export with nxasm rc102 UNDEFINED label: main. The exporter is a library, not an executable; this is a build-entry contract, not a compiler regression. Adapter qualification must use a main-bearing gate/consumer that imports the library. Build admission passed with native policy checks and bounded host measurements. 1970-08-23 json-prerequisite Native nx_sov_build_run qualification job1789881917 built and ran nx_json_test through nx_cc->nxasm_x86, no GCC, run-exit=0. `nx_json: emitter + tokenizer round-trip PASS`; staged artifact only, serving root deliberately unchanged. This qualifies the existing JSON tokenizer prerequisite for the future GLB accessor resolver, not the GLB adapter itself. 1970-08-23 hires-candidate Native nx_body_gen job1789882088 produced a higher-resolution procedural candidate: height1700mm, radial128, sub64, relief300, 20 parts, 110 control rings, 7,202,304 triangles, truncated=0, headroom1,186,304, artifact633,802,840B SHA483425e30501be39a12561cc70f5d3703231de32b52a2726134d9f0882695c7f. This improves geometric sampling only; no skin, rig, material, motion or visual acceptance claim. Resource cost must be measured before repeated use. |
GR3 4D splat motion · cell WATCHING · gate NO-GATE-DECLARED
| Contract | g4_avatar_rig · executor Organ · 1.5 u · stands on GR2 |
|---|---|
| Definition of done | The splat avatar moves with the same gait, breath and expression machinery the mesh uses; replay bit-identical |
| Matrix cell | SPLAT: 4D animated splats driven by the rig · runtime/nx_gsplat4d_lib.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_gsplat4d_lib; declare nx_gsplat4d_lib TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR4 Second species canon · cell WATCHING · gate UNRECORDED
| Contract | bp_species_canon · executor Organ · 1 u · stands on nothing |
|---|---|
| Definition of done | A horse canon through the identical pipeline; a bodybench-style ruler against a CC anatomy oracle of the species proves species-agnostic generation |
| Matrix cell | BODY: a second species through the identical canon engine · runtime/_hdl_build/nx_body_proc.nx · WATCHING decl |
| Gate | nx_aesthetic_canon_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | no log row names this rung yet |
GR5 Part-genome creatures · cell WATCHING · gate NO-GATE-DECLARED
| Contract | cg_genome_parts · executor Organ · 2 u · stands on GR4 |
|---|---|
| Definition of done | Typed interpolatable parts over the existing plans; distinct genomes must measure distinct and interpolation must stay anatomically legal |
| Matrix cell | BODY: interpolatable part-genome across creature types · runtime/nx_creaturegen.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_creaturegen; declare nx_creaturegen TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR6 Indoor rooms · cell PRESENT · gate NO-GATE-DECLARED
| Contract | ig_room_solve · executor Organ · 2 u · stands on nothing |
|---|---|
| Definition of done | Constraint-solved rooms with materials through our engine, deterministic and bounded; REFUSES an unsatisfiable constraint set rather than overlapping furniture |
| Matrix cell | WORLD: indoor rooms by constraint solve -- the Infinigen 2.0 expansion · runtime/nx_indoorgen_lib.nx · PRESENT decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_indoorgen_lib; declare nx_indoorgen_lib TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR7 Advances archive mirrored · cell PRESENT · gate NO-GATE-DECLARED
| Contract | gp_collect2 · executor Organ · 0.5 u · stands on nothing |
|---|---|
| Definition of done | Second root through the ingest organ, fixed-point crawl, partition reconciles per root; the extractor then cites it |
| Matrix cell | RESEARCH: SIGGRAPH Advances course archive mirrored beside it · runtime/nx_gfxpapers_ingest.nx · PRESENT decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_gfxpapers_ingest; declare nx_gfxpapers_ingest TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR8 PBR map set per genome · cell PRESENT · gate NO-GATE-DECLARED
| Contract | ntb_bake_pbr · executor Organ · 1 u · stands on nothing |
|---|---|
| Definition of done | Four maps by four regions regenerated per genome at load; normal-map lighting proven through the existing image path |
| Matrix cell | MATERIAL: full PBR map set regenerated per genome · runtime/nx_nxa_texbake_lib.nx · PRESENT decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_nxa_texbake_lib; declare nx_nxa_texbake_lib TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR9 Pre-integrated SSS · cell LANDED · gate NO-GATE-DECLARED
| Contract | sr_sss_preint · executor Organ · 1 u · stands on GR8 |
|---|---|
| Definition of done | Bound to the baked normal map; carries the away-side equality tooth (the T66 form) so no uniform brightener can fake it |
| Matrix cell | RENDER: pre-integrated subsurface skin · runtime/nx_sdfrender.nx · LANDED decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_sdfrender; declare nx_sdfrender TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR10 Path-traced GI · cell LANDED · gate UNRECORDED
| Contract | gp_bounce_gi · executor Organ · 2 u · stands on nothing |
|---|---|
| Definition of done | Per-frame bounce beyond cached radiosity; the cache-vs-rebuild equality stays as the determinism tooth |
| Matrix cell | RENDER: per-frame path-traced dynamic GI · runtime/nx_gi_path.nx · LANDED decl |
| Gate | nx_gi_path_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | 2026-09-27 measure paced nx_frontier_scan attest graphics 2026-09-27: bars=7 fetches=112 venue-tag fan-out rows=49 UNREACHABLE=0 newer_sources=8 stale_bars=1 verdict=STALE-BARS; the stale bar was SB-RTGI (2026-07): the SIGGRAPH Asia 2026 comment tag answered Gaussian Light Transport 2609.11430 published 2026-09-10, and the bar's own listing re-fetched today also reads newest=2026-09; the unpaced run had reported that tag UNREACHABLE (30 of its first 40 venue-tag rows), an abstention nx_barfresh cannot convert into STALE 2026-09-27 land SB-RTGI re-adjudicated to 2026-09 from the mirrored bytes (knowledge/fetched/rf_a2bd0c41.raw, pinned as ref arxiv_siga_rtgi_2026_09_27, abstract read through nx_arxiv_digest): the September entry expresses GI as a 13D Gaussian mixture fitted on the rendering-equation residual and renders view-independent solutions in milliseconds; July's Split Radiance Cascades and OctaOctree stay named in the bar; nx_barfresh graphics re-read after the edit is the acceptance 2026-09-27 land ref arxiv_siga_rtgi_2026_09_27 re-pinned 2026-09-27: the fleet refs gate read PROV-MISMATCH because the row cited the scanner's raw capture (rf_a2bd0c41.raw) under the feed's self link while the fetch journal holds that mirror under the export.arxiv.org query; the same query was re-fetched through nx_research_fetch as knowledge/fetched/cmp_graphics_siga_rtgi_202609.xml (2782 B, sha256 4a6d43bf) and the row now cites the journaled url, mirror and pin. LESSON: cite the url the researcher fetched, never the url the feed prints about itself 2026-09-27 queue fleet refs gate 2026-09-27 (37/44 RED) names six PIN-MISMATCH rows that pin knowledge/library/gfx_*.html: graphics.refs terraindiff26, kerbl3dgs, karis2021, bitterli2020 and procgen.refs terraindiff26, kerbl3dgs. Those library files are rewritten by the papers-index beat (gfx_sig2026.html 288701 B today carries no NX-DERIVED stamp yet differs from every declared pin), so a pin on them is stale by construction. Remedy for the row owners: cite the Ke-Sen listing mirror in knowledge/fetched (cmp_graphics_kesen_sig2026.html, sha256 7e6c166c, immutable) and re-pin; the beat-written library artefact should carry the NX-DERIVED stamp so no future row pins it 2026-10-01 measure OCTOBER ATTESTATION 2026-10-01: nx_barfresh graphics verdict FRESH, 7 of 7 bars ATTESTED (it read STALE with 7 UNATTESTED before this seat looked), every listing read by the seat; the morning's organ attests could not reach arXiv and export.arxiv.org answered this hour only after 429 retries. Literal readings: SB-RTGI 2026-09 (its own venue-tag query, totalResults 1, Gaussian Light Transport 2609.11430 of 2026-09-10), SB-HUMAN 2026-09 (GenStream 2609.18634 arXiv-posted 2026-09-16, originally ACM MM 2025; on-subject September entries such as Heartian 2609.28539 of 2026-09-22), SB-ADVANCES-2607 2026-07 (course page byte-identical to its 2026-09-27 pin), SB-VENUES-2609 2026-09 (the SIGGRAPH Asia 2026 venue-tag listing, 59 entries, 29 in September, newest 2609.37167 of 2026-09-29). Two refs RE-POINTED with the reason in each row: SB-VENUES-2609 from kesen_sig2026, a static July index that can never carry a September month, to arxiv_co_siga2026, whose entries set it; metahuman from https://www.unrealengine.com/en-US/metahuman, which answered 403, to the MetaHuman 5.8 release notes the estate can read. Reading-month convention, as on this morning's vendor bars on aesthetictwin, skintwin and anatomy: infinigen-princeton (newest dated entry 2026-08, the 2.0.0a2 release), metahuman (5.8 of 2026-06-17) and siggraph-rtr (the July 2026 course) carry newest=2026-09 meaning nothing newer than the September reading, the true newest dated entry named in each barscan (queue row 1790828670). RANK by nx_compare_rank (artefact graphics.rank 24361 B sha256 f84a967e870a085f793abf4993c286da82a3dc9e642b4c1fbeca6576e50ed8c9, target 0.1 milestone M1, 63 rungs, done 13, open 50, finish 3, unmapped 7, PM filed 1 of the top 3): the critical path is one rung, #1 GR2 the genome splat avatar (gs_avatar_genome, priority 1000, DARK-CONTRACT); staged next GR27 ground material 3800, GR14 linear-light pipeline 3000, GR42 the beauty tier's first frame 2025, GR15 sovereign strand system 2000, GR61 camera model as data 1000; the reference-still human rungs rank low on value alone (GR60 support surface 125, GR63 real-versus-render bench 100, GR57 sheer fabric over skin 75, GR58 wet state 66), and every open rung is TREATMENT-UNDECLARED with sponsor=none. 2026-10-01 queue READING-MONTH CONVENTION FOR VENDOR AND REFERENCE BARS NEEDS A DECISION (an estate-wide ruler question for the nx_barfresh owner, raised by the seat that applied it): the barscan grammar says newest is the newest entry's month and nx_barfresh attests only when it equals the bar's month, yet the vendor and reference bars (here infinigen-princeton, metahuman, siggraph-rtr; on aesthetictwin VISIA, 3dMD, Artec and others; on skintwin and anatomy the MetaHuman, UE5 and DARPA rows) were dated by the month they were read, so an honest literal barscan (2026-08, 2026-06, 2026-07 here) reads UNATTESTED, and the only literal fix is re-dating each bar to its newest dated entry, as this seat did for aesthetictwin MB8 and SB-FACEANAT-2609. Both readings now stand on the boards. Decide one: re-date every vendor and reference bar to its newest dated entry (strict, the 2026-09-06 aspect-sentiment precedent), or give barscan an explicit baseline-unchanged kind that nx_barfresh accepts and prints as its own state, so a reading-month bar is never mistaken for a newest-entry bar. Until decided, each convention barscan names the true newest dated entry in its note. |
GR11 Native-resolution GPU delivery · cell WATCHING · gate UNRECORDED
| Contract | cw_webgpu_compute · executor Organ · 1 u · stands on nothing |
|---|---|
| Definition of done | HALF LANDED 2026-08-18: the WebGL2 tier is LIVE DEFAULT on all five worlds and the WebGPU WGSL world pass is LIVE LEVERED (?gpu=1, real-GPU verified via the CDP wire, boot-stall ceiling hardened); REMAINING: the compute-shader form, the NPC+eye passes to parity so WebGPU can default, and the same frame bit-compared against the CPU path at reference size |
| Matrix cell | WORLD: native-resolution delivery on the visitor's GPU · runtime/nx_craft_emit.nx · WATCHING decl |
| Gate | nx_worldvis_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)nx_npc_atlas_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | 2026-09-06 retract OPERATOR 2026-09-06: the WebGPU lane that delivers https://nishifamily.com/world/beach?gpu=1&review=0 is named as a rolling disaster (bad output, random coding) despite every benchmark, oracle and industry document being available -- a heavy-investment climb that stalled short of SOTA, one of the key pillars this work addresses from September 2026 research. On this board GR11 stays HALF LANDED as written and its LIVE LEVERED half reads as PRESENCE only until the same frame bit-compares against the CPU path, the CTS and WPT webgpu suites pass on the real GPU, and the operator ACCEPTs; the ecosystem stall detector (EC40) reads this board's investment against its rank velocity 2026-09-06 queue SOTA RE-BASELINE, September 2026: pin the current WebGPU and WGSL specifications, the WebGPU CTS and WPT webgpu suites and the reference implementations as refs with years; name the oracles this board already holds and did not consume (nx_wgsl_pixel_gate, nx_wpt_runner, the CDP real-GPU wire, the referee panel, the CPU-path bit-compare); split GR11 REMAINING into gated rungs each carrying an oracle receipt; a rung without a receipt never reads done, and a one-step landing without a measured outcome ranks below a longer rigorous one (engineshift ES27) 2026-09-08 requirements September 2026 Beach production requirements B26-RENDER/STREAM/SAND/CHARACTER/HAIR/HOT/DELIVERY/REVIEW captured in knowledge/compare/sovereign-human-world-contract-20260907.md (47776 bytes after append1788872418). Primary-source corrections separate compute capability from guaranteed speed, texture arrays from sparse residency, writeBuffer from zero-copy, device loss from one-frame rollback, and particle-push shaders from MPM/XPBD. Each requirement needs source and served identities, dependency/owner, comparable hardware/workload, quality and correctness oracles, costs and next experiment. Existing graphics/character/softbody/deploy/toolchain domains own the work; no new random pages. This is requirements capture, not frontier parity or delivered Beach improvements; comprehensive current benchmark research remains open. 2026-09-08 requirements Operator target: integrated Beach WebGPU above60FPS with intelligent resource use, assessed on declared device/browser/resolution/NPC-count/scenario and quality tier. Budget for60FPS is16.67ms per frame; report raw intervals and p50/p95/p99/worst, missed-budget frames, CPU simulation/submission and actual GPU intervals when exposed, unavailable counters, memory/residency, network and energy where measurable. Average FPS alone cannot qualify smoothness. Persistent Nishi research must discover dated primary publications, implementations and shipped references; maintain capability coverage, query/source history, access/publication dates, supersession edges, unresolved gaps and evidence freshness. Re-evaluate affected comparisons with pinned reproducible workloads when relevant methods change; do not rerun everything or infer parity from new links. As-of-today ranking requires current evidence and an explicit survey boundary. Native research automation and comprehensive coverage remain unverified; keep deployment and Beach delivery as current implementation priority. 2026-09-08 research WebGPU4K60/Wasm source audit2026-09-08: https://emscripten.org/docs/porting/multimedia_and_graphics/WebGPU-support.html describes Emdawnwebgpu C bindings on top of browser WebGPU, not direct driver access. https://toji.dev/webgpu-best-practices/buffer-uploads.html recommends writeBuffer for Wasm; mapped uploads can require an extra heap copy, staging rings unmap/copy/remap rather than remain concurrently GPU-accessible. Do not claim zero-copy or guaranteed native speed from webgpu.h or prescribe replacing Nishi compiler. https://ir.pku.edu.cn/handle/20.500.11897/738236 dates FusionRender to2024; https://websplatter.github.io/ and https://arxiv.org/abs/2602.03207 identify a distinct2026 Gaussian-splatting pipeline; original pasted WebSplatter link incorrectly pointed to FusionRender. https://arxiv.org/abs/2604.02344 studies LLM dispatch overhead, not proof that4K Beach rendering is rarely GPU-bound. Benchmark native4K3840x2160 separately from upscaled output; retain internal resolution, quality and raw frame intervals. Candidate experiments: batch dirty uploads, bundles for compatible repeated commands, GPU resident simulation/culling with indirect draws, pipeline preparation and bounded staging reuse. Select by measured bottleneck and total quality/cost, not mandatory-method slogans. Full paper reproduction and current broad coverage remain open; this audit does not assert a delivered renderer change. 2026-09-08 priority Operator sequencing clarification2026-09-08: Beach big rocks before microdetail. Integrated priority: reliable scoped delivery and measurement; ground/collision stability with no unintended sinkholes or falling through terrain; reproducible procedural beach variants; coherent shoreline/wave/wetness interaction; grounded character movement and broad lighting/material quality; then finer hair/follicle/skin detail. Reuse existing terrain/gameengine/procgen/water owners and dependency graph; rank by user-visible severity, unlocked dependencies, recurrence and measured total cost per accepted result. Terrain qualification must compare rendered surface and collision surface across seeds, terrain seams, slope transitions, spawn placement, walking/running, timestep changes and stream/unload/reload; distinguish intentional depressions from defects. Procedural qualification pins seed/config/version and tests materially different beach layouts with coherent terrain, water level and contact, using Houdini and other verified procedural systems as workflow/capability references rather than name-only scores. Current runtime defect cause and external parity remain unverified. Fine-detail requirements remain retained but must not displace these blocking integrated outcomes. 2026-09-08 benchmark-target Operator selected False Earth https://github.com/momentchan/false-earth as an external Beach direction, implemented through NishiLang emission. README inspected2026-09-08 describes GPU procedural terrain/grass, shared terrain sampling for blade attachment, stable PCG seeds/grid snapping, distance LOD, character grass push, lighting/postprocessing and adaptive DPR. Map these to existing terrain/procgen/vegetation/renderer owners. Compare matched traversal and camera views for terrain continuity, vegetation anchoring and motion, interaction, shadows, temporal stability and scene coherence. Record actual internal render resolution separately from output resolution because adaptive DPR can hide reduced pixel workload; pin commit, configuration, browser, GPU, seed and capture before timing comparisons. README claims are not reproduced benchmark results. Study mechanisms and build native Nishi equivalents; do not transplant its React/Three/TSL stack or separately licensed astronaut asset. Reference deliberately depicts transforming surreal terrain, so its ground behavior is not an oracle for stable physical Beach ground. Current first defect remains continuous rendered terrain versus voxel player-contact height, reproduced locally with a256Q8 discrepancy; no served correction claimed. 2026-09-08 research Water foundation read2026-09-08: Mark Finch, GPU Gems chapter1 https://developer.nvidia.com/gpugems/gpugems/part-i-natural-effects/chapter-1-effective-water-simulation-physical-models . GPU geometric waves plus fine normal detail; wavelength/amplitude/direction/speed constraints, analytical derivatives and bounded Gerstner steepness are reusable baseline concepts. Historical shipped Uru technique, not September2026 frontier proof or complete shoreline fluid simulation. Beach acceptance should share parameterized water state between visual displacement and contact/buoyancy queries, verify normal derivatives and nonfolding bounds, preserve temporal continuity during parameter changes, and separately qualify shoaling/breaking/runup/foam and wet sand. Implement via native Nishi emission; compare newer methods by workload, quality and cost rather than adding wave count alone. 2026-09-08 benchmark-target Operator selected https://www.shadertoy.com/view/Ms2SD1 as an ocean visual reference. Direct fetch failed2026-09-08, so current original source/render was not inspected. Discoverable copies identify Seascape/TDM and carry restrictive attribution/noncommercial/sharealike headers; original identity/version/terms require primary verification before any code reuse. Treat as visual target only pending capture: broad swell, fine ripples, reflective horizon, crest shading and temporal coherence at measured resolution/cost. Nishi implementation remains native; separate visual reference acceptance from interactive shoreline, buoyancy and breaking-wave physics. No copied shader or verified FPS claim. 2026-09-08 reference-pending Operator reference https://www.shadertoy.com/view/wt3XDj requested2026-09-08. Direct fetch returned402; title, author, contents, appearance, terms and performance remain unverified. Retain exact shader ID for future authorized access; do not invent classification or benchmark scores from an inaccessible URL. Evaluate relevance to current Beach blockers before implementation. |
GR12 Expression morphs · cell PRESENT · gate UNRECORDED
| Contract | fa_expr_morph · executor Organ · 2 u · stands on GR2 |
|---|---|
| Definition of done | FACS axes over the identity morphs, graded against the 710-expression FACS-class basis MetaHuman ships (the bar, not a smile bone); the panel grades expressions readable |
| Matrix cell | MOTION: FACS facial expression morphs · runtime/nx_faceanat.nx · PRESENT decl |
| Gate | nx_lipratio_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | no log row names this rung yet |
GR13 Skin-bound dynamics · cell LANDED · gate NO-GATE-DECLARED
| Contract | dbs_displace_free · executor Organ · 1 u · stands on nothing |
|---|---|
| Definition of done | CONTRACT SYMBOL CORRECTED 2026-09-01 (board owner): this rung declared dyna_bind_skin while its matrix row named runtime/nx_nxa_dyna.nx, a pair that could never resolve, so GR13 held ranker slot #8 as OPEN for a capability that had shipped. Measured full-population, corpus_complete=1: func dyna_bind_skin( is declared once, at runtime/nx_softbind.nx:238, and it ticks three soft-tissue REGION points -- not this rung's subject; func dbs_displace_free( is declared once, at runtime/nx_dyna_bind_skin.nx:203, and is the per-vertex half that procgen.matrix and softbody.matrix already carry LIVE. DONE-RULE, unchanged in substance and now pointed at a gate rather than at a declaration: the proven solver bound to skin VERTICES with collision against the body it rides, the flip being the receipt and nx_dyna_bind_skin_gate on the graphics.gates roster being the proof -- a green cell over a red gate is not done |
| Matrix cell | MOTION: dynamics bound to skin vertices · runtime/nx_dyna_bind_skin.nx · LANDED decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_dyna_bind_skin; declare nx_dyna_bind_skin TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR14 Linear-light pipeline · cell WATCHING · gate UNRECORDED
| Contract | sr_linear_light · executor Organ · 1 u · stands on nothing |
|---|---|
| Definition of done | All lighting math computed in linear, display-encoded exactly ONCE at output; proven by the tonemap A/B that failed 2026-08-26 (ACES over display-encoded values washed the frame -- the banked double-transfer lesson) now passing: filmic curve applied and the A/B screenshot referee grades it richer, not paler; prerequisite to GR9 |
| Matrix cell | RENDER: linear-light pipeline, display-encoded once at output · runtime/nx_game_page_emit.nx · WATCHING decl |
| Gate | nx_game_page_emit_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | 2026-10-01 queue CLAIMED by seat colour-pipeline (cross-board program step 4 toward gameengine GE83): organ-level proof only -- sRGB encode and decode known answers incl. the linear-segment knee, a compositing case where blending in linear light and blending encoded values differ by a computed amount, and a neg-control tooth that catches a double display-encode; the page-level A/B screenshot referee in the done-rule needs a browser capture this headless seat does not have and is NOT claimed here |
GR15 Sovereign strand system (TressFX-referenced) · cell WATCHING · gate NO-GATE-DECLARED
| Contract | hr_strand_sys · executor Organ · 2 u · stands on nothing |
|---|---|
| Definition of done | The Infinigen-fork hair chain, leg 1: a NishiLang strand system whose sim and render structure is read from the mirrored TressFX source (the canonical shipped open real-time strand system) -- guide strands + per-strand constraints on the one solver, rendered with real width and coverage in our WebGL2 tier; done when a girl-framed capture is graded by the panel above the current ribbon baseline and no ad-hoc constant lacks a cited row |
| Matrix cell | HAIR: sovereign strand system vs TressFX bar · runtime/nx_softbind.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_softbind; declare nx_softbind TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR16 Marschner-class hair shading · cell WATCHING · gate UNRECORDED
| Contract | hr_marschner_lobe · executor Organ · 1.5 u · stands on GR15 |
|---|---|
| Definition of done | Hair BRDF per Marschner 2003 through the Karis 2016 / Tafuri 2019 production reductions (mirrored): dual-lobe spec with shifted primary and secondary highlight, azimuthal roughness; constants pinned as cited conf rows before the shader edit, gate compares lobe positions against the published reduction |
| Matrix cell | HAIR: Marschner-class shading · runtime/nx_game_page_emit.nx · WATCHING decl |
| Gate | nx_game_page_emit_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | no log row names this rung yet |
GR17 Groom style zones with pinned dynamics · cell WATCHING · gate NO-GATE-DECLARED
| Contract | gr_style_zones · executor Organ · 1.5 u · stands on nothing |
|---|---|
| Definition of done | nx_nxa_groom gains the cosmetology-brief zone model (hairline, part, nape, density map) with strand dynamics constants extracted from the mirrored Super-Helices / Discrete Elastic Rods papers and PINNED as oracle rows (closing the conf's own uncited hair_k_q10 correction record); style presets are parameters, every band cited or derived |
| Matrix cell | ROADMAP: groom style zones with pinned dynamics constants · runtime/nx_nxa_groom.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_nxa_groom; declare nx_nxa_groom TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR18 Infinigen fork map as data · cell WATCHING · gate NO-GATE-DECLARED
| Contract | if_fork_map · executor Organ · 1 u · stands on nothing |
|---|---|
| Definition of done | The fork strategy is a LIVE census, not prose: a conf mapping each Infinigen subsystem (ProcFunc generators, constraint solver, materials, creatures, ground-truth writers, Infinigen-Sim articulation) to its sovereign NishiLang twin, its governing SOTA reference, its gate, and its measured state -- rendered on this board and re-measured at emit; the differentiators (inside-out anatomy, genome identity, bit-exact determinism, the adult invariant) carried as exceed rows |
| Matrix cell | FORK: Infinigen subsystem map live census · runtime/_hdl_build/nx_graphics_board.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_graphics_board; declare nx_graphics_board TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR19 Render the sections we already generate · cell WATCHING · gate NO-GATE-DECLARED
| Contract | ntb_view_sections · executor Organ · 2 u · stands on nothing |
|---|---|
| Definition of done | MEASURED 2026-08-28: ntb_view finds only VERT and TRIS, so a groomed and dressed asset renders IDENTICALLY to a bare one -- nv=14164 tris=28092 filled3=56482 for both, with 4800 generated hair strands and a 648-vertex measurement-fitted garment present in the file and ignored. We GENERATE hair and clothing and cannot SEE them server-side, so GR15-GR17 and the garment lane have no judgeable artifact and the referee panel can never be pointed at them. Done: ntb_view consumes GVRT/GTRI and HSTR (strands as camera-facing ribbons), reports gverts/gtris/strands SEPARATELY from the body counts, refuses on index overrun, and stays a no-op when the sections are absent; proven by the emitted PNG differing and the reported triangle count strictly exceeding the bare asset's on the same input. Garment reuses the existing rasteriser; the open design question is that tm_set_image is global, so garment and hair need either a flat-region UV or a per-vertex vertex-colour path |
| Matrix cell | ROADMAP: render the sections we already generate -- groom and garment must reach the viewer · runtime/nx_nxa_texbake_lib.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_nxa_texbake_lib; declare nx_nxa_texbake_lib TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR20 Garment and hair penetration ruler · cell WATCHING · gate NO-GATE-DECLARED
| Contract | cc_garment_pen · executor Organ · 1.5 u · stands on GR19 |
|---|---|
| Definition of done | nx_clipcheck already measures the two GROSS self-clipping classes and on the shipped ref9.nxa returns verdict=CLIP (44064 keys, torso_clip_keys=146 worst 21mm, floor_clip_keys=2176 worst 30mm) -- but its axes are limb-through-torso and foot-through-floor. The operator-reported defect is GARMENT-through-body and HAIR-through-body, and NOTHING in the estate measures it: a ruler exists for the neighbouring defect and reads RED while the reported defect has no ruler at all. Done: per-vertex penetration of GVRT against the body surface and of HSTR roots/segments against the scalp, reported as a count plus worst depth in mm plus the offending element index, before and after XPBD settle, with a neg-control that a deliberately oversized garment reports MORE penetration than the fitted one |
| Matrix cell | ROADMAP: garment and hair penetration ruler on the fine axes · runtime/nx_clipcheck.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_clipcheck; declare nx_clipcheck TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR21 Genome asymmetry -- synthetic faces are too perfect · cell WATCHING · gate UNRECORDED
| Contract | bg_asym_permil · executor Organ · 1 u · stands on nothing |
|---|---|
| Definition of done | MEASURED 2026-08-28 on the ADMITTED referee panel (cjc_panel, floors from knowledge/cjc_panel.conf): cjp_sym reads 938 for our generated body and 913 for both MetaHuman frames against 650 for the photographic bar -- every synthetic subject is ~1.4x more bilaterally symmetric than a real person, and the panel measures it today with no rung consuming the signal. This is a realism defect that costs NOTHING to fix in a procedural generator and is structurally unavailable to a finite asset library: a real face is asymmetric and ours is machine-perfect. Done: a per-mille asymmetry gene drives independent left/right ring-scale and feature offsets from the seed, deterministic and reproducible from the genome; proven by cjp_sym moving toward the photographic band without cjp_percept falling below the admitted tier floor |
| Matrix cell | ROADMAP: genome asymmetry -- synthetic faces measure too perfect against the photographic bar · runtime/_hdl_build/nx_body_gen.nx · WATCHING decl |
| Gate | nx_chargen_param_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | no log row names this rung yet |
GR22 Camera rejection sampling · cell WATCHING · gate NO-GATE-DECLARED
| Contract | ntb_cam_reject · executor Organ · 0.5 u · stands on nothing |
|---|---|
| Definition of done | Infinigen's published figures are honestly non-cherry-picked at the IMAGE level, but the VIEWPOINT is rejection-sampled best-of-ten by pixel-depth variance -- so part of the quality gap a reader attributes to their renderer is a composition filter. Cheapest visual win available: the emit path renders N candidate viewpoints, scores each by depth variance over the figure, and keeps the best. Done: N and the score are conf rows not literals, the chosen index and its score are printed so the choice is auditable, and a neg-control shows the WORST-scoring viewpoint grades lower on the referee panel than the best -- if it does not, the heuristic is not measuring what it claims and the rung fails |
| Matrix cell | ROADMAP: camera rejection sampling -- the composition filter behind the rival figures · runtime/nx_nxa_texbake_lib.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_nxa_texbake_lib; declare nx_nxa_texbake_lib TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR23 One colour transform, two execution surfaces, ours · cell WATCHING · gate NO-GATE-DECLARED
| Contract | sr_view_transform · executor Organ · 1.5 u · stands on GR14 |
|---|---|
| Definition of done | SOVEREIGNTY RULE APPLIED (operator 2026-08-28: we dont want webgl2 and other 3rd party unless we are recreating ourselves the capabilities). The transfer curve and the LUT are DERIVED AND GENERATED IN NISHILANG; AgX is an ORACLE we check ourselves against, never a dependency -- borrow the rulers, not the runtime. TWO INDEPENDENT RIVALS CONVERGE ON IT, which is why it is worth recreating: Infinigen never sets a view transform and inherits the Blender 4.2 AgX default with compositor grading OFF (read byte-for-byte in our mirror: distort False, color_correct False), and VR HOT deliberately ADDED AgX in 0.13 on 2026-06-16 calling it a color profile you might know from other high end games. Done: ONE NishiLang implementation of the curve consumed by BOTH the page-side surface (beach lane) and the server-side portrait renderer (GR19), byte-identical LUT from one generator, so the two surfaces CANNOT diverge; the native renderer is the DURABLE home and the browser page is a stopgap execution surface. GR14 is the prerequisite and carries the measured lesson: the 2026-08-26 A/B failed because ACES was applied over display-encoded values and washed the frame -- transform over LINEAR light, display-encode exactly once |
| Matrix cell | ROADMAP: one colour transform derived and generated in NishiLang, two execution surfaces · runtime/nx_sdfrender.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_sdfrender; declare nx_sdfrender TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-10-01 queue CLAIMED by seat colour-pipeline (cross-board program step 3 toward gameengine GE83): mirror the AgX reference as an OUTSIDE oracle, then build the transfer curve and the LUT generator in NishiLang as a lib plus a gate whose teeth read oracle values from the mirrored bytes; in flight, nothing shipped yet |
GR24 Identity parameter space with correlated sampling · cell WATCHING · gate UNRECORDED
| Contract | bg_identity_space · executor Organ · 2 u · stands on nothing |
|---|---|
| Definition of done | THE INTELLIGENT-PROCGEN HALF, and the axis a finite asset library structurally cannot enter. MEASURED RIVAL POSITION: VR HOT ships roughly 15-20 hand-picked skins accumulated 2-4 at a time over five years on a FIXED LICENSED TOPOLOGY (Daz Genesis 8.1, vendor-stated in their own 0.5.4 patch notes), so their identity space is bounded by what one marketplace authored; Infinigen has NO humans at all. Ours is a seed. Done: semantic master parameters (age, adiposity, sun exposure, grooming, dimorphism) fan out into correlated low-level axes through a FULL COVARIANCE draw rather than independent per-channel sampling -- the mechanism behind why a rival palette reads deliberate instead of tinted -- with each axis drawing from its own split stream so adding a parameter never reshuffles existing genomes; every character reproducible and auditable from its genome, proven by bit-identical regeneration from the seed and by the referee panel showing identity spread across a population without any member falling below the admitted tier floor |
| Matrix cell | ROADMAP: identity parameter space with correlated sampling -- the axis a finite asset library cannot enter · runtime/_hdl_build/nx_body_gen.nx · WATCHING decl |
| Gate | nx_chargen_param_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | 2026-09-04 land THE CORRELATED IDENTITY DRAW LANDED AS A GATED CORE: nx_identity_space_lib.nx 8264 B plus nx_identity_space_gate.nx 11080 B, gate 22/22 GREEN. x = mu + L z, where L is GR47 la_cholesky_q30 and is NOT recomputed here -- the rung own words, one ruler never a second decomposition. CLAUSE 1 PROVEN AGAINST THE DEFECT IT REPLACES, which is the only way to prove it: over a population of 2000 the empirical off-diagonal reads 1.974 against a target of 2.000 (1.3 percent) with axis variances 3.984 against 4 and 5.150 against 5, while THE SAME MACHINERY DRIVEN BY A DIAGONAL FACTOR decorrelates to 0.018 -- a 110x separation, so a covariance check that only read the diagonal would have passed the independent per-channel sampling this rung exists to replace. CLAUSE 2 PROVEN BY WIDENING: drawing 5 axes instead of 2 from the same seed leaves axis 0 and axis 1 BIT-IDENTICAL, because every axis draws from its own splitmix-derived stream, so appending a sixth master parameter can never reshuffle the existing back catalogue. CLAUSE 3 is bit-identical regeneration with a different-seed neg-control so reproducible does not degenerate into constant. CLAUSE 4 IS NOT CLAIMED: the referee panel showing population spread above the tier floor needs a panel whose only voting lens is currently self-refusing, and borrowing another rung evidence would be the whole defect this board exists to prevent. |
GR25 Placement solved, not authored · cell WATCHING · gate NO-GATE-DECLARED
| Contract | pl_fit_solve · executor Organ · 2 u · stands on GR20 |
|---|---|
| Definition of done | Their known-issue lists are our exceed axis, measured: VR HOT carries nipple clipping, bra clipping, body-hair-through-cloth and texture seams as PERSISTENT known issues across years, because garments are an authored library conformed onto a licensed topology; Infinigen has no garment system at all. We already generate a garment fitted BY MEASUREMENT (18 body cross-sections, ease margin) and a groom rooted by area-weighted scalp sampling -- so for us non-penetration is achievable BY CONSTRUCTION rather than by hand-fixing each asset. Done: garment and hair placement produced by a constraint solve over the measured body (clearance as a solved constraint, not an authored offset), such that GR20 reports ZERO garment-through-body and hair-through-body penetrations across a declared pose set for an arbitrary genome -- including bodies at the extremes of GR24's parameter space, which is precisely where an authored library fails |
| Matrix cell | ROADMAP: placement solved not authored -- garments and props fitted by the constraint solver · runtime/nx_part_solver_lib.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_part_solver_lib; declare nx_part_solver_lib TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-09-04 measure Live /world/rigview via Marionette (Waterfox 153.1, 1600x1000): __nxa reports clip=187 then clip=192 across two spaced reads, nv=14164 nt=28092 nj=104, fps 74-100. clip is NOT a defect count -- the source comment says __nxa.clip = contacts and the loop increments it ON THE PUSH-OUT, so it counts contacts RESOLVED and a WORKING collider RAISES it. Driving it to zero would delete the collision response and make clipping worse. 2026-09-04 measure ROOT CAUSE of garment-through-body on rigview, arithmetic not opinion: the vertex shader displaces the RENDERED surface by aS.x times (aP.x below zero ? uJ1 : uJ1R) plus aS.y times uJ2, applied AFTER dual-quaternion skinning, while the CPU collision cloud sets bsx from joint skinning ONLY and never sees j1o, j1oR or j2o. Constants measured from source: BRAD=.006*H shell, SRCH=.030*H search radius, soft-tissue clamp plus/minus .032*H. The breast travels 5.3x the shell and 1.07x the WHOLE search radius, so at peak excursion the cloth vertex riding the true surface is farther from the nearest UNDISPLACED sample than SRCH, hits the no-sample-in-range guard and registers NO CONTACT AT ALL. The collider goes blind exactly when the excursion is largest, which is when the clipping is most visible. 2026-09-04 land nx_nxa_rig_emit.nx 50326 to 51710 B on the NAS publishing tree, four CAS-verified edits, each size delta PREDICTED then ASSERTED: +948 soft-tissue term in the collision proxy, +8 missN declared, +10 missN incremented, +418 missN published. The injected term is DERIVED from the shader line rather than invented, so proxy and render cannot drift apart: a 200000-trial random-input equivalence check against the shader expression gives worst absolute deviation 7.105e-15 against a ~10-unit collision shell, with 197402 firing trials and a 2598-trial all-zero control proving the non-soft-vertex path is unchanged. Sovereign build GRANTed and GREEN (asm 292893 B, elf 65409 B). SOURCE-LANDED AND BUILD-PROVEN, NOT YET SERVED. 2026-09-04 lesson A COUNTER NAMED FOR WHAT THE READER ASSUMES RATHER THAN WHAT IT MEASURES SENDS THE READER AT THE WRONG REMEDY: __nxa.clip reads as clipping-defects and counts contacts-resolved, and acting on the name would have removed the collision response. Read the increment site before believing a counter name. SECOND, measured twice today: a REFUSED /api/build DESTROYS the _build fossil -- the 65409 B artifact from a GRANTed sovereign build was gone after nx_organ_ship's /api/build was refused on the I/O storm, so any staleness comparison taken after a refusal measures against an absence. THIRD: check-before-build stopped a duplicate rung -- GR25 already owns garment-through-body and hair-through-body penetration with GR20 as its ruler, so this is a CONTRIBUTION to GR25 and closes nothing on its own. 2026-09-04 queue OWED, blocked only on CAPABILITY not on knowledge: this seat holds jobrun, fswrite and stage_alias but NOT mgmt, so /api/promote is out of reach and nx_organ_ship routes its build through /api/build which admission refused. The build IS queued as BQ-nx_nxa_rig_emit with queue_launch_rc=0 and fires on the next nx_orchestrate pass with headroom. RECIPE TO FINISH, in order: (1) _offc/nx_sov_build_run.elf nx_nxa_rig_emit --build-only, which bypasses mgmt admission and was already proven GRANT-GREEN on this exact source; (2) nx_organ_ship nx_nxa_rig_emit once the I/O storm clears, so the content ruler, behaviour probe and gate all run before the promote; (3) re-read __nxa.clip and __nxa.miss on /world/rigview. miss is the NEW blind-set axis, a SEPARATE axis never folded into clip, and its DELTA across the fix is the reading -- the target is NOT zero, because a hem swinging clear of the body is legitimately a miss. 2026-09-04 measure THE ACCEPTANCE PLANE IS THE GOAL'S REAL WORKLIST AND IT ALREADY HOLDS THIS SESSION'S COMPLAINT VERBATIM, FILED 2026-08-30. knowledge/status/accept_queue.status reads awaiting_operator=1 rejected_awaiting_seat_rework=4 awaiting_seat_or_referee=0 verdict=GREEN, page live and 3465 s fresh. The five rows: beach-garment-objects REJECT (garments as mesh objects), beach-skin-relief UNJUDGED (micro-relief staged in TEXR, not yet on the served page, must be real texture not smooth painting), beach-hair-colour-wind REJECT (HAIR CLIPS THE BODY and moves like a mop, not strand by strand), beach-feet-planted REJECT, and beach-photorealism-overall REJECT whose operator words are the answer to surpass-MetaHuman: "is this truly getting to photorealism -- no: flat light, no shadows/SSS, doll faces painted not modelled, wig-shell hair, painted nipples, cone skirt, voxel sand; PS2-era look, far from MetaHuman/UE5". SO THE OPERATOR'S 2026-09-03 JUDGEMENT WAS NOT NEW INFORMATION -- IT WAS A STANDING REJECT ROW NOBODY HAD CLOSED, WITH SEVEN DEFECTS ALREADY ENUMERATED AND THEIR SYMBOLS ALREADY NAMED. Today's rigview fix closes the CLOTH half of the hair row's clipping (the collision proxy now carries the soft-tissue term); the HAIR half is untouched, and nx_hairdyn_lib hd_tick has NO COLLIDER AT ALL -- it passes only a maxd radial clamp. 2026-09-04 land THE HAIR HALF OF THE OPERATOR'S CLIPPING REJECT IS SOURCE-LANDED AS ONE SHARED COLLIDER, NOT A FOURTH COPY. nx_vecmath gains vm_capsule_push (17806 to 20258 B): point-versus-capsule with the surface point written out and 1 returned only when the point was inside, so a caller pushes N points against M capsules without a sentinel. IT LIVES THERE BECAUSE THE ARITHMETIC ALREADY EXISTED THREE TIMES -- inline in nx_softjiggle's sj_step at Q256 (the only one that actually pushes), and as ss_capsule in BOTH nx_skullsdf trees at Q1024 (distance only) -- while nx_vecmath, which owns the vm_isqrt all three call, had none. Q256 is ADOPTED from sj_step rather than re-derived so the migration is byte-comparable instead of a second opinion about the same geometry. nx_hairdyn_lib gains hd_tick_collide (13520 to 16382 B) taking a flat 7-i64-per-capsule array. THE DEFECT IT ANSWERS, STATED PRECISELY: hd_tick's ONLY spatial bound is maxd, a RADIAL CLAMP about the anchor -- a clamp limits how far a segment travels and says NOTHING about what it travels through, and the two are easy to confuse because both bound a position. hd_tick IS UNCHANGED AND STILL SHIPS: this is a SIBLING because the bake path and nx_hairdyn_gate's ringdown and phase-lag teeth assert hd_tick's exact output, so folding a collider in would have moved every baked frame and every asserted number -- a behaviour change wearing a bug fix's clothes. Its return is CONTACTS RESOLVED, not defects found, and the header says so with the __nxa.clip precedent named, because that exact counter was misread as a defect count this same day and driving it to zero would delete the collision response. BUILD-UNPROVEN: three paced build attempts refused by admission (run-queue saturation then procs_blocked=8 of 8, load 19.31); queued as BQ-nx_percept and re-issuable, and hammering a storming box is itself a defect by standing order. 2026-09-04 measure CORRECTING MY OWN ROW FROM MINUTES AGO: THE HAIR COLLIDER IS NO LONGER BUILD-UNPROVEN. The previous land row said three paced attempts had been refused by admission; the fourth, paced ~8 minutes later, returned VERDICT=GRANT (load1 1524 under a 1643 ceiling, procs_running=1, procs_blocked=5 of 8) and nx_hairdyn_gate COMPILED: 46824 B, sha 0196dacea4338838fb0e422744fcb8b094f9b23f039cec74e5d1a9796e888c4d. That gate's closure carries BOTH new pieces, so vm_capsule_push in nx_vecmath and hd_tick_collide in nx_hairdyn_lib are compile-proven together. STILL HONESTLY OPEN: the gate BINARY is built but not promoted, so its teeth have not been RUN against the new lib -- and its teeth measure hd_tick, which is deliberately unchanged, so a GREEN from it would say nothing about the collider anyway. The collider owes its OWN tooth: a strand seeded inside a capsule must end OUTSIDE it, with a neg-control proving the tooth fires when the push is removed. Until that exists this is a compiling capability, not a proven one, and the acceptance row it serves stays REJECT. THE SAME ADMISSION STORM IS WHY nx_percept's ladderx is still unshipped: four ship attempts, the last two never reaching their build stage, so the leave-one-family-out measurement that would confirm or refute the hue-confound hypothesis is still owed. 2026-09-04 measure THE COLLIDER'S OWN TOOTH EXISTS, RAN, AND CAUGHT A REAL DEFECT IN THE CODE I SHIPPED AN HOUR EARLIER. nx_haircollide_gate (7450 to 8413 B, built 37223 B, run-exit=1) reports passed 11/12 verdict=RED. THE EIGHT GEOMETRY TEETH ALL PASS AT EXACT VALUES: a point inside reports a contact and lands ON the surface (100 against r=100, tol 2), the push is purely radial so the axial coordinate is untouched (500 unchanged), a point OUTSIDE reports no contact AND leaves the out-buffer unwritten, the on-axis degenerate case takes the documented +x, and a point 500 beyond the endpoint is OUTSIDE the capsule rather than inside the infinite line (the commonest way this arithmetic is written wrong). BOTH CONTROLS PASS: ncap=0 resolves nothing, and without the collider at least one segment IS left inside, so the tooth can fail. THE FAILING TOOTH IS THE END-TO-END ONE: WITH-the-collider-NO-segment-is-left-inside actual=4 expected=0, identical to the no-collider control. THE DIAGNOSTIC VALUES NAME THE SHAPE: seg0 after the collided tick is x=0 y=978 dist_to_axis=0, against x=0 y=880 without it, with contacts_RESOLVED=4 and SD_STRIDE=6. So the collider IS changing the system (y moved) and IS firing (4 contacts), but the radial correction is NOT SURVIVING to the end of the tick -- a push of (0,880,0) against a y-axis capsule must land at (100,880,0), x moving and y not, and the measurement shows the opposite. THREE CAUSES REMAIN CONSISTENT WITH THAT AND THEY HAVE DIFFERENT REMEDIES: the pushed position being overwritten by the next segment's sd_step_axq, a slot-convention mismatch inside the 6-wide stride, or the spring re-clamping after the push. NOT GUESSED -- the next cycle prints the per-iteration positions and separates them. THE ACCEPTANCE ROW beach-hair-colour-wind STAYS REJECT, AND THAT IS THE GATE WORKING: a compiling collider that leaves every segment inside the body is exactly the state a green-by-default tooth would have shipped as done. 2026-09-04 land RED TO GREEN ON THE PREDICTED VALUES: nx_haircollide_gate 12/12 verdict=GREEN, run-exit=0, and the hair half of the operator's clipping REJECT is now a PROVEN capability rather than a compiling one. THE DEFECT WAS THE Q256 PARAMETER QUANTISATION IN vm_capsule_push, AND THE GATE'S OWN EMITTED VALUES SOLVED IT BY HAND BEFORE ANY GUESS: for seg0 at (0,880,0) against the axis (0,0,0)-(0,1000,0), t = 880*1000*256/1000000 = 225.28 TRUNCATED TO 225, giving a closest point of (0,878,0) instead of (0,880,0), an offset of (0,2,0) whose direction is +Y rather than radial, and a push to (0,978,0) -- which is the EXACT value the failing run measured. On an axis of length L a Q256 parameter resolves the closest point only to L/256, here 3.9 units, so ANY point nearer the axis than that error gets a direction dominated by axial truncation noise and is pushed ALONG the capsule instead of out of it. FIXED AT THE ROOT, NOT WITH A TOLERANCE: exact integer num/den with the numerator clamped to [0,den], one divide, no quantisation step at all, plus a named den==0 branch for the degenerate sphere case. AFTER: seg0 x 0 to 100, y 978 to 880 (axial coordinate now untouched, as a radial push requires), dist_to_axis 0 to 100 = exactly r, segments_left_inside 4 to 0, and the no-collider control still reports 4 so the tooth can still fail. ★★★★★★A QUANTISED PARAMETER DOES NOT DEGRADE GRACEFULLY WHEN THE QUANTITY IT MEASURES FALLS BELOW ITS STEP -- IT POINTS THE WRONG WAY WHILE STILL REPORTING SUCCESS. ★AND THE CONTACT COUNTER READ 4-OF-4 RESOLVED THROUGHOUT, i.e. the collider reported complete success on every segment while leaving every segment inside the body: the same counter-name trap as __nxa.clip, seen from the other side -- a SUCCESS count that is true and useless. sj_step never exposed this because its points are skin-surface against limb axes and never sit within a few units of one; hair hanging straight down the capsule is precisely the pathological case, which is why the incumbent's Q256 was safe to adopt and unsafe to keep. 2026-09-04 land GE21 PREPARATION CAUGHT A DEFECT IN MY OWN HOUR-OLD CODE, AND IT WAS INVISIBLE ON THE NATIVE TARGET. Checking whether nx_wasm_craft could simply import nx_hairdyn_lib (it imports nx_softbind but NOT the solver; its only hair symbols are rig_hair_y and rig_hair_r, i.e. GEOMETRY, so the chain integration genuinely lives only in the emitted JS and GE21/GE22 is real) surfaced this: hd_tick_collide called sys_mmap(24) PER INVOCATION. The emitted $nx_wasm_mmap is a STATELESS GROW-PER-ALLOCATION helper -- pages = ceil(n/64K), base = memory.grow(pages)*64K -- so on the wasm lane that is roughly 720 memory.grow calls per second at 60 Hz across a 12-girl cast, unbounded, on the exact path GE21 exists to move into wasm. IT WOULD NOT HAVE APPEARED IN THE NATIVE BUILD AT ALL. ★★★★★★NEVER ALLOCATE IN A HOT LOOP -- AND "HOT" IS A PROPERTY OF THE TARGET, NOT OF THE SOURCE: the same function is harmless natively and catastrophic in the browser, so a review that only ever reads the source cannot see it. FIXED by taking a caller-owned scratch buffer as a parameter rather than hiding an allocator inside, with the reasoning written beside it; the gate allocates it ONCE, because a gate that allocated per call would hide the very property the parameter exists to guarantee. NEUTRALITY PROVEN ON REAL DATA: 12/12 GREEN again with EVERY emitted value byte-identical to the pre-change run (seg0 100/880/0, dist 100, inside 0 with the collider and 4 without, contacts 4). ⚠DECLARED, NOT SILENT: this run also printed DEPLOYED to serving root -- because the gate now HAS a live twin from its promote, the default nx_sov_build_run refreshed it. That is the documented twin-refresh behaviour and the deployed artifact is the GREEN one matching source, but an unannounced deploy is exactly what the record warns about, so it is named here rather than left to a later census to discover. 2026-09-04 retract I BUILT A FIFTH COPY OF THE CAPSULE COLLIDER WHILE TELLING THE OPERATOR I WAS CONSOLIDATING THREE, AND THE ONE I MISSED IS THE ONE THAT SHIPS. nx_wasm_craft.nx carries hg_capd2 / hg_inside / hg_push -- a full capsule collider with HG_CAP_MAX=32 slots (head, neck, 2 shoulders, 2 arms, torso, DYNA anchors) in HG_CAP_W=10 format (ax ay az bx by bz r kind sx sy) -- and it references vm_capsule_push exactly ZERO times. WHY THE SEARCH MISSED IT: I grepped for `func .*capsule`, and NOT ONE of the shipping implementation's identifiers contains the word. That is this estate's own law firing on me -- A MARKER LIST KEYED ON THE VOCABULARY OF THE THING YOU ARE DETECTING CANNOT SEE INSTANCES EXPRESSED IN OTHER VOCABULARY, AND THE MISS READS AS "THIS DOES NOT EXIST" RATHER THAN AS "MY DEFINITION IS TOO NARROW". The check-before-build rule was followed in form and failed in substance. ★AND IT REFRAMES THE RUNG, WHICH MATTERS MORE THAN THE DUPLICATE: the operator's REJECT says "hair clips the body", and I read that as A COLLIDER IS MISSING. IT IS NOT. A collider is present, wired, and shipping to the browser; the defect is that IT IS NOT GOOD ENOUGH. Those are different rungs with different work. WHAT THE COMPARISON IS WORTH KEEPING FOR: hg_capd2 quantises its segment parameter exactly as my first cut did, at HG_Q12=4096 rather than Q256 -- the SAME defect class, 16x finer, so the near-axis direction error is L/4096 instead of L/256. And hg_push's degenerate branch (d < 1) pushes along +z, THE STATURE AXIS: correct for the arm capsules, which are roughly horizontal, but PARALLEL TO THE AXIS for the head, neck and torso capsules, which are the ones hair actually lies against -- so an exactly-on-axis point there is pushed ALONG the collider and stays inside, the same failure nx_haircollide_gate caught in mine. ⚠NOT CLAIMED AS THE OPERATOR'S DEFECT: d<1 needs a guide almost exactly on a collider axis and hair roots sit on the scalp SURFACE, so the branch is latent rather than shown to be reached. Measuring it is the next step, not asserting it. ⇒ THE REMEDY IS NOW CLEAR AND SMALLER THAN GE21: have hg_capd2/hg_push COMPOSE vm_capsule_push so there is one ruler, which also puts nx_haircollide_gate's twelve teeth over the SHIPPING path instead of only over the native solver. 2026-09-04 measure BOTH OF MY CONCERNS ABOUT THE SHIPPING COLLIDER ARE REFUTED BY ONE CONSTANT, AND THE REAL MECHANISM IS SOMEWHERE ELSE. HG_UNITS_PER_178CM=171530 means ONE UNIT IS ABOUT 0.01 mm. So (a) hg_capd2's HG_Q12=4096 segment-parameter quantisation costs L/4096 = 42 units = 0.44 mm on a stature-length capsule -- against the 670 units = 7 mm my own Q256 cut cost, which is why mine broke and theirs does not; Q12 was chosen fine enough ON PURPOSE and I was about to "fix" working code. And (b) hg_push's +z degenerate branch needs a guide within d<1 = 0.01 mm of a collider axis, which is never reached by scalp-rooted hair. THE ENGINE'S COLLIDER IS SOUND ON BOTH AXES I QUESTIONED. ★SO WHERE THE CLIPPING ACTUALLY COMES FROM, READ OFF THE RIBBON BUILDER: collision runs on the 128 SIMULATED GUIDES, but the page renders 1200 strands as SIX INTERPOLATED CHILDREN PER GUIDE, and each child is displaced laterally by a hash offset -- ox9 = hsh9(gidx9,k5)*w09*1.2 and oz9 = hsh9(gidx9+7,k5)*w09*1.2 -- APPLIED AFTER THE GUIDE WAS COLLIDED, WITH NO RE-CHECK. With w09 = max(900, span*0.045) that is roughly 900 to 1600 units before the 1.2 factor, i.e. about 9 to 17 mm of UN-COLLIDED lateral displacement on every rendered strand. That is far more than any plausible body clearance and it is enough on its own to push hair through the body. ★★★★★★A COLLISION PASS THAT RUNS ON THE SIMULATED SET AND NOT ON THE RENDERED SET IS NOT A COLLIDER FOR ANYTHING THE OPERATOR CAN SEE -- AND IT LOOKS PERFECT IN EVERY TEST THAT MEASURES THE SIMULATED SET, WHICH IS EVERY TEST ANYONE WOULD NATURALLY WRITE. My own nx_haircollide_gate has exactly this blind spot: it measures chain segments, so it would stay 12/12 GREEN with the render path clipping visibly. ⇒ THE REMEDY IS ORDERING, NOT ARITHMETIC: apply the child offset BEFORE the collision pass, or re-collide each child after offsetting, or bound the offset by the guide's measured clearance. Cheapest correct form is to collide the child, since the offset is deterministic from the hash and needs no new state. NOT YET IMPLEMENTED -- named with its evidence so the next cycle starts from the mechanism instead of from the symptom. 2026-09-04 land THE CHILD-OFFSET FIX IS SOURCE-LANDED, BUILT, ADJUDICATED AND PROMOTED; THE PAGE RE-EMIT IS THE ONLY STEP LEFT. nx_game_page_emit.nx 358904 to 360131 B: the child offset is now clamped so it cannot move a strand INWARD -- rq9/nrx9/nrz9 from the strand root's own xz give the radial direction about the head axis, and only the negative radial component is removed, so all lateral variety survives while a child can never end nearer the body than its collided guide. Built 430126 B; shipped and PROMOTED, live 425241 to 451160 B. THE ADOPTGATE REFUSED FIRST AND EVERY ONE OF ITS FOUR LOST RUNS WAS ADJUDICATED BEFORE ANY FLAG WAS PASSED, because three of them were NOT MINE and that is exactly the AHEAD-class shape where promoting DESTROYS capability: (1) the const-ox9 line, mine, const to let; (2) the __nx_bootfail stall banner, mine, the webgpu branch added earlier today; (3) `then(async r=>{const ex=` -- the SOURCE reads `let ex`, because GE30 threading reassigns ex to the threaded instance's exports, so this is a SIBLING SEAT'S reword; (4) the _skipdraw render/blit line, same threading work. ALL FOUR ARE REWORDS WITH SOURCE AHEAD OF LIVE, i.e. the live emitter was simply STALE -- it predated the GE30 threading, my stall-banner edit, the sky extraction and this clamp. So this is BEHIND, not AHEAD, and promoting UPDATES rather than destroys. Only then was allow_loss=4 passed, with all four named. ★THE HABIT THAT MATTERS: THREE OF THE FOUR LOST RUNS WERE NOT MINE, AND A BLANKET allow_loss WOULD HAVE DELETED A BOOT-FAILURE HANDLER, A STALL BANNER AND TWO THREADING LINES WITHOUT ANYONE NOTICING. Enumerate before allowing, every time. OWED: nx_craft_emit re-emits the six world pages and was launched; its stage-1 native gate passed 60+ teeth and the 3b wasm-vm gate passed FITS and PAINTS, then output stopped advancing at the PAINTS-REAL precondition under host load. beach.html is still 55562 s old, so THE FIX IS NOT YET ON THE SERVED PAGE and beach-hair-colour-wind stays REJECT until a re-emit lands and a frame shows otherwise. |
GR26 Physically-based sky and aerial perspective · cell LANDED · gate UNRECORDED
| Contract | gpe_sky_atmos · executor Organ · 1 u · stands on nothing |
|---|---|
| Definition of done | The sky is a two-stop vertical colour ramp today (recipe spec rows 7 skytop and 8 skyhor) with the EH3 hash cloud sheets over it -- a gradient, not an atmosphere, and on the live beach it is roughly half the frame at default pitch. Single-scattering Rayleigh plus Mie with a sun disc, plus aerial perspective tinting distant geometry, in the same integer determinism as every other render rung. Ordered BEFORE GR27 because ground albedo is lit by sky irradiance, so a flat ramp caps how real any sand can look no matter how good its material is. Operator 2026-08-28 named it first of seven. |
| Matrix cell | WORLD-ROADMAP (GAME TIER): physically-based sky and aerial perspective · runtime/nx_game_page_emit.nx · LANDED decl |
| Gate | nx_game_page_emit_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | 2026-09-04 land gpe_sky_atmos NOW EXISTS AS A SYMBOL IN runtime/nx_game_page_emit.nx, AND THE INTERESTING PART IS THAT THE CAPABILITY WAS ALREADY SHIPPING. This rung is ranked 2 on this board at value 22000 and the operator named it FIRST of seven, and the matrix note still says the sky is a two-stop vertical colour ramp today. IT IS NOT. The emitted shader already carries derived Rayleigh with wavelength-dependent tau (440/lambda)^4 giving (.123,.287,.700), Kasten-Young airmass clamped at 38, single-scatter 1-exp(-tau*am), a SUN DISC at d>.99996 with a mix skirt, the SUNT transmittance dusk tint, and raymarched volumetric clouds with Beer-Lambert extinction and a dual-lobe HG phase; AERIAL PERSPECTIVE is in shade() using the SAME constants, mix(col,sky(rd),fw9). The shader's own comment says the two-stop painted gradient is gone. WHAT WAS GENUINELY ABSENT WAS THE NAME: gpe_sky_atmos occurred ZERO times, because the emission sat inline inside gpe_build_s, a thousand-line function. So the WATCH CELL was correct and the MATRIX NOTE was stale, which is a distinction worth keeping -- the instrument was right and the prose around it had rotted. A CAPABILITY THAT SHIPS WITHOUT ITS CONTRACT NAME IS INVISIBLE TO EVERY INSTRUMENT WATCHING FOR THE NAME, AND IT ROTS IN THE PESSIMISTIC DIRECTION: THE BOARD UNDERSTATES WHAT IS BUILT, WHICH IS THE HALF NOBODY AUDITS. FIXED BY NAMING THE REAL CODE, NOT BY WRITING A STUB: gpe_sky_atmos owns the sky emission point and every sky byte the page emits passes through it. Also removed a genuine duplicate import of nx_atmosphere.nx on consecutive lines. PROVEN NEUTRAL BY THE ESTATE'S OWN PROMOTE RULER rather than asserted: nx_contentdiff between the pre-change and post-change binaries reads runs_scanned=3592 lost_from_live=0 gained_by_candidate=0 symmetry=IDENTICAL-CONTENT, 3/3 GREEN. Both trees byte-identical at 358580. SCOPE DECLARED: the shader SOURCE literal still sits at the call site rather than inside the function. Hoisting it was built and proven content-neutral locally under the same ruler, and deliberately NOT shipped, because it means retyping four kilobytes of shader across a transport that has 503'd repeatedly today and one wrong byte is a corrupted sky nobody would notice until a human looked at the page. |
GR27 Ground material: granular sand and wet-shore darkening · cell WATCHING · gate NO-GATE-DECLARED
| Contract | wc_mat_ground · executor Organ · 1 u · stands on GR26 |
|---|---|
| Definition of done | EH1 fixes terrain SHAPE and EH2 fixes what GROWS; neither touches what the ground IS, so the shore reads as tan voxel paint. Granular high-frequency normal that survives grazing angles, darkening plus specular gain inside the wet band, slope-keyed dune banding. The one input it needs already exists: recipe spec row 66 carries the ABSOLUTE shoreline coordinate, so height-above-waterline is readable today with no new plane. Depends on GR26 for the irradiance that makes wet-vs-dry separate at all. |
| Matrix cell | WORLD-ROADMAP (GAME TIER): ground MATERIAL -- granular sand and wet-shore darkening · runtime/nx_wasm_craft.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_wasm_craft; declare nx_wasm_craft TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-09-06 land SOURCE: nx_world_shader_src.nx 77,418 -> 94,324 B (7 CAS edits, size ladder exact): fn sand(hp,c,rd,t)->vec2f in the ONE shader source -- the field normal tilted by hash grain facets and wind ripples, both Nyquist-faded against the pixel footprint stretched by the grazing angle; the wet band from an 8-direction x 3-radius water probe of the voxel volume at the column top and one below, submerged when water sits above; wet albedo x0.55 with a Blinn film lobe and a Fresnel sky sheen; dry quartz glints. shade() takes it on the field for classes 1, 2 and 5 (nx_wasm_craft gencol: 5 is sand at and below the waterline, SHORE repaints slot 1 as shore sand), suppresses the per-cell hash relief there, and the film loses the sun inside a block shadow. Build of nx_wgsl in flight; gate, promote and the beach re-emit owed. Every constant is a named data hook: the ripple heading is the wind row, the grain scale the recipe's sand grade 2026-09-06 lesson The shoreline coordinate the rung note points at (recipe spec row 66) is NOT needed: the volume already carries water as class 4, and probing the column's neighbours derives wetness from the actual water, which also answers lakes at other heights where a single waterline uniform would not. The per-class ground material word is the real data hook still owed; today the sand classes are keyed by id under the field 2026-09-06 measure nx_wgsl rebuilt on the NAS (second admission try, first refused QUEUE at load 18.09): 1,201,520 -> 1,223,496 B sha 0bc81fa0; nx_contentdiff lost_from_live=0 gained=19 ONE-WAY-GAIN GREEN 3/3; nx_behaveprobe world live 24,891 B vs staged 29,136 B both rc=0 DIFFERS STRONG (the sand function in both dialects); /api/promote PROMOTED with expect_sha256, prev banked; nx_offc_install refreshed the runner twin; nx_wgsl_gate 288/288 GREEN on the promoted binary 2026-09-06 land SERVED: nx_craft_emit re-emitted craft, beach, vale and isle (native gate 95/95, feel oracle 17/17, wasm-vm 19/19, page JS lexes and scopes resolve); the world fragment captured from nx_wgsl world grew to fragment_bytes=14508 and sites/nishifamily/world/beach.html 714,422 -> 716,600 B now carries fn sand(hp:vec3f,c:vec3i,rd:vec3f,t:f32)->vec2f in its WGSL. Shipped, referee grade OWED (no browser capture was launched: the no-GUI-without-asking law), operator ACCEPT owed -- the cell stays incomplete until the three signatures exist |
GR28 Palm flora reaches the SERVED beach · cell WATCHING · gate NO-GATE-DECLARED
| Contract | wc_veg_mesh · executor Organ · 1 u · stands on GR31 |
|---|---|
| Definition of done | RE-SCOPED 2026-09-01 (board owner). THE GROWER LANDED 2026-08-30 and this rung has been ranked #2 for it ever since: func wsp_palm_shape( is declared and CALLED at sh9 == 3 in nx_wasm_craft.nx, so the 2026-08-28 claim that only shapes 1 and 2 have growers is stale. The old contract name wc_tree_palm can never ship because the capability shipped under a different name on gameengine -- repointed, not left to rot. WHAT IS OPEN is the half this rung always also named ("one grower plus ONE RECIPE ROW") and it is deliberately reverted, not forgotten: knowledge/world_recipes.conf carries the beach row with NO spec pairs on purpose, because both densities were judged on REAL-BROWSER frames -- 5:64 an opaque green canopy roof, 5:400 a forest of solid green cubes at eye level, because frond rosettes render as filled voxel crowns. DONE-RULE, taken verbatim from the conf's own revert record rather than invented here: palms return to the beach only when the crown renders as fronds (or the world leaves voxels) -- so this rung now DEPENDS on GR31, and its proof is a captured frame of the SERVED beach graded by the world referee (GR30), never the presence of a grower. CONTRACT NAME ALIGNED to gameengine.matrix the same hour: lane 3 split the same capability onto _ABSENT_:wc_veg_mesh minutes before I named it wsp_beach_flora_bind, and shipping both would have manufactured a third two-names-one-capability defect on the very board being corrected for two of them. Their arithmetic also supersedes the conf's "retune the density" reading: 1 block = 1 m, a frond is a chain of at most four 1 m cubes, so a crown is ~33 cubic metres of solid green against a real rachis of ~0.05 m -- no sparsity value makes a blob stop being a blob, and the exit condition is sub-block crown geometry, i.e. GR31. Operator 2026-08-28: no infinigen generated sand or palm trees. |
| Matrix cell | WORLD-ROADMAP (GAME TIER): palm flora reaches the SERVED beach -- the grower landed, the recipe binding did not · runtime/nx_wasm_craft.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_wasm_craft; declare nx_wasm_craft TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR29 Skin micro-surface on the path the family looks through · cell WATCHING · gate UNRECORDED
| Contract | gpe_skin_microsurface · executor Organ · 1.5 u · stands on nothing |
|---|---|
| Definition of done | SCOPE SETTLED 2026-08-28 BY MEASUREMENT, after two wrong diagnoses of mine that the instruments corrected. NOT missing maps: TEXM is PRESENT, floor gate 7/7, 721,830 B, GL tier samples it. NOT a missing pore model: nx_relief_lib owns a cited three-band law and the baker already consumes it. THE DEFECT IS RESOLUTION, and nx_relief_bands_gate proves it -- a whole-body atlas spans 1.7m, so at 4096 the texel is 415um and secondary and pore both read carried=0; the live cast ships at 2048. Bands are computed then clamped, leaving a coarser band wearing the fine band's name, which is exactly what reads as painted. RES NEEDED: secondary 13600 whole-body, but over a 250mm face only 2000, and pore only 625. SO THE RUNG IS A PER-REGION ATLAS, not a bigger whole-body one -- the resolution we already ship carries the pore band once the footprint is a face rather than a body. Done rule: pore carried=1 at shipping resolution on a region atlas, with the neg-control that the whole-body atlas still reports 0 rather than being quietly redefined. Done rule: the served page shows measurable micro-relief at skin scale under the referee panel, with the neg-control that a flat-albedo build FAILS the same tooth. |
| Matrix cell | MATERIAL: skin MICRO-SURFACE on the live path -- pores and areola structure, generated not painted · runtime/nx_game_page_emit.nx · WATCHING decl |
| Gate | nx_game_page_emit_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | 2026-09-13 measure 2026-09-13 native regional skin asset generated and decoded: resident-region-skin2048-t278.nxa SHA256 1b0ece72f0793618a774ecb2f24dd425393ac89098312b397334abbe7f93dbbf, 41169952 bytes, four regional sets and20 PNGs. Actual native bake3502 exit0 elapsed70650ms VmHWM231244kB; initial3414 timed out at60000ms after three sets and was killed/reaped, no final output, source unchanged. Existing positive inspect3617 and native decoder3616 passed all20 dimensions/RGB byte lengths; all18 source sections byte-equal. Canonical codec and baker safety adopted; native CLI shipping under review. Renderer integration, physical calibration, chart seams and visible photorealism remain unverified. Evidence buildroot/knowledge/gates/region-skin-generated-review-t278.json and texr-renderer-handoff-t278.json. Isolated jobs own temporary allocations until exit; no in-process leak-free claim. 2026-09-13 land 2026-09-13 canonical native regional baker shipped via guarded workflow3965; exactly5 reviewed diagnostic changes allowed, fresh capacity gate10/10, serving and _offc SHA256 e19ad62435b32c75dc8fd40b3d0ce2df0fd15952842ce4adc6c4844e708c42c9 byte-identical. Actual existing registered inspect call3985 reads2048 candidate and20map descriptors. See buildroot/knowledge/gates/region-baker-canonical-adoption-t278.json. Generated map artifact and decoding are qualified; shared-renderer integration and visible photorealism remain open. No Beach page or production asset replaced. 2026-09-13 land 2026-09-13 Native regional skin chart-edge normal correction adopted and shipped via existing baker; canonical six-group gate passed, registered inspect passed, live/offc SHA db135be99a0a127731a08d27ce5874431ec0b2a33b6305721140e8a3f71a69db. Controlled 2048 bake preserves 16 non-normal maps and 18 sections; 16642856 nonedge pixels unchanged, 50253 boundary pixels corrected, zero measured new gradient-direction disagreements. Physical calibration, chart topology and Beach renderer integration remain open. Evidence buildroot/knowledge/gates/region-edge-adoption-t278.json 2026-09-13 land 2026-09-13 Existing native regional inspector now emits artifact SHA256 and explicit normal transfer metadata tied to focused shared encoding owner. Registered call6185 equals qualified4022B fixture; eight normal descriptors, prior fields preserved, producer verification remains a separate hash-joined receipt. Native complete-capture regression and16 boundary checks passed; ship6153 rebuilt stale shared-closure gate before promotion. No Beach shader/page release from this change. Evidence buildroot/knowledge/gates/texr-descriptor-delivery-t279.json 2026-09-13 watch 2026-09-13 Independent asset-cohort discovery retained: existing rigfloor roster has11 ref variants of one geometry; Catwalk glTF has no declared extensions/morphs. Additional available VRM-derived Seed-san, FBX-derived DarkKnight and static OBJ Bunny are regression candidates with exact hashes; no genuinely unseen holdout verified. Native section checks pass for two NXA assets, while provenance API returns NO-ROW and private_use allowed. Existing asset-catalog backing manifest not found at declared roots and4096 enumeration/index hazard remains; no catalogue mutation or rendering claims. Review buildroot/knowledge/gates/asset-cohort-existing-owner-review-t280.json 2026-09-13 watch 2026-09-13 Native all-corner UV preparation preserves all84276 source corners and28092 triangles;102-edge shortest source-graph cut leaves7 charts with disk-consistent Euler/boundary counts. Existing clipped-cotangent Jacobi diverges on main chart; new isolated factorized least-squares solver converges6 small charts in31-132iterations, main chart remains unqualified after1024iterations20.99s. No triangles hidden and no usable atlas or Beach texture release claimed. Development fixtures and familiar NPCs are not unseen holdouts. Evidence knowledge/gates/uv-solver-cut-review-t280.json; actual job1789328087; pending bounded followup8229. 2026-09-13 watch 2026-09-13 UV native qualification exposed compiler dot-store hardcoded integer destination despite resolved f64 field. Isolated existing-parser candidate passes13/13 native exact-value and invalid-field controls including indexed chains; fullreview knowledge/gates/parser-field-final-review-t280.json. Canonical/serving compiler unchanged pending review. Separate retained debt: address-of SSA let returns value as address; MCP64KiB request ceiling. Solver caller conversion misuse corrected using existing explicit numeric intrinsics, not attributed to compiler. No visual acceptance from source gates. 2026-09-13 watch 2026-09-13 Independent nonhumanoid Bunny qualification: exact source1eb35d1e21ce99e5ce911353b6be278990713448dd9e8f5c9387f9de39b32205 2408417B; native oi_read retains35947vertices/69451triangles but fx1024 creates2011 zero-area faces. Independent source/quantized buffers agree; source already has1113 loosevertices and223 boundaryedges, no winding/nonmanifold/duplicate defects. Existing meshcheck22/22 controls pass. Generated normals not allunitlength; occlusion/rendering unmeasured. Asset NOT accepted, not unseen. Exact native buffer identities, failure provenance and coordinate-contract options retained knowledge/gates/bunny-native-review-t282.json. No asset publication or shared parser change. 2026-09-13 land 2026-09-13 Shared normal encoding contract and explicit selected-producer qualification published to Alpha Beach. Native fullpage30/30, legacyCLI30/30, wrong reference SHA refused before page emission. Public https://nishifamily.com/alpha/beach exact1417792B SHA8e6fddbede735b7107dd2f622cdff2d79db1111eb106ce9bbadd6393a93048e7; Playwright WebGPU Intel xe-lpg cast pipeline created, clean scope, four legacy visitors, Frame character interaction reviewed. WebGL isolated samecode qualifies rendering. Diagnostic release note under Controls. Skin/hair/eyes/garments remain visibly inadequate; no TEXR integration or photoreal acceptance. Beta/Production unchanged. Evidence knowledge/gates/normal-alpha-release-t297.json. Correct structured release row replaces interpretation of prior free-text t296 append; historical text retained. 2026-09-13 measure 2026-09-13 UV preparation exposed resolved f64 field stores incorrectly checked/emitted as i64. Reviewed exact function adopted into canonical nx_parse.nx SHA8fb41956e2782826f26c5a445274fc7a2b4293f77f6da11d8d8167eaad2f6915; staged canonical x86 compiler976040B SHA5286c47d09a40271b143c356c66a8502af3e9147b19716de7e8e0261f2b5efd7 passes13/13 real native typed-value and direct/indexed invalid-field checks. Existing gate now reuses only byte-exact fixtures; source history not overwritten. Serving compiler unchanged pending wider toolchain qualification. No usable UV atlas or skin visual gain claimed. Evidence knowledge/gates/parser-canonical-adoption-t298.json. 2026-09-13 measure 2026-09-13 Native staged compiler field gate16/16: added u8/i32 neighbor preservation, signed negative load and pointer-chain checks; existing float64 and invalid-field controls retained. Evidence knowledge/gates/parser-field-narrow-t299.json job1789329333. Serving compiler unchanged; f32 and concurrent gate isolation unqualified. No Beach visual acceptance implied. 2026-09-13 measure 2026-09-13 Existing compiler field gate19/19, now f32 direct/indexed/pointer-chain values and neighbor preservation verified through staged canonical compiler. Evidence knowledge/gates/parser-field-narrow-t299.json retains both16/16 and19/19 iterations. Serving toolchain and public Beach unchanged; full conformance and concurrent harness isolation not claimed. 2026-09-13 watch 2026-09-13 Source-pinned material handoff audit: prepared TANGENT discarded; GPU tangent/normal image absent; imported lit shader uses geometric vN. Changing normal flag alone cannot repair this. Khronos tangent handedness requirements linked in knowledge/gates/material-handoff-audit-t301.json; shared owner integration and both-backend qualification pending. No new visible release. 2026-09-13 measure 2026-09-13 Structured binary comparison adopted into existing shared library: native16/16 and existing display18/18 pass. Correctly distinguishes DWARF count-byte false loss from genuine removals. Evidence knowledge/gates/contentdiff-shared-adoption-t302.json. Serving CLI/ship/staging migration pending; no allowance or new Beach release. 2026-09-13 watch 2026-09-13 Root review confirms candidate Bunny precision result0 introduced collapses vs2011 legacy;32 native controls are scoped evidence. Before adoption: reproduce/fix sign-only OBJ suffix acceptance and expose polygon triangulation qualification in result. Evidence knowledge/gates/obj-root-review-t303.json. No imported-model rendering or live repair claimed. 2026-09-13 measure 2026-09-13 Exact OBJ geometry API adopted in canonical nx_obj_import; actual native35/35 passes, Bunny0 introduced collapsed faces vs2011 legacy, exact source/index hashes retained. Polygon fan state explicit; units/materials/rendering unqualified. Evidence knowledge/gates/obj-canonical-adoption-t304.json. Ingestion-route and Alpha integration remain next; no public visual release. 2026-09-13 measure 2026-09-13 Native UV candidate conditioning counterexample9814 repaired by explicit Dirichlet elimination; same solver planar controls52/52 including cancellation/resume and existing LSCM24/24. Actual resident adaptive run0160 preserved all84276 source corners but remained rejected:10charts9392collapsed+8flipped atlas triangles, next nondisk split refused. Full15154B evidence knowledge/gates/uv-adaptive8-full-t280.out and knowledge/gates/uv-dirichlet-review-t280.json. No canonical unwrap or visible material adoption; caller cancellation is in-process only. Boundary-seeded partition candidate0337 running, not yet accepted. 2026-09-13 measure 2026-09-13 Management approval parsing canonical repair passes40/40 actual-handler/helper checks: exact unique yes required; duplicate approval/confirmation refuse. Shared form lookup reused, legacy lookup unchanged. Evidence knowledge/gates/mgmt-unique-approval-t305.json. Serving binary unchanged; structured-refusal propagation and guarded deployment still pending. 2026-09-13 watch 2026-09-13 Operator requests additional independent models beyond historic NPCs. Cross-model qualification must distinguish source defects from conversion, rig, material and rendered regressions; retain source identity, relevant capabilities and separate held-out assets. Reviewed material-normal-segment-review-t301.json: Catwalk has671 mixed tangent-handedness triangles and0 normal-map parts, so it cannot qualify mapped materials. Existing character worker assigned verified additional mapped asset plus controlled fixtures and no-map compatibility in shared integration; request remains pending, not delivered or published. 2026-09-13 measure 2026-09-13 Existing native UV preparation candidate now reuses only unchanged source-face/cut-corner sets; independent full5393952B packet parity passed after detecting and correcting double-quantization at atlas stage. Matrix applications10528 to4377, observed solve60.36s to37.79s; wall-time samples are not isolated benchmarks. Evidence knowledge/gates/uv-reuse-review-t280.json12614B. Resident UV still rejected6341collapsed+4flipped triangles; no material deployment. Timeout descendant gap retained, liveness0528 matched0 after valid402process snapshot. Caller residual/refinement policy now explicit; bounded32-refinement run0985 pending. 2026-09-13 measure 2026-09-13 Shared structured comparison iterator and complete-file CLI adopted in canonical owners via native MCP. Actual canonical build1081 compares material143320/147288B with state0 lost0 source_removed0 source_added1. Reviewed candidate17/17 and legacy18/18; canonical negative reruns pending. Evidence knowledge/gates/contentdiff-canonical-cli-t307.json. Serving binary and downstream staging/adoption guards unchanged; no Beach visual release or behavior equivalence claim. 2026-09-13 measure 2026-09-13 Canonical native comparison exposes typed result through existing workflow, avoiding console parsing for next adoption integration. Canonical gate1507 passes21/21 including actual material pair, tail loss, unknown/unreadable and stale-result invalidation. Evidence knowledge/gates/contentdiff-typed-workflow-t309.json. Serving and adoption gate not yet migrated; no Beach release. 2026-09-13 measure 2026-09-13 Canonical adoption content stage now calls typed native comparison directly, removing comparison subprocess and GREEN substring parsing. Native1720 passes6/6 actual material, self, tail loss, unknown, missing and forged text controls. Evidence knowledge/gates/adoption-native-content-t310.json. Behavior probe/capture and downstream staging integration remain; serving unchanged, no visual release. 2026-09-13 measure 2026-09-13 Native UV precision witnesses1756/1780 isolate source-bound chart0(7373faces) versus chart1(859faces). Exact dyadic independent raw determinants confirm no raw zeros/flips in either. Chart0 work export introduces8zeros; Q16 atlas introduces200morezeros+1flip. Control remains allpositive. Zero source-position/index mismatches. Full packet and per-triangle evidence knowledge/gates/uv-precision-review-t280.json18879B; chart0 solve3.66s, control0.22s, not hardware-normalized benchmark. Existing Q16/baker contract is a representation choice, not physical law. Adaptive32 run0985 timed out despite retained descendant output; final39chart atlas remains rejected. No further blind subdivision or live material adoption. 2026-09-13 measure 2026-09-13 Canonical staging now compares complete files, separates source loss and rejects UNQUALIFIED even with override; existing raw API preserved. Native2037 passes25/25; shared md_promote always classifies before rename, compatibility2077 passes40/40 (not real promotion). Evidence knowledge/gates/staging-canonical-adoption-t311.json. API/ship/behavior integration pending; serving unchanged. Root reviewed mapped diagnostic images; reversed labels require camera/source parity analysis, no global winding fix approved. 2026-09-13 measure 2026-09-13 Canonical management API deploy/promote now share qualified override predicate and expose content_state/source_removed/override_can_resolve. Native compatibility2271 passes40/40, not full HTTP promotion test. Evidence knowledge/gates/management-staging-propagation-t312.json. Serving unchanged; typed behavior and ship integration remain before release-chain qualification. No Beach visual release. 2026-09-13 measure 2026-09-13 Native symmetric-Dirichlet refinement candidate reuses source-indexed UvGramSolve after bounded inventory. Native2419 passes16/16 including derivative finite differences, cancellation and folded-input refusal. Actual chart0 128sweeps adds2.215s, raw/work faces allpositive and packeddefects209 to70; source-normalized minimum area improves2.19e-7 to0.00120. Rejected for whole-atlas use:64packedcollapses+6flips and worstanisotropy2.36 to606.6 despite lower meanenergy. Control859faces stayspositive allstages. Full source/metrics knowledge/gates/uv-stretch-review-t281.json. No canonical changes; no precision increase or further blind subdivision. 644012-10-12 measure Root t313: full management native build2708 passed without install; typed adoption observation candidate composes canonical content evidence and behavior candidate without parsing approval text, qualification pending. Actual WebGPU mapped/no-map images reviewed: relief visible, orientation unresolved, no NPC/page acceptance. Search503 read-after-accept recovered on same read-only query; default web scope unsuitable for internal-file discovery; root cause unresolved. Evidence knowledge/gates/release-review-followthrough-t313.json. 2026-09-13 measure 2026-09-13 Existing UV distortion-referee extension candidate now has native stage-specific max-anisotropy/orientation/collapse/unknown measurements and caller-owned acceptance. Actual source-bound gate3123 passes25/25: raw incumbent2.363 max versus t281606.622 is rejected under explicit incumbent-preservation policy; packed collapse still fails independently. Control859faces qualifies allstages under its measured caller policy. Nine native anisotropy measurements match independent finite-triangle Jacobian witness within8.41e-15 relative difference; collapse excluded only from undefined ratio, retained as failurecount. Full source/delta/evidence knowledge/gates/uv-quality-review-t282.json; canonical owner nx_uvunwrap_lib unchanged. No renderer/generation/serving changes; global overlap and whole-atlas acceptance remain unverified. 644012-10-12 land Canonical shared source adoption t315: runner35901B, probe13761B, integer4104B, adoption10562B, content inspector25941B. Native canonical behavior25/25 integer20/20 sectionlessELF10/10 content regression21/21 composed observation7/7. Typed observation always requires reviewed acceptance; legacy CLI/ship integration and serving deployment unfinished. No new Beach release. Evidence knowledge/gates/shared-adoption-t315.json. 1970-08-23 plan Weekly investment recommendation, not measured spend: prioritize visible character material/rig/contact and dune occlusion delivery; complete retained integration before expanding features. Shared MCP JSON+SSE repair must pass same attached file/admin calls. Swarm routing requires matched accepted outcomes including root context, delegation, review, rework, cached input, output and latency; current Claude review receipts have rootSupervisionTokens unknown, no positive ROI established. Evaluators require known-good/bad controls and independent visual witnesses: Sept19 anatomy numerical gains failed appearance and product path was retracted. Evidence knowledge/gates/native-mcp-sse-t316.json, shared-adoption-t315.json, claude-public-source-review-t253.json, claude-public-source-revision-review-t253.json; anatomy.plan and aesthetictwin.plan. Source/gate passes are not served visual delivery. 1970-08-23 research Current primary baseline check: MCP2026-07-28 publication https://blog.modelcontextprotocol.io/posts/2026-07-28/ specifies stateless core, header routing, cacheable discovery and task extensions. Existing JSON/SSE candidate repairs current endpoint compatibility; does not establish2026 conformance. Versioned client/server compatibility tests and migration assessment required before protocol adoption. DORA https://dora.dev/guides/dora-metrics/ defines five delivery metrics; combine with accepted Beach quality and full task effort, never use test count as product-quality proof. Current library/code/serving/visual closure gaps remain tracked; permanent repair requires shared-owner deployment and representative consumer verification. 1970-08-23 research Humble Unreal realtime simulation FX suite reference candidate: native fetch1789848115 HTTP200 retained840823B knowledge/fetched/rf_e70139e6.raw SHA8901e69b1f7aea387a9cdaec6d5dfb4aa5aa88fe49615fb8a14931dd1a756ffc. Browser observed USD22/100packs; Beach candidates Nature Water, Atmospheric FX, Leaves, Kick Up, Wind Gust. Evaluate three relevant effects for measured appearance/runtime comparison before conversion investment; no purchase, asset qualification, solver-source access, direct Nishi compatibility or positive ROI established. NPC motion/material/contact delivery remains primary. License and format checks pending actual redemption. Assessment knowledge/gates/humble-fx-reference-20260919.json; https://www.humblebundle.com/software/unreal-engine-realtime-simulation-fx-suite-software 1970-08-23 plan Served Alpha motion release b1789848521, page e673d1ea260dc1856dae419802887fd08276e4023667584e5ec15ff06af3af9c, published through existing site-release/v2 with five hash-bound dependencies. Root Chrome observed full source clip advance, pause holding time/revision, and discrete pose0 to1 collision revisions154 to155. Evidence knowledge/gates/beach-native-motion-release-20260919.json and beach-native-motion-served-review-20260919.json. This is mechanical playback qualification only; manually loaded prepared reference is not an autonomous integrated NPC or photorealism gain. 1970-08-23 plan Photoreal character acceptance remains FAILED: operator reports ankle/foot deformation, witch head/eye separation, inactive arms and static soldier; root resident Character1 face view confirms protruding eyes, hair intersection and inadequate skin. Audit complete resident and imported cast individually; one passing source-reference clip cannot substitute. Investment order: procedural generator source/bind/unit and attachment coherence, parameter-derived geometry/material/deformation, then representative multiview and integrated scene qualification. Imported fixtures serve oracles; target remains reproducible procedural digital humans, not per-model hand tuning. Automatic detection and full cast motion coverage remain incomplete. 1970-08-23 research Native discovery verified existing nx_body_proc (76228B) and nx_body_gen (77675B), plus nx_skelgen registered in authoritative serving tool_allowlist.conf169958B. Reuse and qualify these owners before new generator development. Current body recipe emits parameterized ring/feature geometry; nested fixed-radius bone/muscle shells do not establish anatomical digital-twin accuracy. Current resident FBX-derived rig path remains separate pending lineage verification. nx_procgen is a maze generator and nx_procgen_human_gate is a2D silhouette test, neither qualifies human creation. ProcFunc2026v2 revised2026-09-15 is current procedural-composition research, native fetched41712B knowledge/fetched/rf_8fcd074e.raw SHAa97b9d552e4bf41861d9d27e62c90b119b35e5ee5739aef22b8a20e378d89bef; source https://arxiv.org/abs/2604.26943v2 . Paper demonstrates room/material composition, not validated human generation. Next acceptance is repeatable generated human variants through shared rig/material/attachment pipeline and visible Alpha, not status-only metrics. 1970-08-23 plan Executed existing native procedural pipeline via MCP: nx_skelgen selftest1789849988 passed21/21; rest generation1789850009 wrote568B, 21joints requested1750mm actualvertex1751mm; nx_body_proc1789850015 wrote3563B canon; nx_body_gen1789850023 emitted103824tris/20parts truncated0 to knowledge/gates/procedural-human-20260919.nxmesh. Repeated jobs1789850055/1789850062 samecounts; byte determinism not yet verified. Mesh has no skinning contract established; Alpha inspection/conversion ongoing. Root visually reviewed skullskin-current-20260919_pair.png from existing skull input: severe face/skull geometry failure, not photoreal. Current fresh-skull generation must distinguish stale input from current recipe before repair. Reusable resident rig observation candidate passed21/21 isolatedChrome fixtures; actualserved phase capture pending. 1970-08-23 plan Historical graphics introduction superseded after served visual review; its old scores were not current acceptance evidence. Retained original: Nishi vs Unreal Engine 5 and Godot and Infinigen and Houdini -- the whole graphics estate on one measured surface: sovereign renderer, procedural bodies and creatures, layered anatomy, motion, gaussian splats, five live wasm worlds, referees and the research library. Measured 2026-08-17: capability breadth 827 permil (nx_graphics_board, 27 grounded axes) against a perceptual headline of about 175 permil clay (headline = MIN of statistical and perceptual, twice Goodhart-proofed); figure-crop CHARJUDGE ladder 0 to 257 vs corpus mean 138 over 169 renders; bodybench honest headline 369-389 detail-bound. The distance between breadth and the perceptual number is the whole roadmap, and the leap rung is the genome-driven splat avatar. 1970-08-23 plan Published and root-reviewed Alpha b1789851741 SHA49d75368588da4db775f5bf6c9fc3262c54b4a860476342c1585a66d863eed3b: Progress shows actual native procedural body images and static GLB; geometry remains coarse, unrigged and visually rejected. Native repeat mesh comparison proved byte determinism SHA3b58ea74d21012da6146e787243bb4510841729bec1b0d4638f4d9bb3d90d0bb. Real constrained skull fit job1789851408 completed:5525 samples,72live parameters, objective984 to721centi-mm, onepass; not anatomical or photoreal acceptance. Verified skull emission1789851941 pending. Same-frame resident transitional capture retained in knowledge/gates/resident-rig-moving-a-20260919.json; shared hierarchy correction under qualification, no released gait claim. References and failures retained in existing review records; no Beta or Production promotion. 1970-08-23 plan Operator clarification: the delivery unit is a coherent playable scene, including photoreal, anime and intentionally mixed multiverse directions. Alpha component workbenches feed integrated Beta scene review and accepted Production; component tests do not establish scene acceptance. Assess shared lighting/color/style, motion/contact, environment, audio/interactions and frame-time stability together. Investment denominator is accepted scene improvement; numerator includes all model input/cache/output, delegation, review, retries, rework and elapsed time, with attribution uncertainty retained. The same evidence-driven production pattern applies to physical and digital products, including material/resource efficiency; no current positive token ROI claim. 1970-08-23 plan Operator correction: top-down intended scene outcomes and bottom-up primitive/model verification must meet in one traceable evidence graph. Proposed native contract: claim, intended use/style, observable, units/frame, tolerance rationale, reference/calibration uncertainty, asset/code/backend version, scenario coverage, evaluator version, counterevidence and qualified scope. Validate evaluators using seeded failures, independent reference measurements, holdout scenarios and false-pass/false-fail rates. Passing component tests cannot substitute for missing system evidence. Known failures remain visible in Alpha and Beta; they do not automatically block experimental publication. Production regression invokes scoped incident containment/rollback then repair through Alpha/Beta. This is architecture direction, not an implemented autonomous validation system. 1970-08-23 gap Published generated body remains visually failed: coarse white static silhouette, no readable facial/skin details and no rig. Narrow enabled/disabled centerline proxy test passed, but user reports walk-through. Native movement owner uses zero horizontal player footprint; legacy cast lacks prepared collision coverage. General solid-character claim unqualified. Investigate exact asset/default/backend/overlap and carry user counterevidence forward. Researcher search for NASA STD7009 returned unrelated top results despite available primary standard; relevance and internal-vs-web scope need explicit evaluation. 1970-08-23 plan Nishi studio extension contract: standards must become executable reusable capability contracts, not claims of superiority. Each primitive owns versioned inputs/outputs, units/frame, failure semantics and evidence. Existing release workflow composes preparation, changed-dependency qualification, visible experimental Alpha publication, integrated Beta evaluation, production acceptance and scoped incident recovery. Agent consumes retained evidence and requests bounded tutoring only for unknown or repeated failures. Known visual defects may publish in Alpha/Beta with explicit failed/unreviewed status; production runtime failures trigger containment, repair and regression replay rather than ordinary continuation. No claim this autonomous cycle is implemented yet. 1970-08-23 plan Delivery efficiency must measure operator actions separately from internal operations, root-agent handoffs, elapsed time, complete provider input/output/cache token usage, retries/rework, accepted scene improvements and escaped defects. Compare matched task scope and quality; unavailable measurements stay unknown. One action is useful only if it produces a reviewable reproducible outcome and retains failure evidence. Existing estate owners must be composed rather than adding duplicate services or pages. 1970-08-23 gap Workflow ownership verified from current source and native catalogue: nx_workflow is read-only board/summary/show; nx_plan_run is legacy sequence runner with fixed step/argument/output caps and differs built175712B versus serving152683B; nx_wflow is the durable CLI, built/serving117440B, backed by nx_wflow_engine. Actual reusable release definitions are knowledge/wflow/component-release-preparation.flows and .steps at serving root, not buildroot. Current two steps only set release_scope and run site_prepare prepare-release with manifest/hash; no publication step. Extend this existing owner for explicit experimental Alpha publish and uncertain-effect reconciliation after current camera served review; do not add another executor. Catalogue INVOKED NO is bounded MCP-log evidence, not proof of never used. 1970-08-23 qualification Native complete-person composition retains52parts using intact body/head arenas rather than dropping head features at48. Native oriented mesh job1789857279 emits44994vertices/90028triangles, GLB2161120B; source positions unchanged by Yquad orientation correction, winding disagreements25276 to0. Topology remainsRED:212degenerate triangles22nonmanifold edges; rig absent. Root source review identified INT64_MIN overflow through abs on recipe coordinates, requested direct signed-bound validation before arithmetic. Whole-surface material500/torso is an explicit uncalibrated approximation; source part/material identity is not yet carried into output faces, so eye/hair material fidelity is not established. These are experimental Alpha candidates, not photoreal qualification. 1970-08-23 research-review Native rank8619 refused stale SB-HUMAN. Exact nx_barfresh grant repaired by union mint8663 preserving45 existing grants and expiry; original call8686 verified46 grants. Current stored feed knowledge/fetched/rf_ff76e849.raw updated2026-09-19 includes GenStream2609.18634 submitted2026-09-16, confirmed against live arXiv and author repository. This is semantic human-media streaming with client generative reconstruction and acknowledged compute/generalization limits, not evidence of superior procedural anatomical generation. Existing July photoreal baseline not relabelled simply to clear gate. Rank remains stale pending scoped baseline review. Reuse/repair/replace remains per-component evidence decision; no retirement authorized or performed. 1970-08-23 candidate-verification Frontier candidate t346 now propagates fs_attest_domain failures for single/all CLI paths instead of reporting ATTESTED after domain failure. Write1789860142 verified92636B. Actual native run1789860154 compiled but stopped earlier at missing knowledge/frontier_scan.conf under buildroot qualification cwd; domain-propagation behavior NOT yet exercised. Existing serving-root resource resolution must be reviewed, not silently copied as a workaround. Durable checked worklist publication and dispatch remain pending; no adoption. 1970-08-23 candidate-verification Existing nx_estate_path.ep_artifact_path now composed by frontier candidate for configuration and registry reads; no copied configuration or cwd mutation. Candidate92991B write0264. Actual CLI run1789860271 reaches requested absent domain, reports failed_domains=1 verdict=INCOMPLETE, exits2 instead of ATTESTED. Native helper regression pending; checked durable worklist write, remaining path consumers, review dispatch and canonical adoption remain incomplete. |
GR30 World-quality referee built and wired · cell LANDED · gate NO-GATE-DECLARED
| Contract | wqg_build_wired · executor Organ · 1.5 u · stands on GR1 |
|---|---|
| Definition of done | THE WORLD HALF OF GR1 AND IT WAS NEVER ON THIS PLAN. GR1 admitted a CHARACTER referee panel and the world lane still has none, so every world fidelity claim on this board is graded by the author's own eye. MEASURED 2026-09-01 by nx_catalog: nx_world_quality_grader is SOURCE yes 37940 B, BUILT NO, PROMOTED NO, REGISTERED NO, INVOKED NO, verdict SOURCE-ONLY (S2). The consumer already exists in prose -- nx_layer_verdict.nx carries an adapter for its 12-i64 output and names _wqg_grade_from_8, and _ft_diag.nx imports the organ -- so this rung is a BUILD and a WIRE, not a design. Done when the organ builds, promotes, registers, returns a grade plus 8 axis scores for a named world capture, and at least one world rung's done-rule reads that grade instead of a sentence. Carries the GR1 discipline: floors declared before the experiment, FP and FN measured on a labelled corpus, and the panel refuses rather than acquits when it cannot see |
| Matrix cell | REFEREE: world-quality grader BUILT and wired -- the world half of the referee story · runtime/nx_world_quality_grader.nx · LANDED decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_world_quality_grader; declare nx_world_quality_grader TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-10-01 queue CLAIMED by seat world-referee (second launch, T0 1790856829). State read from nx_catalog: nx_world_quality_grader built and promoted 32285 B and registered, its only live caller nx_worldgrade promoted 91681 B, its gate nx_world_quality_grader_gate built 43338 B and never promoted, contract symbol wqg_build_wired not declared in the organ. Plan: record real-world grades through nx_worldgrade, declare wqg_build_wired as the guarded entry that reports can-see, wire nx_worldgrade through it, extend the gate with a real-engine-world tooth, a planted stair-step control, determinism and emitted values, then promote and register the gate. The grader is a heightmap statistics ruler and not an image referee |
GR31 A second art style of emission beside the voxel raycast · cell WATCHING · gate NO-GATE-DECLARED
| Contract | wc_style_surface · executor Organ · 2 u · stands on nothing |
|---|---|
| Definition of done | OPERATOR 2026-08-30: everything is voxel and we dont need it to be that should just be one art style of emission. MEASURED 2026-09-01 on the served page: the beach WASM module exports 78 functions and NONE of them names a mesh, surface, rig, face, hair or skin producer -- the exports are simulation state and the world is drawn by a per-pixel integer DDA raycast over a voxel field, so voxel is the only style the engine can emit rather than the one it chose. The extractor is already built and graded elsewhere on the estate: nx_csg_extract, marching tetrahedra on the Kuhn six-tetrahedron split, watertight by construction, LANDED on both procgen and dcc. Done when the same worldgen field emits as a surface mesh through that incumbent extractor with the style selected by a recipe row and never by an engine fork, and with the voxel path proven byte-identical when the row is absent -- a second style, not a replacement |
| Matrix cell | WORLD: a second art style of emission -- surface extraction beside the voxel raycast, chosen as data · runtime/nx_wasm_craft.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_wasm_craft; declare nx_wasm_craft TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR32 The sovereign strand organ reaches a live world · cell WATCHING · gate NO-GATE-DECLARED
| Contract | wc_strand_bind · executor Organ · 1 u · stands on GR15 |
|---|---|
| Definition of done | GR15 buys a strand SYSTEM at 2 u as if from nothing. MEASURED 2026-09-01 by nx_catalog: nx_nxa_strands is SOURCE yes 9565 B, BUILT NO, PROMOTED NO, INVOKED NO, verdict SOURCE-ONLY (S2) -- 9565 B of strand source sits unbuilt beside a rung that does not know it exists, so the plan overstates its own cost and hides that the first move is a build. This rung is deliberately the BINDING and not the organ: building it alone would leave the page-side groom the operator is looking at untouched. Done when a strand set generated by the sovereign organ reaches a live world through an engine export and the page copies buffers rather than authoring geometry, which is the same bar gameengine GE21 wc_hair_tick and GE22 gpe_wasm_hair_bind already set for the guide chain |
| Matrix cell | HAIR: the sovereign strand organ reaches a live world -- incumbent exists and is unbuilt · runtime/nx_nxa_strands.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_nxa_strands; declare nx_nxa_strands TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-09-03 measure 2026-09-03 the strand delta the operator sees between /world/rigview and /world/beach is a RENDERER delta, not an asset delta: both pages read HSTR, and rigview loads the older ref9d while the beach loads ref9e. nx_nxa_rig_emit draws every strand as a sub-mm fibre with analytic coverage (its own comments name the trichology width, the length spread and the flyaway minority); the world emitter samples guides out of HSTR into 1200 opaque ribbon cards. The closure this board owns is GR15 plus this rung: nx_nxa_strands (still SOURCE-ONLY, 9565 B, never built) reaches a live world through an engine export and the fibre-coverage shader is emitted ONCE by nx_wgsl from NishiLang for both the viewer and the world -- extract rigview's coverage shading into that emitter, never copy its page JS onto the beach (gameengine GE21 GE22 GE38 ratchet page JS to zero). Until then ecosystem EC22 reads the beach UNBOUND on the groom capability the viewer proves. |
GR33 Strand chain ordered root-to-tip · cell WATCHING · gate NO-GATE-DECLARED
| Contract | hg_seg_axis · executor Organ · 0.5 u · stands on nothing |
|---|---|
| Definition of done | THE CHEAPEST VISIBLE FIX ON THIS BOARD AND IT IS ONE TOKEN. The served page derives a strand segment index from component 1 of the bind position while this mesh space is Z-UP, which the served shader itself declares (vec3 yup=vec3(sp.x,sp.z,-sp.y), beach.html:292) and which the fit pass ten lines above the bug already honours. The chain lag and its squared weight are therefore spread front-to-back across the head instead of root-to-tip down each strand, which is both the mop shear the operator sees and the mechanism that drives hair through the face. DONE-RULE, pointed at a gate and not at a declaration: nx_gpe_hair_geometry_gate carries hg_seg_axis, a tooth asserting the segment index is derived from the up axis the shader declares rather than from a hard-coded component, bite-proven RED against the current served expression and GREEN against the fix, and the gate is listed on graphics.gates so the verdict is executed evidence rather than prose |
| Matrix cell | HAIR: the strand chain is ordered ROOT-TO-TIP, proven by a tooth · runtime/nx_gpe_hair_geometry_gate.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_gpe_hair_geometry_gate; declare nx_gpe_hair_geometry_gate TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-09-08 land Beach hair pause-recovery hotfix h1788833756 is served with a visible release note. A ten-second pause previously left6.933seconds of backlog after60frames; the fixed timing discards excess paused time under the existing four-step budget. Tested normal outputs30/60/120FPS remain identical; pause fixtures1/10/60seconds resume without backlog. Actual native private emission4507 contains the tested function and both scripts parse. Composite nx_craft_emit755840B SHA77e47e930ac2d0b67fb0666ffddd4c3e1c698876806c6e6215a062a747b68964 promoted4585 and restaged4593; exact deployed-copy comparison4595. The one replaced printable timing statement was reviewed; contentdiff4421 remains RED for the intentional replacement. Full served-origin Beach4143 verified. Browser visual acceptance, hair appearance and body collision remain incomplete; GR33 overall completion is unchanged. Reference: https://gafferongames.com/post/fix_your_timestep/ for bounded timestep recovery, not a comparative performance result. 2026-09-08 land Beach groom-axis hotfix h1788835310 served with visible release worklog (HTTP5942 fullread; stage/live hash5927 identical). HSTR geometry was Z-up but ribbon width/jitter construction assumed Y-up; convert at builder boundaries. Actual ref9e asset SHA6e3078707944141164c50860939057173adf3787a48c6ef936e14ce2322319da has4800guides; actual emitted CPU builder preserves604800vertices and allUVs, finite/nondegenerate triangles and repeatability. Vertical width edges fall from100795quads to0 under the horizontal-width oracle; this is a coordinate correctness result, not visual quality or body-contact proof. Native source5310/build5374/private emit5418 verified; composite5493/promote5828/restage5871/hash5876 preserves regeneration. Complete diff5576 reviewed13lost printable runs (two statements and eleven comments), with5gained; RED remains for the intentional replacement. Browser/GPU/appearance/contact and external comparative performance remain unverified; GR33 remains incomplete. 2026-09-08 measure Actual Beach ref9e asset re-hashed6363:11728120B SHA6e3078707944141164c50860939057173adf3787a48c6ef936e14ce2322319da. Reproducible local contact-audit.mjs checks raw bind-pose surface intersections:11663of33600guide segments strictly cross body triangles; corrected emitted ribbons34153of100800quads have at least one strictly crossing edge (prior34508). Body has220boundary edges in6loops, so no unqualified solid inside/outside oracle. BVH agrees with brute force on131distributed guide samples; analytic hit/miss/winding/endpoint/coplanar fixtures pass. Counts omit touching, coplanar and exact-boundary cases and do not measure animated shader-deformed or rasterized clipping. This is a contact baseline, no acceptance or new visual release. Next fix must address guide/body fit and rendered children against posed body geometry. Established technique reference: AMD TressFX4 dynamic skinned-mesh SDF collision https://gpuopen.com/news/tressfx-4-simulation-changes/ (2017 primary source; not September2026 SOTA or a measured Nishi comparison). GR33 remains incomplete. 2026-09-08 land Beach h1788836868 initialized cheek vertex input served: publish1788838826 exit0 746442B; full HTTP1788838842 verified against scoped stage except existing ad insertion. Visible dated release worklog present. Composite adoption8732/restage8798/hash8843 exact. Float64 native scalar checks cover14164 body vertices, not GPU appearance or contact acceptance. User reports scene still looks unchanged: next milestone substantial hair/body-contact before-after, GR33 remains incomplete. Preview capability revoked8682 confirmed reason5. 2026-09-08 measure Beach release marker must be native in-game GPU geometry at starting camera with screenshot, not HTML icon. Native bush source and T96 under test, not delivered. User requires whole-system evaluation: CPU/GPU/RAM/VRAM/transfers/storage/network/frame pacing/thermal-power/cleanup; unavailable counters are coverage gaps, never passes. CPU-only Node simulation result does not evaluate renderer. Latest gate build refused admission; no queued work. Source verification and paced retry in progress. 2026-09-08 measure Native Beach release bush gate1788840419 passes96/96: fixture delta14, overwrites0, restamp unchanged, initial-camera ray hits foliage, native GPU palette verified. Native plain/sharedWASM and private page emitted. Exact-artifact VM suite1788840781 TERMINAL timeout120s after parsing/memory/render and initial terrain checks: partial only, NOT passed. No public release or GPU screenshot. Resolve fixture-failure init handling, shared palette scope, VM validation and browser verification before promotion. Full-system performance coverage remains outstanding. 2026-09-08 land Beach release beach-2026090702 served: publish1788842265 749670B; full HTTP1788842290 exact expected engine/page plus existing edge ad. Native world96/96; final plain/shared WASM19/19. User reports voxel teal bush visible in Waterfox. Automated screenshot and GPU/FPS improvement unobserved; no photoreal or SOTA claim. Worklog below game. Shared foliage recoloring disclosed. Delivery and rollback automation remain open; request-size workaround required17 exact artifact edits. 2026-09-08 requirements B26-HAIR/CHARACTER inside-out state: persistent seeded follicle layout, growth direction and length, shaving/stubble/regrowth, region-specific irritation and hair-removal history; shared skin and eye state continuity. Full qualification fixtures and approximation limits in knowledge/compare/sovereign-human-world-contract-20260907.md, subsection persistent follicle, skin and eye state. Pair close/mid/far visual review with reproducibility, save/resume, temporal stability and measured resource costs. Clinical treatment accuracy requires primary evidence; no complete digital twin or Beach implementation is claimed. Reuse GR29 skin and GR33 hair rather than add isolated pages. 2026-09-08 research Character research references checked 2026-09-08: Meta Codec Avatars https://www.meta.com/emerging-tech/codec-avatars/ provides telepresence datasets and reference implementations; inspect capture requirements, identity generalization and clothing/hair limits before comparison with procedural Beach NPCs. BioWare/Frostbite https://developer.microsoft.com/en-us/games/articles/2024/12/bioware-dragon-age-the-veilguard-strand-hair-technology/ (2024-12-05) reports up to 50000 strands per character on Xbox Series X, compute software rasterization, collision/pinching and extreme-closeup challenges; this is a production reference, not a WebGPU benchmark. EA SEED FAST https://media.contentapi.ea.com/content/dam/ea/seed/presentations/seed-id3-2024-spatio-temporal-sampling-paper.pdf (May2024 DOI10.1145/3651283) optimizes sample distributions for reconstruction filters, evaluated on AO, volumetric ray marching and dithering; qualify with matched filters, error and temporal stability at equal cost. UMBC https://news.cs.umbc.edu/2020/05/umbc-researchers-develop-better-techniques-to-render-characters-with-realistic-skin/ identifies Xie/Olano/Karis/Narkowicz2020 DOI10.1145/3384536, variance-guided adaptive subsurface sampling; underlying paper and current successors need review. Map skin to GR29 and rendering/sampling to existing graphics owners. These four leads are neither a comprehensive September2026 survey nor Nishi parity. Investigate Stanford, USC ICT, Berkeley, Cornell, CMU, NVIDIA, Epic, Disney, Adobe and Meta by specific dated method, reproducible artifact and relevant workload, not institutional prestige. Require source/availability/rights scope, hardware, quality-cost evidence, failure cases and one executable next comparison per adopted lead. |
GR34 The normal map already inside the asset is read · cell WATCHING · gate UNRECORDED
| Contract | gpe_texm_normal_bind · executor Organ · 0.5 u · stands on nothing |
|---|---|
| Definition of done | The served ref9e.nxa carries a four-map TEXM (albedo, spec, gloss, NORMAL; NTB_MAP_NORMAL=3) and the page reads record 0 at both TEXM call sites, so a normal map ships in the file the browser already downloads and nothing samples it. This is layer one of a three-layer consumer gap and is deliberately scoped to layer one. DONE-RULE: the page binds map record 3 as a normal map on the character pass; nx_nxa_texm_gate (re-run 2026-09-01, passed 12/12 verdict=GREEN) stays green so the carry is unbroken, and a new tooth shows the lit result differs from the record-0-only build on the same frame -- a bind that changes no pixel is not a bind |
| Matrix cell | MATERIAL: the normal map that is already inside the asset is actually read · runtime/nx_game_page_emit.nx · WATCHING decl |
| Gate | nx_game_page_emit_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | 2026-09-02 land gpe_texm_normal_bind SOURCE-LANDED in buildroot/runtime/nx_game_page_emit.nx (337,538 -> 339,9xx B, eight anchored CAS edits, every intermediate state a compiling NishiLang string): the body-pass fragment shader gains uniform sampler2D uNrm + int uNrmOn and perturbs n with the asset's TEXM normal map in a screen-space cotangent frame (dFdx/dFdy of world position and UV, no tangent attribute); the resident-cast decode reads the TEXM record whose map_id is 3 by scanning the record table (never assuming slot 3), decodes it once through nxDecodePng and binds it on unit 4 (unit 3 is the dress map uGD); GL.npc carries nrm; the per-frame bind sets uNrmOn=1 only when both the map and the uniform exist; the visiting-cast draw resets uNrmOn=0 so it never inherits the resident binding. uNrmOn=0 is byte-identical to the incumbent shading. Build is QUEUED (BQ-nx_craft_emit, admission-refused all day); the served page still runs the record-0-only build until it fires and the worlds re-emit. 2026-09-02 queue PROOF OWED before the cell may flip: (1) nx_nxa_texm_gate stays 12/12; (2) the new tooth -- a headless capture of /world/beach?review=1 with window.__npc_nmap===1 must DIFFER from the same frame with the map forced off, and the diff must sit inside the figure crop (a bind that changes no pixel is not a bind); (3) cjc_panel grade on the lit crop beside the record-0-only crop, because a normal map that reads worse is a regression on the operator's surface, not progress. 2026-09-02 lesson the ranker's own text was the spec: this rung cost 0.5 because the map was already inside the file the browser downloads. Layer two (per-region relief) and layer three (spec/gloss records 1 and 2 into the BRDF) stay open rungs; do not fold them in here. 2026-09-02 measure LIVE ON THE SERVED PAGE 2026-09-02 (b1788381661, read through the Gecko wire off the page's own GL context): window.__npc_nmap=1 -- the TEXM normal record was located by map_id, decoded once and bound on unit 4 for the body pass; nmapfail empty; tier WebGL2 + npc-atlas 2048; frame p50 75 to 84 ms, p95 136 to 138, p99 160 against a 133 ms budget at 2235x1117. The bind is PRESENT; the pixel-diff tooth (same frame with the map forced off must differ inside the figure crop) and the referee grade are still owed before the cell may read GREEN, and until then this rung reads shipped, unproven. |
GR35 A declared asset ladder, so finished work reaches the cast · cell WATCHING · gate UNRECORDED
| Contract | gpe_asset_ladder · executor Organ · 0.5 u · stands on GR34 |
|---|---|
| Definition of done | Roughly 80 MB of finished character assets are orphaned: ref9f.nxa (40448728 B, ref9e plus TEXR:3590072, every other section byte-identical) and ref9d_r2.nxa (39478840 B) are referenced by NOTHING in buildroot, measured with corpus_complete=1, while the page serves ref9e. The mechanism already exists (a ?npcasset= override and a ref9e/ref9d/ref9 fetch chain), so this is a declaration, not a loader. DONE-RULE: the ladder is a conf row naming the richest asset a build will serve, the emitter refuses by name when a named asset is absent rather than silently falling back, and nx_asset_floor_gate passes on the asset actually selected -- proven by the emitted page naming the richer asset and by a capture whose byte count differs from the ref9e build |
| Matrix cell | MATERIAL: the served cast loads the asset that carries the finished surface · runtime/nx_game_page_emit.nx · WATCHING decl |
| Gate | nx_game_page_emit_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | 2026-09-04 measure RIGVIEW AND BEACH SHOW DIFFERENT MODELS BECAUSE THEY LOAD DIFFERENT ASSETS, AND THE RICHEST ASSET IS SERVED BY NEITHER. OPERATOR 2026-09-04: the rigview still shows a totally different better in someways model where the beach has a better model in other ways. MEASURED, from the SERVED pages and from source, not inferred. RIGVIEW loads /world/ref9d.nxa -- read out of the live page through the browser wire. BEACH walks a hardcoded fallback chain in nx_game_page_emit.nx that prefers ref9e.nxa, with ref9d.nxa and ref9.nxa behind it, and nx_craft_emit.nx names ref9d.nxa separately. ★★★★★★SO THE TWO PAGES RENDER TWO DIFFERENT CHARACTER ASSETS AND NOTHING RECONCILES WHICH PAGE GETS WHICH: nx_nxa_rig_emit.nx names NO asset at all in its source -- rigview's NXA arrives as an emit-time URL argument -- while beach hardcodes its chain, so the selection lives in two incompatible places and neither can see the other. THE OPERATOR'S OBSERVATION IS THEREFORE NOT A RENDERING-QUALITY QUESTION AT ALL: each page is better in different ways because each is showing a DIFFERENT MODEL, and a comparison between them measures asset choice rather than renderer quality. ⚠AND GR35 ALREADY RECORDED THE WORST PART, UNBUILT: ref9f.nxa is ref9e PLUS TEXR:3590072 of micro-relief and is referenced by NOTHING, so the RICHEST finished asset is served by neither page while roughly 80 MB of finished character work sits orphaned. THREE TIERS EXIST, THE BEST ONE IS UNUSED, AND THE SELECTION IS A HARDCODED FALLBACK RATHER THAN A DECLARED LADDER -- which is exactly what GR35 gpe_asset_ladder was costed at 0.5 to fix and has been sitting unbuilt. ★SECOND MEASUREMENT FROM THE SAME READ, and it settles which of today's two fixes actually shipped: rigview's probe returns ok=1 nv=14164 nt=28092 nj=104 hair=2600 clip=202 MISS=83. miss is the blind-set counter added to nx_nxa_rig_emit TODAY -- 83 cloth vertices with NO body sample inside the search radius -- and it is LIVE on the served page. So the rigview soft-tissue proxy fix and its new instrument DID ship while the beach clamp did not, which is consistent rather than surprising: different emitters, different binaries, and only nx_craft_emit was stale by 16.6 hours. ⇒ORDERING CONSEQUENCE: comparing beach against rigview to judge renderer progress is measuring the wrong variable until the asset ladder lands, and any A B that changes the served asset without recording WHICH asset was served is uninterpretable. 2026-09-23 measure 2026-09-23 scoped Alpha inspection repair delivered at /alpha/beach title b1790154993: page1125385B SHA515e85baba3dd29f3d32f7cd64a8b4c0813a799b0733da06ecd4af97d881daa2, actual shared385136B SHA335df0bdd07f4d8b3f527438b5333a24a82627d827a9823bc1dff3a8bdf7045a. Current canonical native camera owner adopted guarded, memory layout preserved; native112/112, actual WASM53/53, pair40/40, pagefit32/32 and full JS referee20647refs0findings. Existing workflow1790155327 published with30 dependency identities verified,29 prior dependencies unchanged and prior1be4d1e retained. Root full-Chromium Intel xe-lpg served review09:27Z: Catwalk load, Fit full body, Body180degree orbit, resident face front/back, Return to gameplay and dated Progress record PASS; pageerrors0. Imported Face explicitly unqualified; refusing Face leaves orbit inert until Body reselected. Imported collision capability absent. Resident close-up eyes/hair/skin rejected; terrain/player/NPC contact, motion, performance and photoreal quality unaccepted. Beta/Production unchanged. Retained delivery buildroot/knowledge/gates/beach-camera-resume-20260923-delivery.json; served proof beach-camera-served-review-20260923.json SHA1984164d86220ef9232449506d305c74010571a040f0aeb0d1c566d8fb1c7b63;8 reviewed screenshots. Next existing owners: native imported-body integration and inspection refusal state; no new viewer. 2026-09-23 measure 2026-09-23 Alpha imported-contact repair published at /alpha/beach title b1790160590: page1216243B SHA79cef81700acf33f5c3daeaa3ce6c60d5b9f44f3aff7ecc7a73493066349d648; shared451054B SHAd986bf9b35d1b4e154f3c2b640383584299fb1d7ec30dd7442ad9e82e407be6e. Existing native body/registry/motion owners composed with scene-derived capacity, camera/saves/morph/layout retained; native140/140, actual WASM78/78, pair40/40, pagefit32/32 and native JS referee20728refs0findings passed. Guarded workflow1790160960 retained predecessor515e85ba and29 unchanged dependencies; all63 publication phases rc0. Root served Chromium Intel xe-lpg review: Catwalk proxy blocks actual player from both sides; disable allows passage and reenable blocks with samehandle1048577/revisions1,2,3. Imported Face remains unqualified but Body orbit stays active without moving player. Distinct id/same source binds2097153; actual native source playback updates proxy and stops at authored clip end. Observed player feet3409Q8 equals native terrain3409; render-source interpolation3409.92578125Q8; non-atomic sample, controller support unavailable, no anatomical/NPC or image-derived geometry acceptance. Reproduced defect: same-id replay refuses while residents are culled, succeeds when framed; existing two-owner readiness repair reviewed for private qualification. Skin/face/hair rejected, arbitrary through-dune occlusion/autonomous navigation/Native Browser parity unqualified. Negative unmatched QA timing: world+cast105.971712ms, raw-loop p50=108ms p95=283ms/256 samples including pauses; no60FPS or attributed improvement. Beta/Production unchanged; no rung closure. Retained delivery buildroot/knowledge/gates/beach-body-resume-20260923-delivery.json; served proof beach-body-served-review-20260923.json41764B SHA812496e63325a3d308435fb6d729b7402461f65637b1837c35221b0c7dc93f0e,3 reviewed local screenshots. Next: durable prepared readiness and identity-first replay through existing renderer; no WASM rebuild. 2026-09-23 measure 2026-09-23 Alpha prepared-character readiness repair published at /alpha/beach title b1790164131: page1218872B SHA52b324caa21e07a1dc793e57050a4201214c108be12a6dc1a59cb6dcade1d695; shared451054B SHAd986bf9b35d1b4e154f3c2b640383584299fb1d7ec30dd7442ad9e82e407be6e unchanged, plain709fc0b3 unchanged. Existing prepared runtime and host renderer converge durable readiness independent of resident culling, snapshotted id+three-artifact replay identity, conflict-aware pending requests and final renderer-lifetime refusal. Matching replay preserves body/allocation/configuration. Shared templates initialize with zero resident rows while row-dependent draws stay gated; zero-row validation cannot relabel old resident submissions. Canonical f3893ffa/24a2dcba guarded adoption and complete511612B readbacks; native pagefit32/32, full JS referee20795refs0reports with negative controls, exact-function isolated Chromium66/66 with synthetic boundary stated; independent reviewcef90c3f accepts scoped change. Existing Alpha workflow1790164641 completed63rc0 phases, all30 dependencies unchanged, prior79cef817 retained. Publication verified, served-review result pending at 2026-09-23 12:07:00 UTC. State: published; waiting for retained actual WebGPU interaction/visual review, next join root proof before scoped acceptance. Retained delivery buildroot/knowledge/gates/beach-prepared-readiness-20260923-delivery.json SHAfd2724eee3ba3586cd3dea488e3013e51a10e3dc7439b6b2bb94bdfb84ad57da; adoption4d6be557 and full publisherd9e3f1d4 retained. No appearance/performance improvement or Beta/Production acceptance claimed; skin/eyes/hair and anatomical support remain rejected/unqualified. 2026-09-23 measure 2026-09-23 served Alpha readiness review complete: https://nishifamily.com/alpha/beach title b1790164131, page1218872B SHA52b324caa21e07a1dc793e57050a4201214c108be12a6dc1a59cb6dcade1d695; shared d986bf9b unchanged. Actual Chromium143 Intel xe-lpg WebGPU confirms offscreen import and same-id already-loaded replay with identical body handle/revision/anchor/counts and artifact requests7→7. Zero resident-row inference uses the existing valid hair-diagnostic guard, not stale draw telemetry. Same-source distinct scene IDs bind1048577/2097153. Frame/orbit/Return pass; imported Face explicitly refuses and retains Body admission. Conservative whole-mesh proxy blocks enabled, permits disabled traversal and blocks opposite-side motion after reenable; same handle revisions1→2→3. One authored clip ends at1252698752Q with changing bounds while second instance stays revision1; world navigation remains unbound. Dated Progress summary visible, prior history preserved, page-script errors0, test context closed. Complete native-hashed served proof buildroot/knowledge/gates/beach-readiness-served-review-20260923.json38183B SHAe7001089603445e100ee7a9f9bea75b7eab0f11d521c4887308071c9328c5896; delivery buildroot/knowledge/gates/beach-prepared-readiness-20260923-delivery.json. Scoped readiness/body-camera interaction accepted; GR35/GE31 remain open for rejected skin/face/eyes/hair, anatomical/sole support, arbitrary occlusion, autonomous navigation/evaluation and matched performance qualification. No60FPS/photoreal/Beta/Production claim. Both telemetry IDs have ERR_ABORTED plus204/accepted observations; cause/attempt counts unknown. Measured aiming, hidden-workspace30s timeout, asynchronous reads and block-local evidence retention are observer corrections, not delivered runtime repairs. Screenshot binaries remain local. Prior negative unmatched performance evidence retained; no token/time/swarm ROI claim. |
GR36 FACE and MORF are consumed, so the face is geometry and not paint · cell WATCHING · gate UNRECORDED
| Contract | gpe_morf_consume · executor Organ · 2 u · stands on GR30 |
|---|---|
| Definition of done | The face is painted in a fragment shader from head-height ratios with NO NOSE TERM at all, while the served asset carries FACE:4700 and MORF:98634 and the page three NXA consumers dereference 15 tags that include neither. The producer nx_nxa_morf is LIVE and its consumer was never written; 7 of 9 face organs are REGISTERED-DARK. CARRIES ITS OWN HISTORY AS A PRECONDITION: face relief v1 was reverted 2026-08-03 for goblin noses because this mesh normals are majority-wrong, so the DONE-RULE is (a) the normal precondition is measured and stated before any relief is applied, (b) FACE and MORF are dereferenced by the page consumers, and (c) the referee panel grades the result at head framing at or above the painted baseline -- a rung that only makes the nose exist has reproduced 2026-08-03 |
| Matrix cell | FACE: the FACE and MORF sections the asset already carries are consumed · runtime/nx_game_page_emit.nx · WATCHING decl |
| Gate | nx_game_page_emit_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | no log row names this rung yet |
GR37 The JS-logic ratchet is pinned against class laundering · cell WATCHING · gate NO-GATE-DECLARED
| Contract | jl_class_pin · executor Organ · 1 u · stands on nothing |
|---|---|
| Definition of done | The ruler that enforces the estate loudest operator order is editable by the party it measures: knowledge/jslogic_class.conf maps each literal hash to a CLASS token, and moving a literal from LOGIC to INTEROP or SHADER removes its bytes from logic_unknown with no code change and no re-baseline. The conf self-invalidates on a literal EDIT (the hash moves, the row reads UNKNOWN) and NOT on a class-token edit, so the one mutation that launders the number is the one mutation it cannot see -- measured: the headline improved 228377 to 128559 on 2026-08-31 while zero JS bytes were removed and the two journal rows total 229911 identically. DONE-RULE: the census reads a declared digest of the class column and REFUSES by name when a class moved without the baseline being re-declared, with TWO neg-controls -- an ordinary literal edit must still self-invalidate the way it does today, and a class move WITH a re-declared baseline must pass. Scoped away from nx_jslogic_census_gate, which a sibling lane wrote the same day for the uncharged-JS axis and which is SOURCE-ONLY: extend that gate, never add a second ruler beside it |
| Matrix cell | REFEREE: the JS-logic ratchet cannot be laundered by editing a conf · runtime/nx_jslogic_census.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_jslogic_census; declare nx_jslogic_census TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR38 The character judge survives the frame it is pointed at · cell LANDED · gate NO-GATE-DECLARED
| Contract | cjc_frame_guard · executor Organ · 1 u · stands on nothing |
|---|---|
| Definition of done | The world lane has no working referee for its own flagship frame: nx_charjudge takes SIGSEGV signal 11 at 0x41510f touching 0x0 on the beach capture and nx_addr2line refuses because the elf was built without -g, while nx_percept anchor corpus is entirely NXFH1 400x240 so a 1920x1200 capture is not comparable to it. The one number the lane CAN produce is the honest headline of this board: nx_frame_score on the native beach capture reads quality=408 verdict=BASIC against the ruler own published random-squares control of 628 RICH. DONE-RULE: the judge grades a full-resolution frame or ABSTAINS by name and never dies, a debug build exists so the next crash is locatable, and the abstention is a THIRD state that can neither acquit nor convict |
| Matrix cell | REFEREE: the character judge survives the frame it is pointed at · runtime/_hdl_build/nx_charjudge.nx · LANDED decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_charjudge; declare nx_charjudge TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-09-04 retract THE RECORDED MECHANISM FOR THIS RUNG IS WRONG AND THE CORRECTION MATTERS MORE THAN THE FIX. The matrix row and this rung both state that nx_charjudge takes SIGSEGV signal 11 touching 0x0 ON THE BEACH CAPTURE, and pair it with the anchor-corpus size mismatch as two independent reasons the panel cannot grade the surface the operator is looking at. RE-MEASURED 2026-09-03 against a live capture taken this session (2868x1604, IHDR depth 8 colour type 2 non-interlaced, 3297194 B): IT GRADES IT, rc=0, CHARJUDGE=533 composition=1000 palette=835 contour=533 face=1000. It also grades 4227x2099 and 1024x8000 at rc=0. THE FRAME WAS NEVER THE TRIGGER AND FRAME SIZE NEVER BLOCKED THIS LANE. THE TRUE MECHANISM, REPRODUCED THREE TIMES: argv is NULL-TERMINATED, so calling nx_charjudge with the image and NO LABEL makes argv[2] the terminator and printing that label dereferences 0x0 -- exit 139 which is 128+11. Two arguments grade; one argument crashes; zero arguments took a DIFFERENT wrong path and reported a missing PATH as a READ failure. A CRASH ATTRIBUTED TO THE DATA WHEN IT BELONGS TO THE CALL SENDS EVERY READER INTO THE WRONG ORGAN -- here it sent them into the renderer and the anchor corpus. The anchor-corpus 400x240 mismatch is a SEPARATE and still-open fact; only the crash half is retracted. 2026-09-04 land cjc_frame_guard SHIPPED into runtime/_hdl_build/nx_charjudge.nx with gate nx_charjudge_frame_gate.nx at 18/18 GREEN. The old main() dereferenced FOUR pointers unchecked (argv entries, the decoded header, the pixel buffer, and two mmap results) and validated WIDTH while never checking HEIGHT at all. The guard is a PURE function taking pointer values as integers, so every refusal path is reachable in-process without manufacturing a corrupt PNG -- a gate that could only test what a real file produces would leave the null-pointer branches, which are the ones that actually crashed, permanently unexercised. TEN NAMED STATES: OK, NO-LABEL-ARGUMENT, NO-PATH-ARGUMENT, DECODER-RETURNED-NO-HEADER, DECODER-RETURNED-NO-PIXELS, NON-POSITIVE-DIMENSIONS, UNSUPPORTED-CHANNEL-COUNT, PIXEL-BUFFER-SHORTER-THAN-DIMENSIONS, FRAME-AREA-OVERFLOWS-THE-BUFFER-SIZE, FRAME-BUFFER-ALLOCATION-REFUSED. DONE-RULE MEASURED CLAUSE BY CLAUSE: it grades a full-resolution frame (byte-identical output to before the change on the documented two-argument call, so no consumer regresses); the one-argument call that was exit 139 now answers abstain NO-LABEL-ARGUMENT graded false at exit 4; the abstention is a THIRD state distinct from graded 0, read 2 and decode 3, and it prints NO axes so nothing can read a score out of a refusal; and --debug now emits debug-line rows=1813 files=15 so the next crash is locatable. The overflow fence is DERIVED from one i64 per pixel rather than picked, and a tooth asserts the largest safe area is still ACCEPTED so the fence cannot be one step too tight and read as robustness. 2026-09-04 retract CORRECTING MY OWN RETRACTION FROM MINUTES AGO, WHICH WAS HALF WRONG AND I PUBLISHED IT TOO FAST. I wrote that the recorded SIGSEGV-on-the-beach-capture mechanism is wrong because a live 2868x1604 capture grades at rc=0. THAT MEASUREMENT WAS TAKEN AGAINST A STALE LAPTOP TREE: nx_charjudge.nx is 2723 B on the laptop against 3328 B on the NAS, and nx_charjudge_lib.nx is 15879 B dated Aug 7 against 29466 B on the NAS -- the lib is nearly DOUBLE and the laptop build could not even compile the NAS CLI because CHJ_NOUT does not exist locally. WHAT STANDS: the argc mechanism is REAL AND PRESENT IN THE NAS SOURCE TOO -- I read it, and argv[2] is dereferenced with no argc check in both copies, so a call with no label dereferences the argv terminator at 0x0 in either. WHAT DOES NOT STAND: my claim that the frame was never the trigger. I proved that only for a lib fourteen kilobytes behind the shipping one, so a frame-related crash in the CURRENT judge is NOT excluded and the row's frame half is UNPROVEN rather than refuted. A CROSS-TREE CONTROL IS NOT A CONTROL: this is the same error as comparing two build lanes and attributing the difference to the code, which this estate recorded earlier the same day, and I repeated it within hours. The anchor-corpus 400x240 mismatch remains separate and open. |
GR39 A character crop that contains a character · cell WATCHING · gate NO-GATE-DECLARED
| Contract | ws_char_crop_nonempty · executor Organ · 0.5 u · stands on nothing |
|---|---|
| Definition of done | nx_world_snap char reports CHAR-FRAME framed=1 ground_y=11 and then CHAR-CROP pixels=0, so framing succeeds and the crop returns no subject image and every downstream character judgement about a world capture is made about an empty picture or not at all. DONE-RULE: a tooth asserts pixels>0 whenever framed=1, the crop refuses by name rather than emitting an empty image, and the hand-rolled NXFH1 writer at nx_world_snap lines ~515-555 is replaced by the shared codec nxl_dump so the estate keeps ONE frame writer -- the duplicate-ruler half is part of this rung, not a separate one |
| Matrix cell | REFEREE: the character crop actually contains a character · runtime/nx_world_snap.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_world_snap; declare nx_world_snap TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-09-07 measure Rigview release rigview-20260907-asset-ui is SERVED, VISUAL ACCEPTANCE PENDING. Added visible asset selection, dated release summary and worklog at /world/rigview; ref9e selects the beach asset. Strict selector18/18 and whole script syntax passed. Origin body5773 contains controls and release feedback. Browser discovery empty, so model rendering and interaction not visually accepted; no model appearance or physics improvement claimed. GR39 completion status is unchanged. |
GR40 The board ranker measures every row · cell WATCHING · gate UNRECORDED
| Contract | rk_measure_all · executor Organ · 1 u · stands on nothing |
|---|---|
| Definition of done | THE META-RUNG, and it gates the trustworthiness of every other number on this page. nx_compare_rank sets mpres=1 for any row not spelled _ABSENT_ and then does if mpres[mi]==1 { done=1 } else { done=rk_measure(...) }, so it measures only the rows already spelled absent and trusts every row spelled present: a rung can be promoted open to done by measurement and can never be demoted done to open, which makes a false green invisible by construction and computes the critical path on corrupt input. The emitter does not share the defect (it calls sd_present on both branches), which is why one row read not-measured on the served page and #1 FINISH in the ranker on the same day. DONE-RULE: the ranker measures both branches; a bite proves a present-spelled row with an absent symbol flips done to open; and because the real risk is a silent fleet-wide re-baseline, the run must publish the before/after done count for every domain in the same output |
| Matrix cell | FOUNDATION (meta, no rival claim): the board ranker MEASURES every row instead of trusting its spelling · runtime/nx_compare_rank.nx · WATCHING decl |
| Gate | nx_compare_rank_caps_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | no log row names this rung yet |
GR41 The review viewers gain the sovereign wasm tier · cell WATCHING · gate NO-GATE-DECLARED
| Contract | nre_sov_tier · executor Organ · 2 u · stands on nothing |
|---|---|
| Definition of done | OPERATOR 2026-09-02 (rigview printed NXA REFUSED no webgl2): nx_nxa_rig_emit, nx_nxa_view_emit and nx_nxa_stream_emit have exactly one renderer, WebGL2, and a refusal leaves a review instrument blank. gameengine GE28 makes the refusal name its cause; this rung removes the dependency. The estate already owns the pieces: nx_trimesh (the sovereign textured triangle rasteriser ntb_view uses server-side) and nx_meshview_wasm (the wasm memory model, 64x64 and 256 verts today). Done when a wasm build of the trimesh rasteriser sized for a 14k-vertex rig renders the same NXA the GL tier renders, skinned from the same SKEL and SKIN sections, and the rigview page falls to it with the badge naming the tier -- proven by a headless capture whose figure crop grades on cjc_panel inside the GL tier's band |
| Matrix cell | REVIEW VIEWERS: the rig, view and stream pages gain the sovereign wasm tier · runtime/nx_nxa_rig_emit.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_nxa_rig_emit; declare nx_nxa_rig_emit TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-09-05 measure rigview inventory against the DanceXR viewer, 2026-09-05 (charsim R26 to R36 carry the rungs): rigview serves one NXA (/world/ref9d.nxa, 104 joints, 14164 verts, 4 influences) with ribbon hair, a chest-pinned dress, body colliders, flesh springs and a Natural or Clothed toggle; DanceXR loads PMX XPS FBX by drag and drop, plays VMD and BVH with on-the-fly bone fixes, stages many actors, and runs alive, physics-tool, camera and audio layers -- GR41's sovereign-tier viewer is the page those rungs land on and GR25 garment penetration stays the ruler for its cloth |
GR42 The beauty tier's first frame: the beach scene rendered offline by a sovereign path tracer and published beside the live frame with its referee score · cell WATCHING · gate NO-GATE-DECLARED
| Contract | env_beauty_scene · executor Organ · 8 u · stands on GR26, GR10 |
|---|---|
| Definition of done | OPERATOR 2026-09-02: get things Infinigen generating to photoreal. Infinigen renders offline (Cycles path tracing, minutes per frame); its twin here is the BEAUTY TIER the matrix already declares ABSENT on nx_gsplat (env_beauty_scene) with the GI contract on nx_gi_path (gp_bounce_gi, also ABSENT: no source exists). Done when (a) nx_gi_path.nx exists as a sovereign integer path tracer over a scene the procgen planes already produce (heightfield terrain from nx_worldgen with a step derived from the pixel footprint so no terracing, water, the physically-based sky of GR26, the cast meshes with their TEXM maps), Monte Carlo bounce with a seeded deterministic sampler so two runs are byte-identical, (b) env_beauty_scene assembles that scene from the same genome and recipe planes the live page reads, (c) the frame is emitted at 1920x1200 with depth, normal and semantic ground truth beside it (the dataset-factory half), (d) it is published on this board beside the live capture and the mirrored Infinigen reference frame, and (e) a referee that PASSES its own ordering ladder grades all three -- until that referee exists the cell reads UNGRADED, never GREEN. The live real-time tier then chases this frame rung by rung (GR26 sky, GR27 ground, GR14 linear light, GR34 normal maps, GR16 hair, GR29 skin) with the same referee. |
| Matrix cell | WORLD-ROADMAP (BEAUTY TIER): film-quality environments -- the actual Houdini-exceed lane · runtime/nx_gsplat.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_gsplat; declare nx_gsplat TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-09-02 queue first byte: nx_gi_path.nx composing nx_worldgen's heightfield and nx_perlin, primary ray + shadow ray + N-bounce diffuse with a hashed sampler, sky as the light source, PNG out through nx_png_write; the laptop builds and runs it (the NAS build lane is admission-refused all day). Reference frame: infinigen.org mirrored 2026-09-02 (cmp_graphics_infinigen_org.html, pin fc604ea3); its media URLs are the next fetch so the comparison has a pinned image, never a remembered one. 2026-09-02 measure FIRST FRAME 2026-09-02, laptop toolchain: nx_gi_path.nx compiled sovereign (asm 228,072 B, ELF 51,371 B) and rendered seed 77 clear-day at 480x288, 4 spp, 2 bounces in 21,959 ms: 751,158 marches, 362,814 sky rays, 188,945 water rays, 199,399 terrain hits, PNG 157,474 B, deterministic by construction (seeded integer LCG). SEEN: continuous slopes with NO terracing (bisection-refined hits), sky-lit shade in the valleys, mirrored lake -- and the frame is PALE: the sky model and palette are display-encoded values that were transported as linear and encoded again, the double-transfer defect GR14 already banked. Fixed in source the same hour (gp_lin decode before transport, one encode at output); second frame owed. Reference pinned the same hour: infinigen.org/img/random_sample.jpeg (447,956 B, pin 134fe4cf, a 3x2 mosaic of Infinigen scenes: birch forest, fern creek, coral reef, beach rocks, cactus desert, canyon). UNGRADED by rule: nx_uiq_visual reports RENDERED for both ours and the live capture (guardrails only), nx_percept's ladder is RED, so the comparison is eyes-only until the referee rung lands. 2026-09-02 land nx_gi_path BUILT ON THE NAS the moment the queue found headroom (59,288 B, sha a2fd860c, src 5b0c6b22) and PROMOTED + REGISTERED (live nishihost/nx_gi_path.elf). Source then split into nx_gi_path_lib.nx (the tracer, 13 functions), nx_gi_path.nx (CLI + the gp_bounce_gi watch symbol) and nx_gi_path_gate.nx (inherits nx_gate_verdict: 15 teeth, 4 named neg-controls, in-process, under a second). Gate GREEN 15/15 on the laptop; BITE-PROVEN against the subject: GP_REFINE 8 -> 0 in the lib turns it RED 13/15 on exactly the refinement teeth, restore verified by hash (513c7f4c) and back to GREEN. Two defects the gate found in its own first run and that are now fixed in source: an 8-bit linear decode could not round-trip display values (c=102 -> 40 -> 100), so radiance now carries 10 fractional bits; and the seed neg-control was VACUOUS because its probe ray landed on the lake (a deterministic mirror event) -- the fixture now searches pitches until a terrain hit is reached and asserts it. All three sources landed on the NAS; NAS rebuilds of nx_gi_path (from the split) and nx_gi_path_gate requested; the gate's promote, registration and organ_gate.conf pairing follow when they build. 2026-10-01 land THE TRACER'S FIRST EVENT TAKES THE CAMERA'S UNIT (2026-09-29, for pano360 PN25): gp_trace_u, gp_march_u and gp_norm3_u in nx_gi_path_lib and wg_sky_u in nx_worldgen carry a per-ray unit u for the primary ray (positions d*t/u in the march and refinement, twater with u, the sky's up and sun terms divided by u), bounce rays stay fx1024 Monte Carlo draws, and every old name is kept as its fx1024 wrapper. PROVEN BYTE-NEUTRAL for every fx1024 caller by a MATCHED-VINTAGE control: the pre-change nx_worldgen and nx_gi_path_lib were rebuilt by reversing the exact edits on scratch copies, proven by hash against the flow-back's agreed base (2560674f, 513c7f4c), compiled in the same lane the same hour, and the control pinhole render (seed 77, 480 x 288) came out eba1436c -- identical to the candidate. nx_gi_path itself is NOT promoted: its served binary (577d0bb3) is 26.6 days old and a rebuild carries other seats' work, so its ship is this lane's owner's, with the stage named 2026-10-01 lesson A NEUTRALITY TEST AGAINST A BINARY OF ANOTHER VINTAGE CONVICTS A NEUTRAL CHANGE: the pinhole rebuilt with the generalised tracer differed from the served nx_gi_path while every ray-event count matched; the served binary predated other seats' edits. Rebuild the pre-change source in the same lane the same hour and compare against THAT 2026-10-01 measure TRIANGULATED BAR as of 2026-10-01 for reflections and global illumination: the best in class is three variance-reduction families, not one method. Resampling: ReSTIR reuses candidate light samples spatially and temporally, unbiased at equal error 6-60x faster than the prior best and biased 35-65x faster at some energy loss, 3.4 million emissive triangles under 50 ms at 8 rays per pixel. Learned guiding: practical path guiding learns the radiance field in an SD-tree, unbiased, no hyperparameter tuning, MSE 0.018 against 0.104 for Vorba et al. at equal time, and Intel ships the family as a production library with per-frame learned fields on surfaces and in volumes and product sampling with BSDF lobes. Fitted transport: the September 2026 entry minimises the rendering-equation residual over a 13D Gaussian mixture and renders view-independent solutions in milliseconds. The September cs.GR listing (170 entries) carries only that paper and RenderFormer-V2 as light-transport titles, none ReSTIR or path guiding, October has no entries yet, and arXiv holds only three ReSTIR entries, newest ToF ReSTIR in May 2026 ; witnesses [bitterli2020] [ppg2017_t1001] [openpgl071_t1001] [arxiv_siga_rtgi_2026_09_27] [csgr202609_gfx_t1001] [csgr202610_gfx_t1001] [restir_search_t1001] ; where they disagree: ReSTIR buys real time by sharing samples across pixels and frames and accepts bias or correlation, path guiding stays unbiased and pays a training phase and memory, the Gaussian fit replaces sampling with an optimisation whose view independence makes glossy reflections its weak case, and the Intel library calls itself pre v1.0 with an API in flux while the 2017 paper calls the same family production compatible ; the estate's running lane builds a brute-force unidirectional integer path tracer (nx_gi_path: primary ray, shadow ray, N-bounce diffuse with a hashed sampler, the sky as the light, a heightfield march with no triangle BVH; first frame 4 spp and 2 bounces at 480x288 in 21.96 s) and it is BELOW the triangulated best because it carries none of the three reducers -- no reservoir resampling, no learned guiding, no cached or fitted transport -- so its noise at equal time is the noise all three witnesses were built to remove 2026-10-01 measure THE ReSTIR FAMILY BEYOND THE 2020 PAPER, MIRRORED AND READ 2026-10-01 from NVIDIA Research's open copies: ReSTIR GI, HPG 2021, five NVIDIA authors: resampling of multi-bounce indirect paths, MSE 9.3x to 166x better than path tracing at one sample per pixel, 8.9 ms against 8.0 ms equal-time at 1080p on an RTX 3090 with MSE improved 15 to 1 [restirgi2021_bf1001]; GRIS and ReSTIR PT, TOG 41(4) 2022, Utah and NVIDIA: variance bounds and convergence conditions for reuse of correlated samples with unknown PDFs, Fig. 2 shows unbiased ReSTIR reuse converging to a wrong result, ReSTIR PT shifts whole paths between pixels [gris2022_bf1001]; Area ReSTIR, DOI 10.1145/3658210, Utah and NVIDIA: reservoirs over each pixel's 4D film and lens ray space for antialiasing and defocus, Fig. 1 times 4.2 ms path tracing, 9.5 ms ReSTIR PT and 14.0 ms Area ReSTIR, its body text not recovered by the extractor [arearestir2024_bf1001]; with Bitterli et al. 2020 [bitterli2020] the family is now read as four methods. COUNTED BY GROUP it is still ONE lineage: every paper has NVIDIA authors and Kettunen, Wyman, Lin or Bitterli recur across them (NVIDIA with Dartmouth in 2020 and with Utah in 2022 and 2024), so these fetches add method witnesses, not independent groups, and the GR42 bar's group independence remains unmet 2026-10-01 queue a ReSTIR witness from a group with no NVIDIA author is still owed (a candidate lead named from memory and unverified: AMD's world-space spatiotemporal reservoir reuse for ray-traced global illumination, 2021); the newest ReSTIR entry on the board's listing is ToF ReSTIR (May 2026, DOI 10.1145/3811299), whose authors were not read [restir_search_t1001] 2026-10-01 measure THE FIRST ReSTIR WITNESS FROM OUTSIDE THE NVIDIA LINEAGE, MIRRORED AND READ 2026-10-01 (the 1790830011 queue row's lead, now verified): Boisse, World-Space Spatiotemporal Reservoir Reuse for Ray-Traced Global Illumination, SIGGRAPH Asia 2021 Technical Communications, PDF from AMD's GPUOpen site, a single author on none of the four NVIDIA-lineage papers: reservoirs of path vertices cached in a GPU hash grid and reused across space and time, so light can be sampled at any vertex of the eye path rather than only where screen-space reuse reaches [worldspace_restir_amd2021_bf1001]. Counted by group the ReSTIR bar now stands on two lineages, NVIDIA with Dartmouth and Utah [bitterli2020] [restirgi2021_bf1001] [gris2022_bf1001] [arearestir2024_bf1001] and the GPUOpen author, both peer reviewed; a third independent group, and a 2026 entry of the family read in full, are still owed |
GR43 A ray-triangle bounding-volume hierarchy for the beauty tier: the path tracer hits the cast meshes · cell WATCHING · gate UNRECORDED
| Contract | gp_bvh_ray_tri · executor Organ · 2 u · stands on GR42 |
|---|---|
| Definition of done | MATH GAP MEASURED 2026-09-03 (nx_capsearch over 7245 organs, corpus_complete=1: nx_bvh_lib is BioVision mocap, no ray acceleration structure exists by name), so nx_gi_path marches a heightfield and cannot hit a triangle; done when a BVH over the cast's NXA triangles (median split on integer centroids, deterministic from the mesh) is traversed by the path tracer, a brute-force every-triangle oracle agrees on every ray of a fixture (same nearest hit, same barycentrics), the build and traverse costs are printed per frame, and a neg-control with one corrupted node bound is caught by the oracle |
| Matrix cell | BEAUTY: a ray-triangle BVH so the path tracer hits the cast meshes (GR43) · runtime/nx_gi_path_lib.nx · WATCHING decl |
| Gate | nx_gi_path_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | no log row names this rung yet |
GR44 A low-discrepancy sampler for the path tracer: Sobol with Owen scrambling, the blue-noise organ reused for the screen · cell WATCHING · gate UNRECORDED
| Contract | gp_sobol_owen · executor Organ · 1.5 u · stands on GR42 |
|---|---|
| Definition of done | MATH GAP MEASURED 2026-09-03: the tracer draws from a hashed LCG and no Sobol or Halton sampler exists by name (nx_bluenoise exists for screen-space masks); done when a deterministic Sobol sequence with per-pixel Owen scrambling drives every bounce, variance at equal spp on a fixture scene is measurably lower than the LCG with the number printed and the LCG kept as the control, two runs are byte-identical, and the sampler is one lib that both the offline tracer and the real-time reconstruction (performance PF4) read |
| Matrix cell | BEAUTY: Sobol with Owen scrambling drives every bounce, the LCG kept as the control (GR44) · runtime/nx_gi_path_lib.nx · WATCHING decl |
| Gate | nx_gi_path_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | no log row names this rung yet |
GR45 Spherical-harmonics irradiance from the physically-based sky for the real-time tier · cell WATCHING · gate NO-GATE-DECLARED
| Contract | sr_sh_irradiance · executor Organ · 1.5 u · stands on GR26 |
|---|---|
| Definition of done | MATH GAP MEASURED 2026-09-03: no spherical-harmonics or irradiance-probe organ exists by name and the live tier lights with a two-stop ramp; done when the GR26 sky is projected to SH9 in NishiLang integer arithmetic (nx_fixq30_lib), the shader emitted by nx_wgsl from the same source evaluates it per normal, a reference integration agrees within a stated band on a fixture sky, and the frame is graded by the world referee against the ramp with the neg-control of the projection zeroed |
| Matrix cell | LIGHT: spherical-harmonics irradiance projected from the physically-based sky for the real-time tier (GR45) · runtime/nx_sh_irradiance.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_sh_irradiance; declare nx_sh_irradiance TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR46 Cascaded shadow maps on the kernel and WebGPU tiers: the cast casts a shadow on the ground · cell WATCHING · gate UNRECORDED
| Contract | rk_shadow_cascade · executor Organ · 2 u · stands on GR42 |
|---|---|
| Definition of done | MATH GAP MEASURED 2026-09-03: no shadow-map or cascade organ exists by name (the served beach draws dark blobs under feet); done when the GE42 kernel renders a sun-depth map for N cascades chosen from a conf ladder row, the world pass samples it with a bias DERIVED from the texel footprint, a fixture with a known occluder produces the analytically expected shadow footprint within a band, and the served beach frame is graded by the world referee with the maps forced off as the neg-control |
| Matrix cell | LIGHT: cascaded shadow maps on the kernel and WebGPU tiers, the cast shadowing the ground (GR46) · runtime/nx_wgsl.nx · WATCHING decl |
| Gate | nx_wgsl_compute_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | 2026-09-04 measure I LOOKED AT THE SERVED FRAME FOR THE FIRST TIME TODAY AND IT INDICTS THE WHOLE SESSION'S PRIORITY. METHOD, reproducible by any seat with the laptop wire and zero NAS load beyond one page fetch: python nishi-ops/ac_cdp.py https://nishifamily.com/world/beach?review=0 35 document.getElementById(c).toDataURL(image/png) piped to a file, base64 decoded locally -- 302621 B PNG, magic verified, canvas 519x260, npc_up=1, gl2=1, door webgl2, mt n=8 shared=true. WHAT THE FRAME SHOWS, and every item is visible in the capture rather than inferred: ZERO shadows and ZERO ambient occlusion, twelve figures none of which is grounded by a contact shadow; voxel terrain built from large cubes carrying one repeating tile, with stacked-block walls; skin flat and plastic with no subsurface and no specular variation; hair reading as a SOLID SHELL with no strand separation at this scale; no sky, only a flat beige horizon band. The cast pipeline itself WORKS -- twelve mobs, textured swimwear reading teal pink and red, plausible proportions -- so the failure is entirely SHADING AND ENVIRONMENT, not geometry plumbing. THE OPERATOR'S RECORDED REJECT WORDS ARE CONFIRMED VERBATIM AGAINST PIXELS: flat light, no shadows/SSS, wig-shell hair, voxel sand, PS2-era look far from MetaHuman UE5. ★★★★★★AND THE FINDING THAT MATTERS MOST: AT THIS QUALITY THE HAIR CLIPPING IS NOT VISIBLE. The hair does not read as strands, so the GR25 child-offset clamp -- a 9 to 17 mm inward correction on interpolated children -- corrects something no viewer can perceive. I spent a full session on a sub-millimetre geometric detail while the render fails at the level of lighting, materials and terrain, after the operator had already said below PS2 level in the first message. RANKED BY WHAT THE FRAME ARGUES, not by what I was working on: 1 shadows and AO, GR46 rk_shadow_cascade, because nothing is grounded and flat light reads as fake instantly and its own rung text already says the served beach draws dark blobs under feet; 2 the voxel terrain and its tiling texture, which reads as a 2003 block game regardless of character quality; 3 skin subsurface, GR29, because the plastic mannequin look survives any resolution increase; 4 a hair BSDF, GR16, anisotropic highlight and strand separation, which must precede any further hair GEOMETRY work; 5 sky and atmosphere, GR26. OWED AND NAMED RATHER THAN GUESSED: this capture is exactly the artifact the acceptance ledger lacks for its four ref_rows=0 subjects, and it CANNOT be delivered by this seat -- nx_accept referee takes a NAS-ABSOLUTE png path, nx_fs_write is text-only by design, and this seat holds no contentput.cap. THE UNBLOCK IS ONE CAPABILITY, NOT A DESIGN: grant contentput.cap or have a NAS-side beat take the browser capture itself. Until then the referee rows stay honestly empty rather than filled with a tier I cannot prove. 2026-09-04 measure THE FIRST NUMERIC A B BASELINE FOR THE BEACH, AND IT CORROBORATES THE PIXELS INDEPENDENTLY. nx_frame_score on beach.nxfh, the FULL-RESOLUTION native capture this session produced: w=1920 h=1200 pal=599 det1=11 det4=33 grad=35 cov=998 cohere=974 quality=500 verdict=BASIC. THE DELTA: GR38's log records the same ruler on the same subject class at quality=408 BASIC, so this reads +92. ⚠THE CAVEAT TRAVELS WITH THE NUMBER AND IS NOT OPTIONAL: cjc_panel.conf states that rows predating the 2026-08-28 resolution fix CANNOT be compared with anything captured after it, and GR38's capture resolution is unrecorded here, so +92 is a CANDIDATE improvement and not a confirmed one until that resolution is known. Whoever confirms it should record both resolutions in the same row. ⛔AND THE COMPARISON EVERYONE WILL REACH FOR IS FORBIDDEN BY THE RULER ITSELF: the random-squares control scores 628 against the beach's 500, and quoting that as quality is explicitly refused in the organ's own calibration line -- composite is a STRUCTURE floor, not a quality rank, and values order by statistics rather than perceived quality without a same-class perceptual comparison. A seat that publishes our world scores below random noise has committed the session's recurring defect, measuring a narrower thing than the claim. ★★★★★★THE COMPONENTS ARE THE VALUABLE PART AND THEY AGREE WITH THE EYE, WHICH IS THE INDEPENDENT CONSENSUS THE OPERATOR REQUIRES RATHER THAN A SEAT'S OPINION: det1=11 and det4=33 say there is almost no fine or coarse detail; grad=35 says the gradient field is nearly flat, which is SHADOWS AND FALLOFF BEING ABSENT expressed as a number; cov=998 and cohere=974 say the frame is near-totally covered and highly uniform. Low detail plus low gradient plus near-perfect coherence is the numeric signature of flat unshadowed untextured surfaces, and it was written down BEFORE this run from a browser capture of the served page -- twelve figures, none grounded by a contact shadow, voxel terrain carrying one repeating tile, hair reading as a solid shell. TWO INSTRUMENTS, ONE CONCLUSION, NEITHER DERIVED FROM THE OTHER. CONSEQUENCE FOR THE RANKING: grad=35 IS GR46 rk_shadow_cascade's case stated numerically, and it outranks every geometry rung on this board including the hair work this session spent itself on. THE BASELINE IS NOW REPRODUCIBLE BY ANY SEAT IN TWO CALLS: nx_world_snap char beach writes beach.nxfh at res_class=FULL, then nx_frame_score beach.nxfh prints the vector above -- so the next capture after a shadow landing is a direct A B against these exact numbers, on the estate's own ruler, with no eye involved. 2026-09-04 measure THE FEET-PLANTED REJECTION IS A SHADING DEFECT MISFILED AS A LOCOMOTION DEFECT, AND THAT IS A THIRD INDEPENDENT LINE OF EVIDENCE FOR SHADOWS BEING THE TOP RUNG. MEASURED by cropping the served-page capture to the front figure's legs and feet, box 322,292 to 432,368 of the 735x368 review capture upscaled 5x: THE FEET ARE PLANTED. Toes are in contact with the sand, the geometry is correct, the ground intersection is right. AND THERE IS NO CONTACT SHADOW OF ANY KIND, so the figure reads as FLOATING despite being geometrically grounded; the legs carry a single flat tan with almost no form modelling. ★★★★★★THE ROW beach-feet-planted-0830 IS FILED AGAINST ga_accept_locomotion with targets charsim:ap_foot_ground and gameengine:mob_step_gait -- BOTH GEOMETRY ORGANS -- so a seat reading that row is sent to fix a foot-placement solver that is already correct, while the actual cause is an absent shadow pass. A MISFILED SUBJECT SENDS EVERY FUTURE SEAT AT THE WRONG ORGAN, and the row's own targets field is what does the sending. THREE INDEPENDENT WITNESSES NOW AGREE ON GR46 AND NONE WAS DERIVED FROM ANOTHER: nx_frame_score on the full native capture returns grad=35, a near-flat gradient field, which IS the absence of shadow and falloff expressed as a number; a browser capture of the served page read by eye shows twelve figures none of which is grounded by a contact shadow; and this leg crop shows correct foot geometry with no shadow beneath it. ⇒GR46 rk_shadow_cascade outranks every geometry rung on this board, including the hair clamp this session spent itself on and including the foot-placement work the locomotion row points at. ⚠AND IT PREDICTS A CHEAP FALSIFICATION: if shadows land and the feet still read as floating, this diagnosis is wrong and the locomotion row was right -- so the next capture of this same crop box against these same numbers settles it either way, with no eye involved. 2026-09-04 measure THE SHADOWS ARE NOT MISSING, THEY ARE VERTICAL -- THE BEACH RENDERS AT LOCAL NOON AND CLIMBS TO EXACTLY OVERHEAD. This corrects a diagnosis I published twice today and it changes the remedy completely. WHAT EXISTS, read from source: nx_game_page_emit carries a full cast-shadow pass at uShd=1 described in its own comment as Blinn 1988 planar projection, the published no-shadow-map answer, plus sun shadows at uSun=1 via a voxel DDA march from the hit point toward the sun, plus contact ambient occlusion over the last .008 H, with the shadow depth DERIVED from sun height and the sky-fill chroma rather than tuned. It is not missing and it is not crude. WHAT GATES IT: uShd and window.__sund and sund[1] greater than NXSHD_MINY and no npcshd=0 in the query. MEASURED LIVE on the served page: NXSHD_MINY=.1 and sund[1]=0.9955, so the gate PASSES BY A FACTOR OF TEN and the pass runs on every frame. THE REASON NOTHING IS VISIBLE: sund[1] is the SINE OF SUN ELEVATION, so 0.9955 is 84.55 degrees -- almost straight down -- and a vertical sun puts every cast shadow DIRECTLY BENEATH ITS OWN CASTER where the caster hides it. Twelve figures stand on sand with nothing under them while the shadow pass works perfectly. THREE SAMPLES FROM ONE PAGE SESSION PROVE IT IS ALSO GETTING WORSE: 83.37 degrees at 20 s page age, 84.55 at 25 s, 86.33 at 75 s -- roughly 3.2 degrees per minute, CLIMBING TOWARD 90. So a visitor's first minutes, and every capture anyone has ever taken including this seat's and the referee's, land in the WORST AVAILABLE LIGHTING. THE SOURCE: sel = ex.sunel() over 4096 from the wasm engine, and THE FALLBACK IS sel = 1 WHICH IS EXACTLY VERTICAL -- an absent export would pin the sun overhead forever, so the default failure mode is the worst possible frame. There is NO time-of-day lever in the query string at all: greps for tod, time, sun, hour, noon and dusk return nothing. THE CORRECTION TO MY OWN NUMBER: nx_frame_score returned grad=35 on this world and I published that a near-flat gradient field IS the absence of shadow. It is not. It is the absence of RAKING LIGHT. Same measurement, opposite remedy -- one calls for a shadow-mapping project, the other for a starting angle. ONE VARIABLE EXPLAINS THE WHOLE OPERATOR REJECT: no visible cast shadows because they hide under their casters; flat plastic skin because vertical light strikes every body surface at a similar angle so there is no form falloff; and voxel sand reading as one repeating tile because no raking light means no dune relief. CONSEQUENCE FOR THIS RUNG: GR46 cascaded shadow maps remain worth having but they are NOT the highest-leverage change. Starting the day at 25 to 35 degrees instead of 90 costs no renderer work, uses the pass that already ships, and is bounded below only by NXSHD_MINY=.1 which is 5.7 degrees. THE A B IS RUNNING AND NEEDS NO CODE: the cycle advances on its own, so the SAME page at roughly 15 minutes of age gives the same scene under raking light -- one variable, same assets, same renderer. If that frame shows grounded figures and modelled bodies the diagnosis is proven; if it does not, this row is wrong and the shadow-map project is back at the top. |
GR47 Integer Cholesky for the correlated identity draw · cell LANDED · gate NO-GATE-DECLARED
| Contract | la_cholesky_q30 · executor Organ · 1 u · stands on nothing |
|---|---|
| Definition of done | MATH GAP MEASURED 2026-09-03: nx_matrix carries determinant and inverse to 3x3 and no Cholesky decomposition exists anywhere by name, so GR24's full-covariance draw has no primitive; done when a Q30 fixed-point Cholesky over nx_fixq30_lib arithmetic factors any SPD matrix up to the identity-space dimension, L times L-transposed reproduces the input within a stated Q30 band on random SPD fixtures, a non-SPD input is refused by name rather than producing garbage, and GR24 composes it for its draw (one ruler, never a second decomposition) |
| Matrix cell | IDENTITY: integer Cholesky so the correlated identity draw has its primitive (GR47) · runtime/nx_linalg.nx · LANDED decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_linalg; declare nx_linalg TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-09-04 land INTEGER CHOLESKY LANDED IN THE ORGAN THE MATRIX ROW NAMES: la_cholesky_q30 plus la_is_symmetric, la_chol_max_resid_q30 and la_chol_status_name added to runtime/nx_linalg.nx, byte-identical both trees at 12378 B, with nx_linalg_chol_gate.nx 11037 B at 26/26 GREEN. Q30 Cholesky-Banachiewicz over nx_fixq30_lib (fq_mul, fq_div, fq_sqrt) as the rung required, composing that arithmetic rather than adding a second one. CLAUSES 1 TO 3 OF THE DONE-RULE ARE PROVEN AND CLAUSE 4 IS NOT CLAIMED: GR24 composing it for the correlated draw is GR24 own rung and remains open, so this is the primitive landing and not the draw. 2026-09-04 measure THE BAND IS MEASURED AND THE FIRST ONE I WROTE WAS DECORATIVE. Worst absolute entry of A minus L L-transposed, in Q30 units: identity 0, textbook 3x3 0, irrational-root 3x3 3, dimension-6 4, and dimension-32 4 at the declared cap. THE RESIDUAL IS FLAT IN DIMENSION, so the error is dominated by per-operation truncation rather than by accumulation -- a result worth having, because it is the opposite of what the header first asserted and the header was corrected to say could grow rather than grows. I first declared the band at 4096 and the gate own ratchet tooth REFUSED it: a bar three orders above the worst observation would never catch a regression. Tightened to 64, which is 16x the worst measured and about 6e-8 relative to FQ_ONE, and pinned in BOTH directions so it can be neither loosened into decoration nor tightened into a permanent red. THE LOAD-BEARING TOOTH IS NOT THE RESIDUAL: a residual check passes for any L whose product happens to reconstruct A, so the factor is compared against a HAND-DERIVED L for the textbook 3x3 and for the 2x2 positive control, and both match EXACTLY at 0 Q30 units. |
GR48 The rendered strand is the collided strand: character geometry leaves emitted JS for NishiLang lowered to wasm · cell NO-MATRIX-ROW · gate NO-MATRIX-ROW
| Contract | gpe_char_geom_wasm · executor Organ · 4 u · stands on gameengine:GE40 (UNRESOLVED: no such rung on this plan) |
|---|---|
| Definition of done | OPERATOR 2026-09-04, the loudest order restated as an absolute: printf-emitted foreign-language logic is not a thin shim we have to accept, it is hacky programming, and first byte up every rung until it HAS to translate is the only way anywhere in the estate. MEASURED THIS SESSION, and it is the ROOT CAUSE of the standing REJECT beach-hair-colour-wind: the hair a visitor sees is built in hand-written JavaScript inside a NishiLang string literal in nx_game_page_emit.nx -- Catmull-Rom subdivision, NXH_GUIDES 1200 by NXH_KIDS 6 child interpolation, per-strand lateral axis, taper, quad emission and UV generation, hundreds of lines of geometry maths running on the main thread at load. The COLLIDER is in wasm (nx_wasm_craft hg_capd2 hg_inside hg_push) and runs on 128 SIMULATED guides. So the collider structurally cannot see the geometry that is rendered, which is why a clamp shipped into that JS on this date changed nothing an operator could see. THREE CONSEQUENCES, each measured: the collider is correct and invisible; the JS is outside the compiler, outside every gate, and outside grep itself (nx_shelltool skips the long emitted lines and reports linetoolong, so absence over that code is UNPROVEN by construction); and a fix in that layer can only ever be a patch. DONE-RULE: the strand construction is NishiLang compiled through nx_compile_wat and lives beside the collider so there is ONE geometry owner, the collider resolves against the RENDERED strand set and a tooth asserts rendered equals collided rather than 128 versus 7200, the JS path is retired only after the wasm path reproduces its vertex buffer within a stated band on a fixture groom, and the referee panel grades a head-framed capture at or above the JS baseline. NOT DONE if the JS block merely shrinks: the acceptance test is that no geometry byte on the page is hand-authored. |
| Matrix cell | NO-MATRIX-ROW — no matrix row carries gpe_char_geom_wasm; the ranker reads this rung as unmapped and no cell can flip for it. Add the row and the join completes on the next publish. |
| Gate | the gate join needs a matrix row: no organ is named for this rung |
| Work log | 2026-09-04 measure THE WASM HAIR ENGINE ALREADY CARRIES THE FOLLOW-STRAND SCHEME AND THE PAGE REIMPLEMENTS ITS SELECTION IN JAVASCRIPT. Seven measurements, each with its instrument, taken 2026-09-04 under the operator order that evidence accompanies every claim everywhere. ONE: nx_wasm_craft.nx:298 const HG_REND_MAX 1200 with the comment render strands the page's shipped NXH_GUIDES budget PRESERVED and moved here. TWO: :299-302 const HG_SIM 128 commented PICKED named as picked, 1200 over 128 is 9.4 render strands per guide, inside the TressFX 4 to 16 follow-hairs band. THREE: :306 const HG_TAB_W 8 with the render-strand table row spelled rest-idx nb1 nb2 w1 w2 lockw spanz 0, so two neighbour guides and two weights per rendered strand already exist as data. FOUR: :5326-5328 hg_x hg_p hg_d are each based at k times HG_SIM times HG_W times 8, so simulated positions are stored for 128 guides and never for 1200. FIVE: HG_TEX_SR is DEFINED at :303 and referenced only inside the comment at :302, matches=2 corpus_complete=1, i.e. the constant that sizes the guide texture has ZERO code uses -- the same defined-and-never-referenced class GR37 and the layout-token finding already name. SIX: the wasm copies render strands with src = r*nstr/nrend at :5532 while the emitted JavaScript computes s5 = Math.floor(q5*nstr/nsel) at nx_game_page_emit.nx:889 -- ONE selection rule implemented twice in two languages, and the wasm comment says so itself, the page's own selection rule. SEVEN: hverts, the hair vertex buffer name, returns matches=8 corpus_complete=1 and every hit is in the emitter. DELIBERATELY NOT CONCLUDED: whether some wasm export feeds hair positions to the page under a name not yet searched. Until that is measured this row does NOT claim the collision fails to reach the screen; it claims only the seven facts above. THE NEXT MEASUREMENT IS NAMED SO THE NEXT SEAT DOES NOT REDERIVE IT: enumerate the wasm exports nx_compile_wat emits for nx_wasm_craft and check every one against the emitted page's ex. call sites, then read hg_tick at :6023 to see what it writes. CONSEQUENCE FOR THE RUNG: if the follow-strand table is unconsumed then GR48 is far smaller than scoped -- delete the JS duplicate and consume hg_tab, rather than port a pipeline that already exists in NishiLang. 2026-09-04 retract I REBUILT THE WRONG BINARY, AND THE BUILD STAMP I CITED AS PROOF WAS EVIDENCE ABOUT A BINARY THAT NEVER RAN. Published earlier today: the clamp IS live, proven by the served title carrying build stamp b1788529014 matching beach.html mtime. FALSE, and the discriminating experiment is structural rather than textual: sites/nishifamily/world/beach.html:2149 reads const ox9=hsh9(...) with line 2150 the segment loop ADJACENT, while the source at buildroot/runtime/nx_game_page_emit.nx:889 reads let ox9= followed by the clamp, and the clamp MUTATES ox9 with ox9-=rad9*nrx9 which is impossible under const. The same pre-clamp const form sits in craft.html:2153 vale.html:2149 isle.html:2151. ROOT CAUSE, found by a cross-lane critic and by no single lane: nx_game_page_emit IS A LIBRARY imported in-process, NOT the organ that writes the page. The writer is nx_craft_emit, which declares CE_OUT2 as sites/nishifamily/world/beach.html and calls gpe_build from that lib. I edited the LIB and promoted the LIB's standalone binary, which emits nothing for the worlds; nx_craft_emit.elf carries mtime 1788472251, built 56164 s or 15.6 HOURS before the clamp existed, and beach.html was then re-emitted at 1788529014, 599 s AFTER my edit, BY THAT SIXTEEN-HOUR-OLD BINARY. That is why the page reads freshly written and pre-clamp at the same time, and it is why the mtime ordering source then elf then page looked like it refuted a stale-artifact story: the ordering was true of the wrong artifact. LESSON, and it is the general one: A BUILD STAMP PROVES THE PAGE WAS REGENERATED, NEVER THAT THE CODE YOU EDITED WAS IN THE THING THAT REGENERATED IT -- and before rebuilding, ASK WHICH ORGAN DECLARES THE OUTPUT PATH, because an emitter named for the page is not necessarily the writer of the page. All six worlds come from that one organ, craft beach vale via CE_OUT CE_OUT2 CE_OUT3 and isle crags ridge via the world_recipes.conf loop at nx_craft_emit.nx:618, so ONE rebuild and promote covers them. SECOND FINDING WHICH CHANGES GR48'S PREMISE: hg_tick, the sole entry point of the entire hg_ collision system, occurs EXACTLY ONCE across 23622 files with corpus_complete=1 -- its own definition, with NO CALLER. hg_solve_guide is reachable only from inside it. So HG_CAP_MAX 32 capsules, the 128 simulated guides, hg_push, hg_inside and hg_capd2 are UNREACHABLE DEAD CODE, and the groom baker nx_nxa_groom.nx contains no collide or capsule token either. The live per-frame hair is nxHairTick in the emitted page: spring, viscous damping, wind, length clamp, and NO BODY TERM OF ANY KIND. CONSEQUENCE: GR48 was scoped as move the geometry to where the collider can see it; the measurement says there is no collision running to see anything, so the rung must FIRST wire hg_tick and prove it fires, and the radial clamp is currently the ONLY body-awareness the rendered hair would have once the writer is rebuilt. THIRD FINDING, the operator's testing complaint made mechanical: beach.html:2809 const REV=location.search.indexOf(review=0)<0 means REVIEW IS ON BY POLARITY -- absent token gives -1 which is less than 0 -- and beach.html:2954 if(REV)setInterval(...,16) installs a 60 Hz look-at servo that writes through ex.look, the same door the mouse uses, sending dd/6 against a LOOKS gain of 6 so the division EXACTLY CANCELS and the full remaining angular error is applied every tick, clamped to plus or minus 1800 it-units of a 25736 it-unit turn, i.e. 25.2 degrees. Any mouse-look under about 25 degrees off-target is erased within ONE sim tick and larger deflections are slewed back, while summon re-forms the ring every 900 ticks. Position is yours, ORIENTATION IS TAKEN. review=0 is the only true escape and it also unpins the render scale that REV forces to full. The header comment claiming review=1 is a lever is STALE: no code tests for it. |
GR49 The translation ladder is declared and counted per target: zero hops on NishiOS and Nishi browser, minimum hops elsewhere, an API layer only where a third party mandates one · cell NO-MATRIX-ROW · gate NO-MATRIX-ROW
| Contract | tl_translation_census · executor Organ · 2 u · stands on gameengine:GE40 (UNRESOLVED: no such rung on this plan), gameengine:GE44 (UNRESOLVED: no such rung on this plan), GR37 |
|---|---|
| Definition of done | OPERATOR 2026-09-04: do everything first byte up in NishiLang and estate and engines, then ship to the translation layer if it requires one, but make it so that in NishiOS and Nishi browser we do NOT have a translation layer so we can be superior in performance and capabilities and delivery, and if third party ship via an API layer if mandatory -- the least possible translation is the best. THE ESTATE ALREADY ENFORCES THIS ON TWO AXES AND NEITHER COVERS THE THIRD: GE40 wp_provenance demands every served wasm byte trace to a named .nx, and GE44 rk_webgpu_door demands ZERO shader bytes on the page that do not trace to nx_shader_ir and reduces WebGPU to a submission pipe with no geometry pipeline. NOTHING MEASURES TRANSLATION HOPS PER TARGET, so the JS layer grew to roughly 128559 bytes of logic (GR37) and the character geometry ended up hand-authored inside it with nobody able to see the number rise. DONE-RULE: a conf ladder row per delivery target declares its permitted hop count and its permitted translation kinds, NishiOS-native and Nishi-browser-native declare ZERO with no wasm and no JS and no WGSL in their path, third-party-browser declares the minimum with each hop NAMED and justified by a mandate rather than by convenience, a census walks every served artifact and counts the hops it actually took, the count is a RATCHET that may not rise, and the census REFUSES by name when a target exceeds its declared ceiling. TWO NEG-CONTROLS: a hop added to a zero-hop target must fail, and a hop merely re-classified without being removed must also fail -- that second control is the GR37 laundering lesson applied here before it can happen rather than after. |
| Matrix cell | NO-MATRIX-ROW — no matrix row carries tl_translation_census; the ranker reads this rung as unmapped and no cell can flip for it. Add the row and the join completes on the next publish. |
| Gate | the gate join needs a matrix row: no organ is named for this rung |
| Work log | 2026-09-04 measure THE SOVEREIGN THIRD WITNESS IS COMPOSITION, NOT CONSTRUCTION -- THE WEBSOCKET STACK ALREADY SHIPS IN NISHILANG. OPERATOR 2026-09-04 asked, of the Python browser driver this seat used to read the served page: is it a tool we need to duplicate into nishi lang and use as a 3rd witness. MEASURED ANSWER, by listing buildroot/runtime rather than by estimating: nx_websocket.nx, nx_websocket_client_upgrade.nx WITH nx_websocket_client_upgrade_test.nx, nx_websocket_frame.nx WITH nx_websocket_frame_test.nx, plus nx_json.nx, nx_json_emit.nx, nx_base64.nx, nx_net.nx and nx_hmac_sha1.nx. THE RFC 6455 HANDSHAKE AND THE FRAME CODEC -- which ac_cdp.py's own header calls the roughly forty lines below and treats as the hard part -- ARE ALREADY WRITTEN AND ALREADY TESTED HERE. CDP to a local browser is plain ws with no TLS, so nothing else is needed underneath. ⇒WHAT IS ACTUALLY MISSING IS ONLY THE CDP GLUE: open the DevTools endpoint, send Runtime.evaluate, read the result frame. That is a small organ composed from tested parts, not a project, and it is buildable on the LAPTOP toolchain where a brand-new file has no laptop-versus-NAS fork to converge first -- which matters today because NAS build admission has refused every lane for hours. ★CHECK-BEFORE-BUILD PAID OFF EXACTLY AS THE LAW PREDICTS: I was one decision from writing a WebSocket client from scratch, in a tree that already ships one with tests. THE WITNESS STRUCTURE THIS ENABLES, and each rung is independent of the others by construction: ONE the third-party browser over CDP, which stays outside and exists to falsify us; TWO a NishiLang CDP client -- OUR wire, THEIR engine -- which removes Python from the EVIDENCE path even though it was never in the BUILD path; THREE nx_browser rendering the same page natively -- OUR wire, OUR engine -- which is the zero-translation rung of the operator's own ladder for NishiOS and Nishi browser. ⚠WHY THIS MATTERS BEYOND TIDINESS, measured today: the estate's own searcher is structurally blind to the lines it ships, nx_shelltool skips any line longer than its buffer and every emitted world page is a handful of 200 KB lines, so THE ONLY INSTRUMENT THAT COULD READ OUR OWN SERVED PAGE TODAY WAS A THIRD-PARTY BROWSER DRIVEN BY PYTHON. A sovereign estate whose PROOF depends on someone else's engine has outsourced its witness, and that is the half nobody was counting while congratulating the build path on being clean. |
GR50 The ship loop adjudicates itself against the SERVED artifact and refuses to report success for a provably inert promote · cell NO-MATRIX-ROW · gate NO-MATRIX-ROW
| Contract | os_served_delta · executor Organ · 2 u · stands on nothing |
|---|---|
| Definition of done | OPERATOR 2026-09-04: the delivery pipeline must be self-adjudicating, a human should not be in this loop, and the math was clear as day from the first failed ship. MEASURED, and it is a three-instance class not a one-off. INSTANCE ONE: nx_organ_ship nx_game_page_emit ran the whole loop GREEN -- build, adoptgate with four lost runs adjudicated, promote 425241 to 451160 B, offc twin, register, verify -- and changed NOTHING a visitor can see, because the target is a LIBRARY whose standalone binary emits no served artifact. Everything needed to say so was one command away: grep for the import edge returns 3 importers with corpus_complete=1, nx_gpe_hair_geometry_gate nx_world_recipe_gate and nx_craft_emit, and only the last declares served outputs CE_OUT CE_OUT2 CE_OUT3. The loop never asked which served artifact it changes. INSTANCE TWO: the same loop reports verify against the BINARY, so a fresh page emitted by a 15.6 hour old writer reads as a successful delivery. INSTANCE THREE: /api/promote already reads organ_kind.conf and refuses daemons BY NAME, so kind data is on the path and a lib promote is permitted and SILENT. DONE-RULE, two teeth and both cheap: FIRST the loop resolves the target's served artifacts from the organ's own output declarations and, when the set is EMPTY, it NAMES every importer and REFUSES to report success, because shipping a lib alone is inert BY CONSTRUCTION rather than by bad luck; SECOND it hashes each served artifact before and after and a ship whose served bytes did not move is reported FAILED however green every stage was. TWO NEG-CONTROLS: a genuine writer ship whose page bytes DO move must still pass, and a lib ship must fail even when build prove adoptgate promote and register are all green. NOT DONE if it merely warns -- the failure mode being fixed is a loop that reported success, so the verdict must change, not the prose. |
| Matrix cell | NO-MATRIX-ROW — no matrix row carries os_served_delta; the ranker reads this rung as unmapped and no cell can flip for it. Add the row and the join completes on the next publish. |
| Gate | the gate join needs a matrix row: no organ is named for this rung |
| Work log | 2026-09-04 retract MY OWN DONE-RULE COMMITTED THE ERROR IT WAS WRITTEN TO STOP. GR50 as landed centres on a SERVED BYTE DELTA, and the operator named it immediately: believing byte changes equals functionality shipping. The direction I wrote is sound -- bytes did not move therefore FAILED -- but building the rung around the delta invites its converse, and the converse is exactly the fallacy I committed this morning when I read 425241 to 451160 B as delivery of a fix that was never in the emitted page. A BYTE DELTA IS A NECESSARY PRECONDITION AND NEVER A PROOF. CORRECTED DONE-RULE for os_served_delta: the loop resolves the target's served artifacts from the organ's own output declarations; an EMPTY set means the target is inert by construction, so it NAMES every importer and reports FAILED; and where the set is non-empty the ship passes only on a BEHAVIOURAL assertion about the served artifact that the pre-ship artifact demonstrably fails -- the artifact's own gate re-run against the new bytes, not the size of the change. The byte delta stays as a CHEAP PRECONDITION that can only ever refuse, never as the thing that grants. |
GR51 The hair gate gains a tooth the broken output must fail: rendered strands inside the body are counted, not assumed · cell NO-MATRIX-ROW · gate NO-MATRIX-ROW
| Contract | gpe_hair_penetration_tooth · executor Organ · 1.5 u · stands on GR48 |
|---|---|
| Definition of done | MEASURED 2026-09-04 on the LIVE gate: nx_gpe_hair_geometry_gate reads passed 16 of 16 verdict=GREEN while the operator is looking at hair clipping through the chest, and BOTH statements are true because they are about different subjects. Every tooth it carries asks whether the pipeline is WIRED -- emitted-consumes-HVRT/HTRI-groom-MESH, emitted-consumes-HSTR-guide-strand-geometry, emitted-indexes-the-chain-table-BY-VERTEX-SEGMENT, emitted-UPLOADS-the-displaced-geometry-each-frame, emitted-ANNOUNCES-absent-hair-geometry-rather-than-reporting-success, and a neg-control that the superseded RIGID block is gone. NOT ONE tooth asks whether a rendered hair vertex is inside the body, so no possible penetration can fail this gate and its GREEN is evidence of a connected pipeline, never of correct hair. THIS IS THE SAME CLASS AS THE OTHER TWO FINDINGS TODAY: hg_tick is arithmetically correct and has NO CALLER across 23622 files with corpus_complete=1, and the ship loop verifies a binary rather than a served artifact -- three green signals, three real defects, and in every case the subject measured was narrower than the claim made. DONE-RULE: a tooth counts rendered strand vertices lying inside the body capsules of the same fixture the gate already builds, the count is EMITTED with gv_kv beside its bar so a second method can contradict it, the tooth is BITE-PROVEN by mutating the clamp away and seeing the count RISE and the gate go RED, and an anti-vacuity control asserts the fixture actually reached a pose where penetration is possible -- a fixture whose hair never approaches the body would pass this tooth for the wrong reason and is not a test. |
| Matrix cell | NO-MATRIX-ROW — no matrix row carries gpe_hair_penetration_tooth; the ranker reads this rung as unmapped and no cell can flip for it. Add the row and the join completes on the next publish. |
| Gate | the gate join needs a matrix row: no organ is named for this rung |
| Work log | no log row names this rung yet |
GR52 The estate's searcher can read the artifacts the estate serves · cell NO-MATRIX-ROW · gate NO-MATRIX-ROW
| Contract | st_longline_windows · executor Organ · 2 u · stands on nothing |
|---|---|
| Definition of done | OPERATOR 2026-09-04: invest in our tooling and do this properly rather than routing around broken instruments. MEASURED, FOUR DEFECTS HIT IN ONE SESSION AND ALL FOUR WORKED AROUND RATHER THAN FIXED: nx_shelltool grep declares linetoolong and SKIPS any line longer than its buffer, and the estate's primary shipped artifact -- an emitted world page -- is a handful of 200 KB lines, so the searcher is STRUCTURALLY BLIND TO EXACTLY WHAT WE SERVE and every absence claim over a served page is unprovable by it; the same verb takes a DIRECTORY and refuses a file path, which reads as the organ being missing; a directory scan reports files=200000, a ROUND DECIMAL that is a CAP printed in a field that reads as a count, and it was quoted as a population fact in this very session; and BUDGET-EXCEEDED returns corpus_complete=0 with matches=0, which is UNPROVEN and reads as absence. THE INCUMBENT IS WELL REASONED AND MUST NOT BE REPLACED: nx_shelltool.nx:200-210 explains that it reads in chunks and carries the trailing partial line so coverage is complete at any file size, and that it refuses to split a long line because half a line matched as a whole one is a FABRICATED match, the opposite failure and the worse one. That reasoning is right and the fix preserves it. DONE-RULE: at nx_shelltool.nx:250 the overlong branch stops discarding to end-of-line and instead scans the same line in OVERLAPPING WINDOWS carrying pn-1 bytes across each cut, pn-1 being exactly the largest pattern prefix that can straddle a boundary, so no match is missed and none is fabricated; a hit on an overlong line emits a WINDOW around the match rather than the whole line; linetoolong is retired as a coverage caveat and replaced by a longline= count that is INFORMATIONAL rather than a hole; files= and any other cap is renamed or accompanied so a ceiling can never be read as a census; and the whole thing is BITE-PROVEN by planting a pattern in the middle of a 200 KB line that the current binary MISSES and the fixed one FINDS, with a neg-control proving a pattern that straddles a window boundary is reported exactly ONCE. |
| Matrix cell | NO-MATRIX-ROW — no matrix row carries st_longline_windows; the ranker reads this rung as unmapped and no cell can flip for it. Add the row and the join completes on the next publish. |
| Gate | the gate join needs a matrix row: no organ is named for this rung |
| Work log | no log row names this rung yet |
GR53 The estate reads its own served pages with its own wire: a sovereign CDP witness composed from the shipped WebSocket stack · cell NO-MATRIX-ROW · gate NO-MATRIX-ROW
| Contract | nx_cdp_eval · executor Organ · 1.5 u · stands on GR49 |
|---|---|
| Definition of done | OPERATOR 2026-09-04, asked of the Python browser driver this seat leaned on all day: is it a tool we need to duplicate into nishi lang and use as a 3rd witness. MEASURED ANSWER, by listing buildroot/runtime rather than estimating: IT IS COMPOSITION, NOT CONSTRUCTION. Already shipped and already tested here are nx_websocket_client_upgrade.nx, which performs the RFC 6455 client handshake on an already-connected TCP fd and validates the 101 status plus Upgrade, Connection and Sec-WebSocket-Accept against the SHA-1 and Base64 transform of the key it sent, with nx_websocket_client_upgrade_test.nx beside it and ten named verdict constants; and nx_websocket_frame.nx, giving ws_parse_frame, ws_build_header and ws_apply_mask with the full opcode set, with nx_websocket_frame_test.nx beside it. Underneath sit nx_json.nx, nx_json_emit.nx, nx_base64.nx, nx_rand.nx, nx_net.nx and nx_syscalls.nx. ac_cdp.py's own header calls the handshake and frame codec THE ROUGHLY FORTY LINES BELOW and treats them as the hard part; that part is done, in NishiLang, with tests. THE ONLY MISSING GLUE, named so nobody rebuilds what exists: connect TCP to the DevTools port, GET /json/version or /json/list to resolve the page's webSocketDebuggerUrl, run the client upgrade on that fd, send one Runtime.evaluate with returnByValue true, then read frames until the matching id and hand back the result verbatim. CDP to a local browser is plain ws, so no TLS is required and nothing else is needed underneath. WHY THIS IS WORTH A RUNG RATHER THAN A NOTE, measured today and it is the uncomfortable half: nx_shelltool grep SKIPS any line longer than its buffer and every emitted world page is a handful of 200 KB lines, so THE ESTATE'S OWN SEARCHER IS STRUCTURALLY BLIND TO THE ARTIFACTS THE ESTATE SERVES, and the ONLY instrument that could answer whether our own code reached our own page was a third-party browser driven by Python. Every false claim this seat published about that question traces to exactly that blindness. THE BUILD PATH IS SOVEREIGN AND THE EVIDENCE PATH IS NOT, and nobody was counting the second half. DONE-RULE: nx_cdp_eval opens a page, evaluates one expression and prints the result verbatim with zero judgement of its own, exactly as the Python wire does; a neg-control proves it REFUSES rather than returns empty when the endpoint is absent, when the upgrade is rejected, and when the evaluate returns an exception; and the acceptance test is that it reproduces, byte for byte, the answer ac_cdp.py returns for the same expression on the same served URL -- two witnesses agreeing is the point, so the Python one is KEPT as the oracle and never retired. SCOPED AWAY from nx_browser: rendering the page with our own engine is the THIRD witness and the zero-translation rung of the operator's ladder, a separate and larger piece; this rung is our WIRE against their ENGINE and is buildable on the LAPTOP toolchain, which matters because a brand-new file has no laptop-versus-NAS fork to converge and NAS build admission refused every lane for hours on the day this was scoped. |
| Matrix cell | NO-MATRIX-ROW — no matrix row carries nx_cdp_eval; the ranker reads this rung as unmapped and no cell can flip for it. Add the row and the join completes on the next publish. |
| Gate | the gate join needs a matrix row: no organ is named for this rung |
| Work log | 2026-09-04 land nx_cdp_eval LAPTOP-ONLY 2026-09-04: sovereign CDP witness (buildroot/runtime/nx_cdp_eval.nx 6798 B; built _build/nx_cdp_eval.sov.elf 40253 B sha bfe1756b), usage nx_cdp_eval <port> <ws-path> <expression>; imports nx_syscalls, nx_websocket_client_upgrade, nx_websocket_frame; hand-built sockaddr_in to loopback; the fixed mask key CE_MASK is named as a loopback-only choice. PROVEN: three refusal paths plus a full masked round trip against a WebSocket fixture (FIXTURE-GOT masked=1 len=107, reply printed verbatim, exit 0). NOT promoted, NOT on the NAS, not yet driving a real browser tab -- that is the remaining leg, and it retires the Python ac_cdp.py wire. MEASURED on the way: nx_net.nx does NOT compile standalone on the laptop toolchain (8 errors, Result of i64 and NetError unresolved while NetError is an enum at line 70), so it was dropped from the closure rather than worked around -- a lang-domain defect to adjudicate. |
GR54 ONE HAIR SYSTEM: five parallel implementations collapse to one, and this rung is measured by DELETIONS not additions · cell NO-MATRIX-ROW · gate NO-MATRIX-ROW
| Contract | hair_one_system · executor Organ · 4 u · stands on GR48 |
|---|---|
| Definition of done | OPERATOR 2026-09-04, and the evidence for it is THIS SEAT'S OWN WORK: this feels like you are just dumping shit everywhere because you have no plan and we have multiple systems in play so sometimes one system gets the work sometimes another and thats garbage and not sota or any level of good development. HE IS RIGHT AND THE COUNT IS FIVE, ALL TOUCHED IN ONE SESSION BY ONE SEAT: a soft-tissue term added to the collision proxy in nx_nxa_rig_emit which is the RIGVIEW shader path; a child-offset clamp added to nx_game_page_emit which is emitted JAVASCRIPT on the BEACH path; vm_capsule_push added to nx_vecmath which is a FIFTH copy of closest-point-on-capsule; hd_tick_collide added to nx_hairdyn_lib which is the BAKE-time integrator; and nx_haircollide_gate built 12 of 12 GREEN as a gate for hg_tick, WHICH HAS NO CALLER ACROSS 23622 FILES WITH corpus_complete=1. FIVE ADDITIONS TO FIVE SYSTEMS IN ONE DAY, INCLUDING A GATE FOR DEAD CODE AND A FIFTH DUPLICATE PRIMITIVE, while the same seat catalogued duplicate-system defects as though observing from outside. ★★★★★★A SEAT THAT FIXES WHICHEVER SYSTEM IT HAPPENED TO BE READING PRODUCES MOTION IN EVERY SYSTEM AND PROGRESS IN NONE, AND EVERY SUCH FIX MAKES THE NEXT CONSOLIDATION MORE EXPENSIVE. THE RUNTIME TRUTH, MEASURED: the hair a visitor sees is built at load in hand-written JS from the rest-pose asset section and moved per frame by nxHairTick -- spring, viscous damping, wind, length clamp, and NO BODY TERM OF ANY KIND. The wasm collider hg_capd2 hg_inside hg_push with HG_CAP_MAX=32 capsules and HG_SIM=128 guides is UNREACHABLE. So four of the five systems cannot affect a served pixel and the fifth is outside the compiler. DONE-RULE, AND IT IS COUNTED IN DELETIONS: closest-point-on-capsule exists exactly ONCE in the estate and the other four sites are retired through nx_retire_path with the survivor named here; hg_tick has at least one production caller and a bite proves the gate goes RED when collision is disabled, so the collider stops being a comment; the rendered strand set is built by NishiLang lowered to wasm beside that collider with the emitted-JS geometry RETIRED once the wasm path reproduces its vertex buffer within a stated band on a fixture groom; and nx_haircollide_gate's subject changes from a dead function to the shipping path or the gate is retired with it. THE ACCEPTANCE TEST IS A COUNT THAT FALLS: capsule implementations 5 to 1, hair integrators 2 to 1, geometry builders 2 to 1, and hg_tick callers 0 to at least 1. A rung that ADDS a sixth anything has failed by construction. SUPERSEDES the scattered pieces this seat landed today -- GR25's clamp becomes a property of the one system rather than a patch on the JS one, and GR48 is the geometry half of this rung rather than a separate lane. |
| Matrix cell | NO-MATRIX-ROW — no matrix row carries hair_one_system; the ranker reads this rung as unmapped and no cell can flip for it. Add the row and the join completes on the next publish. |
| Gate | the gate join needs a matrix row: no organ is named for this rung |
| Work log | no log row names this rung yet |
GR55 Skin-tone fidelity of the LIVE render is measured in Delta E and ITA error per class under named lights · cell WATCHING · gate NO-GATE-DECLARED
| Contract | skin_tone_fidelity_dE · executor Organ · 1.5 u · stands on nothing |
|---|---|
| Definition of done | SEPT-2026 BAR, read from the mirrored paper: 19848 MetaHuman renders scored against Chicago Face Database references show fidelity collapsing for darker classes under non-reference lighting (class VI median ITA error 11.37 to 53.85) [truetotone26]. Our generator spans the ITA classes by construction (nx_skin_ita, gate T4), so the gap is MEASUREMENT of the rendered result, not generation. DONE-RULE: the ITA ladder is rendered through the live beach path under three named lights, each render is scored in Delta E CIE76 and ITA error against its reference, per-class medians are EMITTED with gv_kv beside the published medians, and a neg-control proves a deliberately desaturated render RAISES the error -- an anti-vacuity tooth asserts the fixture spans at least four ITA classes so a single-tone ladder cannot pass |
| Matrix cell | SKIN: skin-tone fidelity as CIELAB error and ITA-class preservation · runtime/nx_skin_ita.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_skin_ita; declare nx_skin_ita TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR56 Our subsurface term is calibrated against a measured face BSSRDF and its RMSE is published · cell WATCHING · gate NO-GATE-DECLARED
| Contract | skin_bssrdf_calibration · executor Organ · 2 u · stands on GR55 |
|---|---|
| Definition of done | SEPT-2026 ORACLE: a polarimetric multispectral capture of dynamic faces fitting spatially varying single scattering and refractive index, validated at a mean reconstruction error of 0.028 against known-index objects [polbssrdf25]. The oracle NEVER enters the build path: one measured BSSRDF is adopted as DATA with its license beside it, our skin model is fitted to it, and the RMSE of our fit against the measurement is the row's number. DONE-RULE: the fit RMSE is emitted with gv_kv, a neg-control proves a flat-albedo control fits WORSE by a named margin, and the fitted parameters are the ones the live path reads (a fit that never reaches the page is a comment) |
| Matrix cell | SKIN: measured ground-truth BSSRDF as the calibration oracle · runtime/nx_skin_pbr_gate.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_skin_pbr_gate; declare nx_skin_pbr_gate TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR57 Sheer and stretched fabric over skin · cell WATCHING · gate UNRECORDED
| Contract | _ABSENT_:gr_sheer_bsdf · executor Organ · 4 u · stands on GR42, GR14, skintwin:ST5 (UNRESOLVED: no such rung on this plan), procgen:PG29 (UNRESOLVED: no such rung on this plan) |
|---|---|
| Definition of done | The reference-still class this rung answers (2026-09-27) wears a sheer swimsuit: the garment is a THIN LAYERED MEDIUM, not an opaque surface. Reflectance core is skintwin ST5 km_layer_over (Kubelka-Munk two-flux layering of fabric over skin); transmission RISES with weave strain read from the drape solver (procgen PG29 gm_xpbd) and with wetness (GR58); the skin term is visible through the fabric in the beauty tier and the live tier. The field's bar this month: Eurographics 2026's texture-free woven-fabric model and neural garment fold inter-reflection, the 2024 real-time neural woven fabric and reflection-transmission capture lineage; nothing this month renders sheer fabric OVER skin with strain-driven transmission, which is the exceed axis. Done-rule: a rendered sheer swatch over a skin patch matches the two-flux prediction within a DECLARED DE2000 band at three strains, and the beach garment shows the skin term through it under the sky. [kesen_eg2026] |
| Matrix cell | FABRIC: sheer and stretched fabric over skin renders as a thin layered medium · runtime/nx_gi_path.nx · WATCHING decl |
| Gate | nx_gi_path_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | 2026-09-27 queue Landed 2026-09-27 from the beach-lounger reference-still analysis. Estate incumbents found by nx_capsearch: nx_gm_cloth_lib (PG29 XPBD), nx_garment_gen (garment shells), nx_kubelka_lib (KM core), nx_skinfit_lib; no sheer-over-skin BSDF anywhere in 2181 tools and 6011 libs (the ranker's corpus_complete=0, so a floor). The skintwin ranker reads ST5 km_layer_over as DARK-CONTRACT: the organ exists, the symbol is undeclared, so the layered core itself is unbuilt. |
GR58 Wet as a surface state on skin, hair and sand · cell WATCHING · gate UNRECORDED
| Contract | _ABSENT_:gr_wet_state · executor Organ · 3 u · stands on skintwin:ST7 (UNRESOLVED: no such rung on this plan), skintwin:ST9 (UNRESOLVED: no such rung on this plan), GR27, GR16 |
|---|---|
| Definition of done | One wetness FIELD drives three materials: skin darkens and its specular sharpens (skintwin ST7 fo_wetness), hair clumps by capillary adhesion and its cuticle reads index-matched (skintwin ST9), sand darkens by the wet-porous term (GR27's wet-shore darkening generalised). The estate's wet-appearance scan read 2026-09-27 (11 entries, newest 2025-12, a homonym) holds no 2026 wet-appearance paper; the standing references are the micro-ellipsoid wet-porous model (2401.15628) and Jensen's 1999 wet-materials model, so this axis has NO current-month bar and says so. Done-rule: dry-to-wet DE2000 and specular-lobe shift published per material against mirrored measurements, and the field is one row of data the beach scene reads. |
| Matrix cell | WET: one wetness field darkens skin, clumps hair and darkens sand · runtime/nx_gi_path.nx · WATCHING decl |
| Gate | nx_gi_path_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | no log row names this rung yet |
GR59 Sun-tan and tan lines as a melanin field · cell LANDED · gate NO-GATE-DECLARED
| Contract | _ABSENT_:gr_melanin_field · executor Organ · 2 u · stands on GR55, skintwin:ST6 (UNRESOLVED: no such rung on this plan), skintwin:ST1 (UNRESOLVED: no such rung on this plan) |
|---|---|
| Definition of done | The skintwin board's own goal names a tan line as an ABSORBER change, never a pixel edit. Melanin concentration is a per-vertex field DERIVED from garment coverage history and declared sun exposure (data rows), read by the skintwin chain (sf_fit, km_r_from_ks) and relit correctly by construction; the ISO/DTR 26478 skin-tone patches and NIST's 100 traceable skin reflectance spectra (250-2500 nm) are the calibration oracles skintwin ST6 pins. Done-rule: a tan line follows a garment edge, its inside-outside ITA difference lands in the band the skintwin referee (ST1) reports, and a relight moves both sides consistently. |
| Matrix cell | SKIN: sun-tan and tan lines are a derived melanin field, never paint · runtime/nx_skin_ita.nx · LANDED decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_skin_ita; declare nx_skin_ita TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-10-01 land SUN-TAN AND TAN LINES AS A DERIVED MELANIN FIELD BUILT AND PROMOTED 2026-09-30. gr_melanin_field landed in runtime/nx_skin_ita.nx (11339 -> 19116 B, additive, pure integer, no new imports): per-vertex melanin = constitutive + facultative, the facultative part computed from each vertex's OWN garment-coverage history and the declared exposure, both data rows in knowledge/skin_tan.conf. The absorber chain is the new runtime/nx_skin_tan_lib.nx (21525 B): melanin -> epidermal absorption -> a finite Kubelka epidermis over a semi-infinite dermis through the estate's km_layer_rt, km_layer_over and km_r_from_ks -> CIE XYZ on the CIE's own 1931 2-degree CMF, D65 and A tables at the 5 nm abridgement CIE 15 accepts over 380-780 nm -> cs_lab_from_xyz -> ITA, and a relight that re-integrates the same reflectance under A and adapts it to D65 through cs_cat_build (Bradford). The constitutive melanin is SOLVED from the genome's ITA on this organ's own locus, never typed. Dose-response and persistence from Ravnbak 2008 [ravnbak_tankinetics_2008] (MMD per Fitzpatrick type; the pigmenting dose fraction lowered 2x and 3x by 6 and 12 exposures; Table II steady state; peak a week after the last exposure; a linear fade over the published days to total fading); absorber and scatterer spectra from Jacques 1998 [jacques_skinoptics_1998]; the band from Amano 2020 [amano_naturaltan_2020]; colorimetry on [cie_cmf1931_2deg] [cie_illum_d65] [cie_illum_a]. nx_skin_ita_gate 9 -> 25 teeth (all 9 incumbents still GREEN), built sovereign 127720 B sha256 5dfb04e0b8d0c00a68fa9d50fd8b57f1d693537fe6abe9d3e78d7ede23912253, staged run 25/25 GREEN, contentdiff lost exactly 1 run (the gate's own rewritten verdict note, named), promoted under allow_capability_loss with prev banked (pre-change live 35d5fdee 28666 B), _offc refreshed, LIVE run 25/25 GREEN (nx_job_run pid 32241). The first render-side adoption of the skintwin ST1 referee cs_de2000 2026-10-01 measure THE ESTATE'S NUMBERS BESIDE THE SOURCES' (live nx_skin_ita_gate 5dfb04e0, 25/25 GREEN, values by gv_kv, one declared genome and one declared scenario, not a population): genome hue 790 asks ITA 49.863 deg and the solved constitutive melanin is 55404 ppm, 5.54 percent, INSIDE the 1.3-6.3 percent Jacques gives for light-skinned adults [jacques_skinoptics_1998], a range the solver was never shown. A three-week beach holiday at the density of Ravnbak's 12-exposure protocol (12 sessions of 3.2 SED, about 0.6 MMD for type II, a UPF 30 garment whose edge wanders 296-304 mm) read at its peak, day 26: the line falls at 301 mm, always-covered skin keeps 0 ppm facultative, always-exposed skin gains 21774 ppm (Table II 9.0 over the type-II baseline 22.9 [ravnbak_tankinetics_2008]). Inside Lab 72.75 3.01 19.18 (ITA 49.86), outside 66.83 4.83 21.22 (ITA 38.42) against the measured outer-forearm mean ITA 38.8 [amano_naturaltan_2020]; DE76 6.51 against the measured natural tan 6.03 SD 2.64 (band 3.39-8.67); DE2000 4.99 by cs_de2000; dITA 11.45 deg inside the band 4.60-13.59 that the same measurements imply at this subject's protected colour; relit under illuminant A and adapted to D65 by Bradford, dITA 11.27, a shift of 0.18 deg against a bound of 4.49; the line's optical-density log-log exponent 2.01 against melanin's 3.33 and a neutral paint's 0.00 (nearest class decides, the paint is told apart); the D65 white summed from the mirrored tables Xn 95.042 Zn 108.880 against 95.047 108.883 [cie_cmf1931_2deg] [cie_illum_d65]; persistence: 21774 ppm at day 26, exactly half (10887) at day 110, 0 at day 194, 175 days after the last session as the paper measured 2026-10-01 measure TRIANGULATED BAR as of 2026-10-01 for facultative pigmentation (UV dose to melanogenesis and tan persistence, ITA of sun-exposed versus protected skin): HOW A TAN IS SCORED -- as a difference from a constitutive baseline site: degree-of-tan is the ITA of buttock minus the ITA of the exposed site, measured every 6 to 7 weeks for a year with personal UV dosimeters [schmalwieser_tan_2025], or outer minus inner forearm, whose natural-tan difference averaged Delta E 6.03 (SD 2.64) [amano_naturaltan_2020]. DOSE TO TAN -- under controlled sub-MMD UVB the absolute pigment increase is linear in dose and independent of skin type, repeated exposure lowers the dose to minimal pigmentation by factors of 2 and 3, the protocol takes the peak at one week after the last exposure, and fading takes 4 to 5 months (132 to 175 days to total fading by linear regression on 1 week, 2 and 4 months) [ravnbak_tankinetics_2008]; in outdoor farmers the change in tan regresses linearly on daily mean erythemal dose with the zero crossing at 1.4 SED per day (above it the tan grows, below it fades) and an obvious tan persists through late winter [schmalwieser_tan_2025]; the newest on-subject measurement (August 2026, L'Oreal) adds depth: in vivo multiphoton microscopy measured melanin density and its epidermal depth distribution, and melanin density and skin colour rose from winter to summer under moderate real-life exposure but not under daily SPF and PPD protection [epmc_tanning_202609]. THE COLOUR PATH -- a natural tan lowers lightness without raising yellowness, following Chardon's curved (L*, b*) skin-colour volume with peak yellowness near L* 50, while a straight L*-b* shift reads as unnatural [amano_naturaltan_2020]. THE INSTRUMENT -- ITA is weak outside controlled colorimetry: fixed-threshold ITA from uncontrolled images scored 22.9 percent against a 51.3 percent majority baseline [alipour_ita_limits_2026], and ITA was built for the suntanning pathways of Caucasian skin, so apparent-colour work reports L* and hue angle [beyondskintone_iccv2023]. LISTINGS -- the arXiv skin-tanning search returns 3 hits, none on this axis (one is leather tanning) [arxiv_skintanning_202609], the arXiv cs.GR October 2026 listing reads no updates [csgr202610_gfx_t1001], and the newest on-subject biomedical month is August 2026 [epmc_tanning_202609] ; witnesses [schmalwieser_tan_2025] [ravnbak_tankinetics_2008] [amano_naturaltan_2020] [epmc_tanning_202609] [alipour_ita_limits_2026] [beyondskintone_iccv2023] [arxiv_skintanning_202609] [csgr202610_gfx_t1001] -- independent groups: the Copenhagen Bispebjerg photobiology group with Vienna (Schmalwieser and Ravnbak share Philipsen and count as ONE), Liverpool and Leeds colour science (Amano, Xiao, Wuerger, Meyer), L'Oreal Advanced Research, Technological University Dublin, Sony AI with the University of Tokyo -- five, all in peer-reviewed venues ; WHERE THEY DISAGREE: the controlled study's abstract says fading was unrelated to ethnic origin, yet its own subgroup medians fade faster in Scandinavians (101 and 104 days) than in Indian and Pakistani volunteers (152 and 132 days); the field study finds the 1.4 SED per day crossing only in farmers, with no dose correlation at any site in spouses and children, and finds the supposedly unexposed inner arm also tans and fades; the instruments differ (a reflectance pigmentation value, tristimulus ITA, multiphoton melanin density) and the imaging papers reject ITA from photographs; the defining erythema standard (ISO/CIE 17166) was mirrored but did not extract, so the SED unit is witnessed through the papers that use it, not through the standard ; VERDICT on the running lane (corrected after reading its land row 1790824660, which preceded this row): GR59 landed 2026-09-30 as gr_melanin_field over nx_skin_tan_lib and is AT the triangulated best on controlled tanning kinetics and baseline -- Ravnbak's per-type minimal melanogenic dose in SED, its 2x and 3x lowering by repeated exposure, a peak one week after the last exposure and a linear 2026-10-01 measure THE NEW TEETH BITE THEIR OWN SUBJECT, PROVEN BY AN AIMED MUTANT (2026-09-30): nx_gate_bite over nx_skin_ita.nx (aim gr_melanin_field) killed its first compiling comparison mutant at byte 6805, which sits in the pre-existing locus code, so it proves the gate non-vacuous and NOT the new block (the aim limit already filed as debt 1790543778); the aimed proof was taken by hand instead: gr_melanin_field's dose line mutated to ignore the coverage history (the garment blocks nothing), built sovereign, run staged: RED 18/25, killed by T12, T12b, T13, T13b, T14, T15 and T16 while the zero-dose, open-fabric and persistence controls stayed GREEN; the source restored byte-identical (sha256 38164447, 19116 B) within the same minute, the scratch rows reaped, and the bite harness's twin custody reported restored=2 mutant_live=0 with live still 5dfb04e0. A rebuild against the CURRENT closure (nx_colorsci_lib 38414 -> 46771 and nx_kubelka_lib 18047 -> 29088 moved under two sibling lanes during this leg) built 28970df2 at the same 127720 B and ran 25/25 GREEN with output byte-identical to the promoted binary's (sha256 92db0439, 4048 B): the sibling changes are behaviour-neutral for GR59 2026-10-01 land NO SECOND SPECTRA READER: skintwin ST6 landed km_spectra_load in nx_kubelka_lib DURING this leg (the estate's one pinned-spectra reader: the bytes proven against the refs pin by sha256, a truncated, reordered or unpinned table refused by name), so the CSV parser written beside it this afternoon (tn_isdig, tn_dec_q30, tn_csv_load) was retired the same hour and nx_skin_tan_lib now reads the CIE CMF, D65 and A tables through km_spectra_load and km_spec_interp under the pins of [cie_cmf1931_2deg] [cie_illum_d65] [cie_illum_a] (21525 -> 20399 B). The gate gained T10c, a biting control: illuminant A offered under the D65 pin refuses KM_SPEC_PIN (-3) by name. Rebuilt sovereign 154232 B sha256 4fd921dcd406757ff4ae0da1ea588693db6b8b5d99600fb671652ba8bd9e3ac5, staged 26/26 GREEN with EVERY published value byte-identical to the first build (the parse changed, the numbers did not), contentdiff lost exactly the four retired symbol names (fq_to_int no longer referenced, tn_isdig, tn_dec_q30, tn_csv_load) and gained T10c, promoted under allow_capability_loss with every lost run named, _offc refreshed, LIVE run 26/26 GREEN (nx_job_run pid 13097). NAMED COST: two promotes in one leg rotated the original pre-change binary (35d5fdee, 28666 B, 9 teeth) out of the prev slot, banked by hash only; prev now holds 5dfb04e0, the first GR59 promote, whose 25 teeth include all nine of the original's 2026-10-01 measure NO CURRENT-MONTH BAR EXISTS FOR THIS AXIS, AND THE READING SAYS SO (2026-09-30): the arXiv abstract search for skin tanning, newest first [arxiv_skintanning_202609], holds three entries and all three are OFF-SUBJECT by subject (a skin-lesion classifier of January 2026, a non-Hermitian skin-effect paper, leather-tanning defect segmentation); the export API answered HTTP 429 eight times the same hour, so the melanin-AND-rendering query is UNATTESTED rather than empty. Nothing the field published renders sun-tanning as a dose-driven absorber field. The bars this rung is held to are therefore measurements, not rival methods: Ravnbak 2008's photobiology [ravnbak_tankinetics_2008] for dose, threshold and persistence, Amano 2020's 60-subject natural-tan colorimetry [amano_naturaltan_2020] for the band, and the vendor bar this board already carries for painted skin (MetaHuman 5.8 SB-MHSKIN-2609, palm and sole tone applied automatically, no exposure history). Against that vendor bar the estate's line is DERIVED, relit by re-integration, and persistent by a measured fade; parity on absolute skin colour is NOT claimed while the dermis is bloodless 2026-10-01 lesson TWO INDEPENDENT MEASUREMENTS AGREED THROUGH ONE MODEL WITH NO KNOB BETWEEN THEM, BECAUSE THE TAN WAS MADE RELATIVE: the facultative melanin is the constitutive melanin scaled by Ravnbak's increase over pre-exposure pigmentation, and the constitutive melanin is SOLVED against the genome's ITA, so the two-flux convention (K = 2 mua, S = musp, against the diffusion-matched 3/4) cancels out of the line instead of being tuned into it. What fell out: the solved constitutive melanin, 5.54 percent, inside Jacques's published 1.3-6.3 percent for light skin, and the tanned side's ITA 38.42 against Amano's independently measured outer-forearm mean 38.8. A SECOND LESSON, PAID FOR TWICE THE SAME DAY: a sibling lane mirrored the SAME Jacques page and CIE tables under its own keys while this one did, and a skintwin.refs append created a duplicate key that only a find caught (removed within minutes, the sibling's keys cited instead); and the reader this lane wrote landed beside the one that lane was writing. Before appending a ref or writing a parser, search the board's refs by PIN and the libs by CAPABILITY, not by the name you were about to use 2026-10-01 queue NEXT STEPS, EACH WITH ITS RECIPE: (1) HAEMOGLOBIN IN THE DERMIS, the named fidelity gap: the inside skin reads a* 3.0 where measured skin reads near 10, and the line raises b* by 2.04 where Amano's natural tans raise none [amano_naturaltan_2020]; add mua.blood = ln10 x epsilon x 150 g/L / 64500 g/mol (the OMLC conversion) at Jacques's f.blood 0.2 percent [jacques_skinoptics_1998] to tn_reflect's dermis, read through km_spectra_load from a mirrored, pinned Prahl TABLE (the omlc_hb row mirrors only the summary page) with an oxygen saturation taken from a pinned source, then re-measure a*, b*, DE2000 and the band teeth -- the constitutive melanin is re-solved against the genome automatically. (2) THE LIVE FRAME: the field is gated, the beach frame does not read it yet; feed gr_melanin_field from the garment lane's per-vertex coverage and bake tn_reflect through the beauty tier, so GR55's per-class ITA error is measured on a tanned body. (3) POPULATION: one genome and one scenario are measured; sweep the six ITA classes (Ravnbak's absolute increase is skin-type independent, so the relative facultative fraction must fall as constitutive melanin rises) and publish dITA per class beside GR55's medians. (4) re-attest the arXiv melanin-AND-rendering listing once the export API stops answering 429. (5) the watch cell _ABSENT_:gr_melanin_field should read LANDED after the orchestrator's next regen of /compare/graphics; check the cell, not this row 2026-10-01 measure THE CHAIN LIB IS BITE-PROVEN AND THE SCRATCH IS DISPOSED (2026-09-30): nx_gate_bite nx_skin_ita_gate over nx_skin_tan_lib.nx killed its first compiling comparison mutant at byte 5800 (inside the new chain lib), valid_mutants=1 killed=1 not_reached=0, verdict GREEN, the lib restored byte-identical (20399 B, sha256 3b35fb18) and both twins restored (custody twins_checked=2 restored=2 mutant_live=0, live still 4fd921dc). Final sources: runtime/nx_skin_ita.nx 19116 B sha256 38164447 (the gr_melanin_field contract), runtime/nx_skin_tan_lib.nx 20399 B sha256 3b35fb18, runtime/nx_skin_ita_gate.nx 21461 B sha256 658b8d20, knowledge/skin_tan.conf 3529 B. The four scratch binaries (two staged candidates, the current-closure rebuild and the coverage-blind mutant) were retired reversibly into knowledge/retired with nx_retire_path, and every scratch registry row (nx_skin_ita_gate_gr59, gr59m, gr59c, gr59p _staged) was reaped; the first GR59 promote 5dfb04e0 survives both as prev and as knowledge/retired/nx_skin_ita_gate_gr59.sov.elf.new-1790826503 2026-09-30 measure ADVERSARIAL VERIFIER, method class 3 (independent hand re-derivation, not the gate's arithmetic) over a fresh run of the promoted nx_skin_ita_gate 26/26 GREEN on the NAS (job /tmp/verify_gr59_ita.out). ITA=atan((L-50)/b) by hand from the printed Lab: covered L 72.748 b 19.180 -> 49.864 deg vs printed 49.863; tanned L 66.828 b 21.218 -> 38.418 vs 38.418; dITA 11.446 vs 11.445 (inside the 4.599..13.585 band); the 1 mdeg gap is inside the milli quantisation of the printed Lab (about 1.2 mdeg). CIE76 from the same Lab 6.519 vs printed 651 centi (the gate truncates, 603 plus or minus 264 band 339..867 holds). Fade from skin_tan.conf: last exposure day 19 + Ravnbak peak 7 = 26 vs 26; 19 + Ravnbak 175 = 194 vs 194; midpoint (26+194)/2 = 110 vs 110 at 21774/2 = 10887 ppm vs 10887 (the half-at-midpoint result holds only because the decay between peak and zero is linear, a model choice, Ravnbak publishes only the total fading time). Bare melanin: Jacques 6.6e11 nm^-3.33 at 400 and 700 nm = 1427.9 and 221.5 cm-1, ln ratio/ln(700/400) = 3.330 vs printed 3.329 (the printed value is the conf exponent echoed through Q30, not a fit). NOT REPRODUCED: the line OD exponent 2.013 cannot be derived from Jacques law at two wavelengths, which gives 3.33 for any Beer-Lambert double pass; 2.013 is a least-squares fit over the diffusion-model reflectance (tn_od_exponent over tn_reflect), plausible for a diffusion regime between sqrt(mua) and linear but unverified by this class, owed a re-derivation through an independent reflectance model. Verdict PARTIAL: ITA, dITA, DE76 and the fade schedule CONFIRMED by hand, line OD exponent UNREPRODUCED, nothing changed. 2026-10-01 measure BACKFILL-HUMAN (c) tanning kinetics: THREE independent groups now counted, the two new ones read in full text: Copenhagen (Ravnbak 2008 and Schmalwieser 2025, one cluster through Philipsen, existing rows); Fudan University, Shanghai (Ma 2024 Phenomics, refs gr59_ma_fudan_2024_uvtime, mirror knowledge/fetched/cmp_graphics_bfh_ma2024_hanchinese_tanning.xml sha256 bf16a7b5: six UV doses on 31 backs, erythema index peaks on D1 and melanin index on D7 regardless of dose over 14 days, confirmed in a 664-person cohort); NCI-NIH with FDA CDRH (Coelho 2009, refs gr59_coelho_nihfda_2009_llp, mirror knowledge/fetched/cmp_graphics_bfh_coelho2009_longlasting_pigm.xml sha256 24e45089: pigmentation robust at day 16 and visible at 9 months or more in 32 percent after 2 MED and 61 percent after 4 MED). Independent count 3. SED: was 0 witnesses, now 1 primary: the CIE's own International Lighting Vocabulary term 17-26-069 (1 SED is an erythemal radiant exposure of 100 J per square metre, referring to ISO/CIE 17166; refs gr59_cie_eilv_17_26_069_sed, sha256 5e809d7c), with the CIE and ISO catalogue pages for ISO/CIE 17166:2019 (refs gr59_cie_pub_17166_2019, sha256 a7f03424; ISO page knowledge/fetched/cmp_graphics_bfh_iso_17166_page.html). Still open: the 12-page standard text with the s_er(lambda) table is sold (CIE Webshop), not mirrored; a first CIE URL variant answered 403. Search listings: cmp_graphics_bfh_epmc_tankin_search.json and cmp_graphics_bfh_epmc_tankin_search2.json 2026-10-01 land BACKFILL-HUMAN refs rows landed this lane in graphics.refs, each mirror pinned by nx_filehash, cited here so the refs referee's inline-citation tooth sees them: [gr59_ma_fudan_2024_uvtime] [gr59_coelho_nihfda_2009_llp] [gr59_cie_eilv_17_26_069_sed] [gr59_cie_pub_17166_2019] |
GR60 The support surface and its cover · cell WATCHING · gate NO-GATE-DECLARED
| Contract | _ABSENT_:gr_support_cover · executor Organ · 4 u · stands on softbody:SB15 (UNRESOLVED: no such rung on this plan), softbody:SB13 (UNRESOLVED: no such rung on this plan), softbody:SB10 (UNRESOLVED: no such rung on this plan), dcc:DC10 (UNRESOLVED: no such rung on this plan), procgen:PG29 (UNRESOLVED: no such rung on this plan) |
|---|---|
| Definition of done | The reference class reclines on a cushioned lounger: the cushion COMPRESSES under the body (softbody SB15, a support surface that spreads load) and its cotton cover DRAPES and wrinkles (procgen PG29 cloth XPBD, dcc DC10 pattern-sew-drape), with body-cushion contact free of interpenetration (softbody SB10; S4R-class resolution 2609.20524 is the September reference for procedurally assembled scenes). Done-rule: contact area and sink depth reported against the tissue and cushion stiffness rows, a non-trivial wrinkle field on the cover, zero interpenetration at the declared contact distance. [arxiv_co_siga2026] |
| Matrix cell | CONTACT: the lounger cushion compresses under the body and its cover drapes · runtime/nx_gm_cloth_lib.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_gm_cloth_lib; declare nx_gm_cloth_lib TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | no log row names this rung yet |
GR61 A camera model as data · cell WATCHING · gate UNRECORDED
| Contract | _ABSENT_:gr_camera_row · executor Organ · 3 u · stands on GR14, GR23, GR42 |
|---|---|
| Definition of done | A photograph is the SCENE TIMES A CAMERA. Exposure, sensor response and tone curve, focal length, aperture and depth of field, vignetting, flare and chromatic aberration become ONE camera row that the beauty tier (GR42) and the live tier read identically, on the linear-light pipeline (GR14) through the one colour transform (GR23). Done-rule: a rendered grey card and a photographed one agree within a DECLARED DE2000 band under the declared exposure, and depth of field matches a measured circle of confusion at two distances. |
| Matrix cell | CAMERA: exposure, sensor response, lens and depth of field are ONE camera row both tiers read · runtime/nx_gi_path.nx · WATCHING decl |
| Gate | nx_gi_path_gate · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) |
| Work log | no log row names this rung yet |
GR62 Reference-still referee: per-region DE2000 against a licence-clean photograph class · cell LANDED · gate NO-GATE-DECLARED
| Contract | _ABSENT_:gr_refstill_referee · executor Organ · 2 u · stands on skintwin:ST1 (UNRESOLVED: no such rung on this plan), GR55, GR30 |
|---|---|
| Definition of done | A reference photograph CLASS (licence-clean stills under the refintake lane's licence state, never an identity, never an unlicensed image) is measured per region in CIELAB: skin ITA, sand, fabric, cushion, sky. Every beach frame publishes its per-region DE2000 against that class beside the world referee grade (GR30), through the skintwin CIEDE2000 ruler (ST1 cs_de2000, built and promoted, ZERO adopters outside the garment lane). Done-rule: ST1 wired to the beach page, bands from measured references, the number on the page. This is the axis the operator's 2026-09-27 goal turns on: a real photo benchmarked, a published frame, no visible difference. |
| Matrix cell | REFEREE: per-region DE2000 against a licence-clean reference photograph class · runtime/nx_colorsci_lib.nx · LANDED decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_colorsci_lib; declare nx_colorsci_lib TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-09-27 measure nx_ladder graphics 2026-09-27: 56 rungs placed, targets 3, verdict LADDERED before this landing; skintwin and softbody, the two boards GR57-GR60 depend on, read NO-LADDER and are laddered the same session. 2026-10-01 measure REFERENCE 1 INTAKE AND ITS FIRST DEFECT, 2026-09-30: the operator supplied Wikimedia Commons private-ref:6a8ea809; licence read from Commons' own extmetadata: CC BY 2.0, author Tino & Tina - fr (Flickr Patmos rest), Restrictions empty; mirrored private-ref:6a8ea809 sha256 d421bd74 at exactly Commons' declared 186403 B (1174 x 673), metadata mirror private-ref:6a8ea809 5672c82a. Scene-class colour reference only, never the person's twin, not a refs row (refs mirrors are republished). nx_skinsample measure on the whole frame: mask_permil=518, L 62.014 a 4.075 b 12.912, ITA 42.9 deg -- the YCbCr skin rule admits the beige sand and stones, so a whole-frame beach skin mean is a sand mean. OUTSIDE ORACLE landed for scoring the fix: private-ref:6a8ea809 (13 conservative interior boxes: 4 skin, sun and dappled shade; 3 sand, sun and shade; sea, towel, rock, vegetation, hill; hair as a skin-coloured negative), read by Claude vision from a byte-identical copy; a mask's recall is its admitted share of skin-box pixels, its false admission the share of sand, rock, towel and hair pixels. 2026-10-01 land THE REFERENCE-STILL REFEREE LANDED, contract symbol present: gr_refstill_referee with gr_rs_acc_new and gr_rs_acc_add appended to buildroot/runtime/nx_colorsci_lib.nx (+8357 B additive, 38414 to 46771 B, sha256 eec69f03; compile-proven inside nx_refstill_lib and gated there first, only then moved, so no importer of colorsci ever saw a broken state). buildroot/runtime/nx_refstill_lib.nx carries the two-layer two-chromophore skin model (the dermis IS sf_forward over km_r_of_mix, the epidermis a double-pass melanin filter), every coefficient computed at run time from private-ref:fe89c6bb (sha256 31c01400) whose rows cite the mirrored Jacques 1998 Skin Optics and the Prahl haemoglobin table; the scale-free residual sf_sse(k t, m)/sf_sse(m, 0); the bound (sqrt(floor)+sqrt(quant))^2 with the lattice floor measured over every cell centre and edge midpoint and the quantisation term half the larger adjacent 8-bit code step per pixel; the declared regions private-ref:b7de2687 (sha256 a8d08af3, CHOSEN boxes, stated so); the class reducer. nx_refstill PROMOTED 142080 B sha256 4ee8bcbc, offc installed, registered (measure, class, oracle). nx_refstill_gate PROMOTED 110897 B sha256 cb8ffdac, offc installed, registered, rostered and mapped in graphics.gates; live run 22/22 GREEN (nx_job_run pid 9609): T3 sand-only neg-control BITE-PROVEN (5 sand colours the AT23 YCbCr rule admits, the model admits 0, closest residual 1413779 e-9 against its bound 176400 e-9), T4 positive control 16 of 16 chain-made colours recovered, T5 Sharma pairs 17 and 20 at 27.149 and 19.454 exact, T6 ITA quartiles exact, T8 sf_fit recovers the dermis blood fraction (22583 vs 22571 micro). Class file private-ref:d9101c46 (sha256 a2a387ad): 4 ADMITTED Commons stills (private-ref:6a8ea809 CC BY 2.0, private-ref:06caaa3c CC BY 2.0, private-ref:ff3b177c CC BY-SA 3.0, private-ref:80de5b96 CC BY 2.5; each mirror equals Commons' declared size and is pinned by nx_filehash) and 2 NOT-ADMITTED (licence No restrictions); not refs rows, nothing published. nx_refintake not used: its unit is a declared subject (a person slug), the opposite of this class's never-an-identity contract. 2026-10-01 measure THE DEFECT AND ITS FIX SIDE BY SIDE, nx_refstill class on the promoted binary 4ee8bcbc (nx_job_run pid 26615, /tmp/nx_refstill_class_live_1.out), every pixel to one region, partition SUMS on all four. private-ref:6a8ea809 790102 px: model skin 369 permil, L 49.465 a 8.086 b 10.735, ITA median -15.5 deg (q1 -72.8, q3 52.4) against the AT23 whole-frame YCbCr mean 518 permil, L 62.014 a 4.075 b 12.912, ITA 42.9; the model refuses 131293 YCbCr pixels and the bright sand becomes its own region (137 permil, L 78.570 a 0.462 b 15.158). private-ref:06caaa3c 3812256 px: model 146 permil L 49.390 a 9.448 b 9.685 ITA -23.0 vs YCbCr 142 permil L 55.505 a 8.505 b 10.220 ITA 28.3. private-ref:ff3b177c 9980928 px: model 400 permil L 59.304 a 3.658 b 9.212 ITA 47.7 vs YCbCr 408 permil L 59.614 a 3.540 b 9.336 ITA 45.8. private-ref:80de5b96 5018112 px: model 64 permil L 44.605 a 12.966 b 35.318 ITA -4.0 vs YCbCr 157 permil L 62.954 a 6.025 b 39.556 ITA 18.1. Bands private-ref:1d118014 (sha256 b882576e, 1348 B, NX-DERIVED, carries a CAVEAT row): skin centre L 50.691 a 8.540 b 16.238 radius dE00 10.801; sky 72.936 -1.823 -15.640 r 9.848; sea 50.095 -2.707 -14.652 r 7.178; sand 71.526 0.982 17.119 r 11.905; fabric 48.505 13.937 25.160 r 26.910. 2026-10-01 measure SCORED BY THE OUTSIDE ORACLE private-ref:6a8ea809 (13 boxes, scored by the oracle file's own rule), nx_refstill oracle on bytes identical to the promoted 4ee8bcbc (nx_job_run pid 8035, /tmp/nx_refstill_gr62_oracle1.out): RECALL over 22500 skin-box px, model 831 permil vs YCbCr 938; FALSE ADMISSION over 26400 sand, rock, towel and hair px, model 151 permil vs YCbCr 395. Per box, model vs YCbCr: skin sun 574 vs 1000, 906 vs 922, 1000 vs 986; skin shade 886 vs 892; sand sun 306 vs 995 and 140 vs 988; sand shade 484 vs 689; rock 175 vs 189; towel 0 vs 0; hair 7 vs 0; outside the negative set hill 739 vs 845, vegetation 0 vs 120, sea 0 vs 0. The model cuts false admission 2.6x for 107 permil of recall; what it still admits is shaded sand and the sunlit brown hill. 2026-10-01 lesson A COLOUR MODEL REMOVES THE DEFECT IT WAS BUILT FOR AND EXPOSES THE NEXT: the two-chromophore mask refuses yellow sand, but a colour-only mask cannot tell skin from a mineral of the same chromaticity -- private-ref:ff3b177c's model skin is 400 permil of the frame at a* 3.66 with an L* spread of 4.4 (grey-beige sand, by its statistics), and reference 1's sunlit brown hill is 739 permil admitted. The CAUSE is UNMEASURED; the leading hypothesis is the model's redness at its low-blood edge (the domain floor is Jacques' typical average 0.2 percent), and the deciding experiment is the model's own L*a*b* along its domain against MEASURED skin (queued) -- not a floor moved until one image scores better, which would be a threshold tuned to flatter one oracle. ALSO: the conf's checkable criterion caught a too-coarse lattice before any photograph was read (41 per axis: floor 200535 e-9 above a mid-grey pixel's 80059 e-9; 81 per axis: 60815 e-9 below), and the gate now holds it as tooth T1b. 2026-10-01 queue REMAINING DONE-RULE LEG, exact next steps: (1) the beach page -- a new nx_refstill verb that reads private-ref:1d118014 into gr_refstill_referee's band rows (GR_RS_BAND_HAVE, L, a, b of each centre), labels a captured beach frame with rfs_measure_rgb and prints each region's dE00 against its band, then the page emitter's row beside the GR30 world grade, through the beach page's own release lane. (2) BEFORE the skin band is shown as skin: calibrate the skin model against MEASURED skin (mirror a published skin reflectance or colorimetry set, walk the model's domain with cs_de2000 against it, fix the KM conversion or the channel wavelengths it indicts), re-score with the outside oracle, and land a second oracle file for another class reference before any domain change is believed. (3) recalibrate regions.conf against the class: private-ref:80de5b96 reads 470 permil fabric with mean a* -15.1 b* 47.8 (hue about 108 deg), a yellow-green surface the sand box (40 to 100 deg) misses, and 289 permil unmeasured (clipped). (4) mirror van Gemert and Star 1987 for the KM conversion row of skin_model.conf. 2026-10-01 measure TRIANGULATED BAR as of 2026-10-01 for the render-versus-photograph colour referee (skin segmentation in photographs and the colour-difference metric): SURFACE-COLOUR METRIC -- CIEDE2000 (CIE 2001, Technical Committee 1-47, for industrial colour differences) is still the standard formula for small differences between region-mean colours under declared reference conditions, but a correct implementation must pass the Sharma supplementary pairs because the CIE worked examples miss common implementation errors, and the formula is mathematically discontinuous in mean hue [sharma_ciede2000_2005]. DISPLAY METRIC -- for display-referred HDR and wide-gamut signals the standard is ITU-R BT.2124 DeltaE ITP from ICtCp, scaled so 1 is a just noticeable difference, deliberately over-predicting by presuming the most sensitive adaptation, and only ordinal for scene-referred signals without an assumed nominal peak luminance [itu_bt2124_2019]. PHOTOGRAPH METRIC -- the image-quality school finds CIEDE2000 no better than plain CIELAB on photographic pairs, pixelwise CIELAB fragile to slight misregistration, and a learned metric better on photographic colour distortions (STRESS 14.1 against 18.2 on the TID2013 colour subset) though worse than CIEDE2000 on uniform patches [cdflow_cvpr2023]; the newest listing entry (submitted 2026-09-30) is again a data-driven regressor that reports CIEDE2000 inconsistent with judgments, beside a September revision of a luminance-scaled Euclidean line element for surface colours, and October has no entry yet [arxiv_colordiff_202609] [csgr202610_gfx_t1001]. SKIN IN PHOTOGRAPHS -- apparent skin colour is read on extracted facial skin regions and reported as L* and hue angle rather than ITA alone [beyondskintone_iccv2023], because ITA from uncontrolled images is lighting-sensitive and unreliable [alipour_ita_limits_2026]; the closest published precedent to this rung, a photo-to-virtual-human pipeline, scores CIELAB Delta E and ITA under cheek and full-face masks over about 19848 renders and finds consistently higher error for darker skin [truetotone26]; the newest arXiv skin-segmentation entry is SkinMap 2510.05296 (October 2025, full-body skin regions weighted for remote photoplethysmography) and that listing holds no 2026 entry [arxiv_skinseg_202609] ; witnesses [sharma_ciede2000_2005] [itu_bt2124_2019] [cdflow_cvpr2023] [beyondskintone_iccv2023] [alipour_ita_limits_2026] [truetotone26] [arxiv_colordiff_202609] [csgr202610_gfx_t1001] -- six independent groups plus the current listings (Sharma at Rochester; ITU-R Study Group 6; Ma's group at City University of Hong Kong; Sony AI with the University of Tokyo; Technological University Dublin; PUCRS), one standard and four peer-reviewed papers (True to Tone is a preprint) ; WHERE THEY DISAGREE: standards keep handcrafted formulae (CIEDE2000 for surface colours, DeltaE ITP for HDR displays) while the photographic-image school says neither transfers to photographs and 2026 preprints keep proposing replacements no standards body has adopted; the skin-measurement papers reject ITA as the single reported number; and none of the witnesses read today builds a skin mask from a chromophore-fit residual, an absence not proven by exhaustive search (their skin regions come from landmark-based facial extraction or cheek and full-face masks) ; VERDICT on the running lane (updated after its land rows 1790825400 and 1790826001): GR62 landed gr_refstill_referee and the nx_refstill frame verb, which print each region's CIEDE2000 against its band centre beside the class radius through ST1's ruler, with the skin mask taken from the two-chromophore fit residual; it is AT the bar in comparing region statistics rather than pixels with CIEDE2000 under a declared illuminant, and BELOW it in four named places: prove cs_de2000 on the Sharma supplementary pairs (neither board cited them before today), add DeltaE ITP for the display-referred HDR frame, report L* and hue beside ITA, and widen the mask's validation beyond one reference photograph sco 2026-10-01 measure THE DECIDING EXPERIMENT, FIRST HALF -- THE MODEL'S OWN COLOURS, MEASURED: nx_refstill model on the promoted binary f03a67db (nx_job_run pid 17063, /tmp/nx_refstill_model_walk_1.out) walks the skin model as absolute reflectance (white diffuser 100000) at eighths of f.mel and at the floor, middle and ceiling of f.blood, 27 rows. At the blood FLOOR (0.2 percent) a* is 3.600 at no melanin, 3.105 at f.mel 5.4 percent (L 77.305 b 11.918 ITA 66.4), 3.575 at 10.75 percent (L 68.274 b 19.840), 4.590 at 16.1 percent, rising to 10.670 only at 43 percent; at the middle of the blood range (2.6 percent) a* is already 19.3 to 21.8 at every melanin; at no melanin b* stays near 0 (1.046, -0.150, 0.942) at every blood level. So the hypothesis in the lesson row is now a MEASUREMENT: the model's low-blood edge for light-to-moderate melanin is a pale beige at a* 3 to 5, exactly where private-ref:ff3b177c's model-admitted sand sits (a* 3.658 b* 9.212), and the model reaches no yellowness without melanin. The second half (the same walk against MEASURED skin colorimetry through cs_de2000) is the queued step; the floor is NOT moved on this evidence alone. 2026-10-01 land THE NUMBER THE PAGE OWES NOW EXISTS AS A VERB: nx_refstill frame <jpeg> <bands> reads private-ref:1d118014 into gr_refstill_referee's band rows (tc_atoi_signed, one row per region the bands name), labels the frame and prints each region's CIEDE2000 against its band centre beside the class's own radius. Promoted 174656 B sha256 32ac3a1b (contentdiff GREEN, lost_from_live=0, ONE-WAY-GAIN), offc installed; chain this session 4ee8bcbc (measure, class, oracle) then f03a67db (+model) then 32ac3a1b (+frame), the oracle scores reproduced exactly on f03a67db (recall 831 vs 938, false admission 151 vs 395; pid 17145). First frame refereed, reference 1 ITSELF (in-class, so within-class is expected and proves only the plumbing): skin dE00 3.875 vs radius 10.801, sky 4.831 vs 9.848, sea 7.178 vs 7.178 (private-ref:6a8ea809's own sea defines the radius, so the reducer and the referee agree through the one ruler), sand 5.297 vs 11.905, fabric 15.473 vs 26.910; other and unmeasured carry no band (nx_job_run pid 22393, private-ref:6a8ea809). 2026-10-01 queue SUPERSEDES step (1) of the 1790825404 queue row (the verb now exists): what remains of the done-rule is the PAGE -- capture a beach frame from the live beach page as a JPEG the sovereign decoder reads (baseline or progressive; a PNG needs a decode path the verb does not have), run nx_refstill frame on it against private-ref:1d118014, and have the beach page's emitter print each region's dE00 and within-class beside the GR30 world grade, through the beach page's own release lane, never by this lane editing the page. Steps (2) to (4) of 1790825404 stand: the skin band is NOT shown as skin until the model walk is compared with measured skin colorimetry and the outside oracle re-scores; a second oracle file before any domain change; regions.conf recalibration; the van Gemert and Star mirror. Also owed: the class verb should check each mirror against the sha256 its refs row pins before measuring it. 2026-10-01 land THE CIEDE2000 RULER IS NOW PROVEN PAIR BY PAIR ON THE FIRST AUTHOR'S OWN DATA FILE: ciede2000testdata.txt mirrored from Sharma's Rochester page through nx_research_fetch (knowledge/fetched/cmp_graphics_sharma_ciede2000_testdata.txt, 1830 B, sha256 44aebb39, refs row sharma_ciede2000_testdata, class dataset), 34 rows with the signs printed. nx_colorsci_gate T2b adds one gv_check_near tooth per pair (actual, expected, delta and tol emitted in 1e-6 units; tolerance 60e-6 = half an ulp of the four-decimal table, 50e-6, DERIVED by cg_half_ulp_micro, plus the incumbent Q30 chain headroom 10e-6, both written into every tooth name from the constants), tags the paper's own arctangent and mean-hue cases (pairs 7-16, read 2026-10-01 from the mirrored PDF through nx_pdf_text), and COMPUTES which pairs straddle the 0/360 cut from the lib's own G and cs_hue_deg under the lib's knife-edge epsilon (7 pairs: 11 12 15 17 18 19 20; pairs 10 and 14 tagged ON-THE-180-DEGREE-KNIFE-EDGE). T2c: the KAT table whose signs were INFERRED from the PDF hue column equals the data file on 238 of 238 L a b dE fields. Planted fixture: the naive mean hue without the 360 wrap is refused against the published h_bar_prime on all 7 boundary pairs while the lib's own is accepted. Staged build da5969fe (127128 B; nx_job_run pid 1676, /tmp/gr62_colorsci_gate_staged_1.out) passed 64/64 GREEN, worst pair 23 at 49e-6 (inside the half-ulp alone); nx_contentdiff live dc7b9777 (80042 B) vs candidate GREEN lost_from_live=0 ONE-WAY-GAIN; PROMOTED nx_colorsci_gate sha256 da5969fe09846ed2f35b2c9855a7a68dfc8e325e05a49eeaea3051b5298c288b with prev banked; offc installed. cs_de2000 itself is UNCHANGED (no defect found), so no skintwin ST1 pointer row is owed. 2026-10-01 measure BITE-PROVEN BY A REAL MUTATION OF cs_de2000 HUE HANDLING, ISOLATED FROM EVERY OTHER LANE: the lib and the gate were copied to private module names (nx_colorsci_lib_gr62bite, nx_colorsci_gate_gr62bite, importing only each other) and the wrap branch of the mean hue was replaced by the paper's named mistake, the naive (h1'+h2')/2; built 110435 B and run (nx_job_run pid 3758, /tmp/gr62_colorsci_gate_bite_mutant_1.out): 54/64 RED, and the ten failures are exactly the seven boundary-pair teeth (deltas from 23751e-6 on pair 20 to 4784644e-6 on pair 19; the mutant gives pair 11 7.179153 against the published 7.2195), the two T2 aggregates (dE00 within half an ulp, h_bar_prime) and the naive-mean fixture (FALSE-POSITIVE on the good input); all 27 non-boundary pair teeth stayed PASS, so the new teeth discriminate this bug class and nothing else. The shared lib was never mutated, so no restore was needed; the pristine staged run above is the GREEN half. Scratch retired reversibly into knowledge/retired (both sources and the staged elf); registry rows reaped (nx_toolreg_reap reaped=2); the _build fossil of the mutant is refused by nx_retire_path as a protected .elf name and remains an unregistered, sourceless build artifact. 2026-10-01 lesson A ZERO HUE IS NOT ALWAYS ZERO IN Q30: for the (2.5, 0) sample the lib's cs_hue_deg returns 359.999999 on three of the five pairs that carry it (16, 18, 20) and exactly 0 on the other two (17, 19), because CORDIC atan2(0, +a') leaves a one-micro-degree residual that depends on a' = 2.5(1+G). The mean-hue branch therefore takes the wrap on 18 and 20 where exact arithmetic would not, and skips it on 16 where exact arithmetic would take it, and all three still land on the published h_bar_prime and dE00 because the hue circle makes the two branches agree. dE00 is unaffected (34 of 34 within half an ulp); a reported hue of a pure +a* colour can print 359.999 instead of 0. A boundary class derived from a fixed-point hue is a property of the representation, so the tooth tags say which branch the LIB took, never which branch exact arithmetic would take. 2026-10-01 measure THE FIRST REAL EXPERIMENTS ON THIS BOARD, AS DATA (ecosystem EC86 renders them on every page; both PRIVATE: the subject is a real person, so the page carries their numbers and never an image). E1 (ab, mask YCbCr control against the two-chromophore skin model, the oracle boxes of private-ref:6a8ea809 as the unit of replication), re-measured tonight by nx_refstill oracle on promoted 32ac3a1b (direct call, verdict=ORACLE-SCORED, the same pixel totals as the land row: recall 938 vs 831 and false admission 395 vs 151 permil, pixel-weighted): per skin box YCbCr 1000, 922, 986, 892 against the model 574, 906, 1000, 886; per negative box (sand sun left strip, sand sun mid beach, sand shade, towel, rock, hair) YCbCr 995, 988, 689, 0, 189, 0 against the model 306, 140, 484, 0, 175, 7. Four skin boxes and six negative boxes are far under the incumbent sample floor, so the page reads INSUFFICIENT with that reason rather than calling the model better on recall or worse on false admission: the box means favour the model on false admission (the sand boxes) and YCbCr on recall (the sunlit thigh), and only more oracle boxes or more references decide it. E2 (factorial, mask x reference over the four admitted references private-ref:6a8ea809, private-ref:06caaa3c, private-ref:ff3b177c, private-ref:80de5b96), re-measured tonight by nx_refstill measure (nx_job_run pids 5224, 5795, 5994, 6105; /tmp/exl_measure_*.out): skin share permil and the admitted skin's mean L*, a*, b* under each mask; one frame per cell, so the mask main effect rests on four frames a side and reads INSUFFICIENT, while the effect sizes are on the page. The YCbCr side is the whole-frame rule (at23_ycbcr_whole_frame), the model side the region=skin row. 2026-10-01 queue DeltaE ITP FOR DISPLAY-REFERRED HDR FRAMES IS NOT BUILT, AND WHY: the standard is in hand (ITU-R BT.2124-0 (01/2019), mirrored 2026-10-01 at knowledge/fetched/cmp_graphics_itu_bt2124_2019.pdf, pinned, refs row on GR62) but no second, INDEPENDENT witness that applies it is mirrored, and the triangulation rule refuses a metric that stands on one document. Exact next steps: (1) mirror through nx_research_fetch and read through nx_pdf_text one peer-reviewed paper, from a group other than the standard's proponents, that evaluates DeltaE ITP on HDR or wide-gamut imagery, and pin a refs row naming the claim it backs; (2) build cs_de_itp in nx_colorsci_lib from the equations read off the mirrored BT.2124 and BT.2100 bytes (absolute linear BT.2020 light, the PQ inverse EOTF, ICtCp, the BT.2124 scaling), its constants as named UNIT rows, never typed from memory; (3) gate it in nx_colorsci_gate with gv_check_near teeth against any worked values the standard prints plus an independent re-derivation of at least two pairs (one at SDR white, one at a highlight), tolerances derived from the printed digits; (4) only then give nx_refstill an HDR frame verb. 2026-10-01 land L*, C*ab AND HUE ANGLE NOW REPORTED BESIDE ITA PER REGION, LIVE: the promoted nx_refstill (sha256 95c9d457b96e9c03e1071c904cf8f89d841f7c6710b87e4aeae89748a4b8a71b, 175448 B; serving root, _offc twin and buildroot/_build fossil byte-identical by nx_filehash cmp; supersedes 32ac3a1b) appends Cab_e3 and hab_e3 at the END of every measured region row beside the incumbent L_e3, a_e3, b_e3 and ita_* keys (rfc_lch_kv, composed from fq_hypot and cs_hue_deg, the same arithmetic cs_de2000 uses, so no second colour formula exists; every earlier key keeps its name and order). Witness run on the admitted private-ref:6a8ea809 reference (nx_job_run pid 18426, /tmp/gr62_lch_live_verify_1.out, verdict=MEASURED, partition=SUMS): skin L_e3=49465 a_e3=8086 b_e3=10735 Cab_e3=13440 hab_e3=53012; sky Cab_e3=24934 hab_e3=271039; sand Cab_e3=15165 hab_e3=88254. Hand re-derivation from the printed a and b agrees to the last printed digit (skin sqrt(8.086^2+10.735^2)=13.440, atan2(10.735,8.086)=53.01 deg; sky atan2(-24.930,0.452)=271.04 deg). 2026-10-01 queue OPEN, journaled at the lane deadline (NAS clock past 1790832600, so no new build was started): (1) nx_refstill_gate (promoted 110897 B) carries NO tooth on the new Cab_e3 and hab_e3 keys (nx_txtscan find hab in its source: total=0), so the additive fields are proven by the live witness run and a hand re-derivation only, not by a gate; owed: gv_check_near teeth on a synthetic region with known a and b (one in each hue quadrant, one at a=b=0 asserting h=0 per the CIE convention, one pure +a* row asserting the Q30 359.999999 residual named in the 1790829802 lesson), plus a contract tooth that every pre-existing key precedes Cab_e3 in the row; (2) DeltaE ITP stays queued exactly as row 1790830110 states (BT.2124 mirrored, the independent applying witness not); (3) nx_refstill_gate is REGISTERED-UNAUTHORISED per nx_catalog (no cap names it). 2026-10-01 land THE FIRST REAL EXPERIMENTS ARE PUBLISHED ON THIS BOARD (ecosystem EC86): scoped regen job 1790858482 verdict=GREEN, FORK-FRESH matrix e99dd74f, index.html 1365034 B (prev 1169619), section anchor #experiments with #exp-GR62-E1 and #exp-GR62-E2, both privacy private, no image rendered for either; nx_page_verify job 1790858564 VERDICT=GREEN (page 200, 5 of 5 declared assets PNG DECODE-OK, all of them estate world renders, a11y-issues=0). WHAT THE PAGE HONESTLY SAYS: E1 per box (unweighted, unit of replication = an oracle box) recall YCbCr 950.0 against skin model 841.5 permil, n=4 per arm, and false admission YCbCr 476.8 against 185.3 permil, n=6 per arm; every corrected verdict reads INSUFFICIENT because the incumbent sample floor is 30 units per arm and one photograph's oracle file carries 4 skin and 6 negative boxes. E2 is a 2 by 4 factorial with one frame per cell, so its 16-contrast family reads INSUFFICIENT throughout and the factor-effects table prints the mask main effect as a difference of marginal means, descriptive. The pixel-weighted totals of the 1790829840 row (938 vs 831, 395 vs 151) are a different aggregation of the same boxes and are not what the table prints. 2026-10-03 land THE PRIVATE REFERENCE STORE IS PARTITIONED BY LICENCE AREA, AS DATA (operator 2026-10-01: "we want to make sure that we actually keep media sources in their proper license areas not just reference things"). private-ref:5c6374be (3135 B): ordered rule rows (contract, NC, ND, SA, BY, CC0, PDM, Public; restricting tokens first, NO MATCH = reserved) and seven area rows each carrying its obligations as key=value (publish, attribution, sharealike, commercial, derivative): pd-cc0, cc-by, cc-by-sa, cc-nd, cc-nc, reserved, contracted. Each area is a directory under private-ref:a74101e4; the two admitted images were copied into their areas byte-verified (private-ref:6a8ea809 186403 B, private-ref:052c57ab 385722 B) and the flat copies retired reversibly to knowledge/retired/ (rename only). private-ref:e051925b (3539 B) gained an 18th field, area, and its header now admits every licence class each in its own area. New organs, all PROMOTED and REGISTERED: nx_refarea_lib (rfa_classify, rfa_check: ALLOW or a NAMED refusal REFUSE-PRIVATE, -RESERVED, -NONCOMMERCIAL, -NO-DERIVATIVES, -NO-ATTRIBUTION, -SHAREALIKE, -NO-GRANT, -UNKNOWN-AREA, -UNKNOWN-USE, -NO-ROW; privacy decided before the licence, every unreadable input fails closed), CLI nx_refarea (classify, check, audit; promoted e31de083, 33037 B), gate nx_refarea_gate (promoted cd19b346, 206996 B, _offc installed, row in knowledge/compare/graphics.gates) 2026-10-03 measure THREE METHOD CLASSES on the licence-area and privacy guarantees: (1) nx_refarea_gate 40/40 GREEN on the staged and again on the promoted binary, every refusal tooth printing the outcome NAME beside the expected one and paired with an allowed positive control (pd publish without attribution, cc-by publish and commercial with attribution, nc publish non-commercially, sa adaptation under the same licence), plus neg-controls (a token must be WHOLE: BYTE is not BY; an unreadable conf classifies everything reserved and refuses every public use); the publisher predicate rp_private_in (nx_swcompare_lib, in front of rp_publish) refuses a store path, a ../ spelling and a manifest-named also_at copy while an unnamed mirror and a lookalike SUBSTRING of a named path stay publishable; (2) bite-proven: nx_gate_bite nx_refarea_gate nx_refarea_lib.nx compiled a comparison mutant and the gate went RED (killed=1, restore byte-identical 14006 B); (3) independent live witnesses: nx_refarea audit reads rows=2 clean=2 (area equals the classification of each row's own licence words, privacy private, filed in the area directory, stored bytes equal declared) and nx_refarea check private-ref:6a8ea809 publish attrib=yes reads REFUSE-PRIVATE. The publisher refusal is LIVE on both generators: the promoted nx_swcompare_matrix (cb4a476f, 1441520 B) and nx_swcompare_sota (e4b5ae25, 812896 B) each carry the private-ref:e10e037d refusal (nx_txtscan count 2) and are byte-identical to their buildroot/_offc forks (nx_filehash cmp identical=1); nx_absent over buildroot/knowledge/compare *.refs finds no row naming private-ref:e10e037d, private-ref:6a8ea809 or private-ref:052c57ab (ABSENT-PROVEN, corpus_complete=1) 2026-10-03 measure EXPERIMENT GR62-E3 LANDED AS DATA, PRIVATE, NUMBERS ONLY: a 2 by 2 factorial (mask YCbCr vs skin model, by ref private-ref:6a8ea809 vs private-ref:052c57ab, both cited by private-store id, no path and no image on the page), nx_refstill measure on the area-filed copies. Skin share permil: private-ref:6a8ea809 518 vs 369, private-ref:052c57ab 905 vs 906; mean skin L*/a*/b* (thousandths): private-ref:6a8ea809 YCbCr 62014/4075/12912, model 49465/8086/10735; private-ref:052c57ab YCbCr 70581/9915/11250, model 67828/9745/11415. Re-derivation by hand from the instrument's own pixel counts: 291780/790102 = 369.3, 3915579/4320000 = 906.4, 3911529/4320000 = 905.4 permil; the private-ref:6a8ea809 model row reproduces E2 and the 1790856400 land row to the last digit. What it says: on a studio frame with no sand the two masks agree to 1 permil, so E1's large false-admission gap is a property of sand, not of skin; one frame per cell, so every contrast reads INSUFFICIENT and the result is descriptive. The model labels 68 permil of the studio ground as sand (L* 79.8, b* 4.7): the region vocabulary has no studio-ground class 2026-10-03 lesson A RETIRED COPY LEAVES THE PRIVACY FENCE: nx_retire_path renames into knowledge/retired/, a path carrying no private-ref:e10e037d word and named by no manifest field, so the two retired flat copies would be publishable by rp_publish if a refs row ever named them. No row does (ABSENT-PROVEN above), but the fence is by path, not by content: a retired private image must keep its name inside the store, or the manifest must list retired copies 2026-10-03 queue OWED, exact: (1) rp_private_in should also refuse private-ref:abf0ed69 (the retire rename keeps the word refstill inside the leaf, so extending the word rule to match refstill_ as well as private-ref:e10e037d closes it with one tooth); (2) the other three beach references of E2 (private-ref:06caaa3c, private-ref:ff3b177c, private-ref:80de5b96, stored in knowledge/fetched/ and named in private-ref:d9101c46) are real-person photographs outside the private manifest: register each with its area and also_at so the publisher fence covers them; (3) rp_publish_gated could compose rfa_check so a public refs mirror is refused by licence area too, once refs rows carry a licence word (today the class field is a source-type vocabulary); (4) a studio-ground region class for nx_refstill so a light backdrop is not counted as sand 2026-10-03 land PUBLISHED: scoped regen job 1791041959 verdict=GREEN, FORK-FRESH matrix cb4a476f, graphics/index.html 1390566 B (prev 1378701) carrying #exp-GR62-E3, privacy private, no image for it; nx_page_verify job 1791042000 VERDICT=GREEN (page 200, img=5 all estate world renders PNG DECODE-OK, a11y-issues=0). Every .jpg string on the page is a filename inside journal prose, none is an img or a link (nx_txtscan find .jpg: 12 hits, all prose) 2026-10-03 land THE PRIVATE-STORE TEXT LEAK IS CLOSED IN THE EMITTER, ONE RULE FOR BOTH GENERATORS (independent verifier 2026-10-03: the store was never served as an image, but plan journal text rendered verbatim, 34 store-path occurrences on this page and the manifest path on aesthetictwin). New leaf lib buildroot/runtime/nx_privref_lib.nx (prvr_redact, prvr_redact_in, prvr_token_for): a span around any case-insensitive match of the store word, a manifest row id, a former id, or the STEM of a stored path, source url, description page, meta mirror or also_at copy is replaced by private-ref:<8 hex of the row id> (a store-word span no row claims is hashed from its own text; a query url contributes its whole basename so the word index is never a needle; - and none are never needles); no match returns the same pointer. Wired at the three text funnels: wnote and wj in nx_swcompare_lib (every note, plan, log and api.json string) and exl_esc in nx_experiment_lib (every experiment cell, redacted before escaping). rp_private_in now reads also_at as a comma LIST (rp_priv_field_find). Store items carry OPAQUE ids rs-<first 8 hex of the sha256 pin> (private-ref:052c57ab, private-ref:6a8ea809, private-ref:06caaa3c, private-ref:ff3b177c, private-ref:80de5b96), the manifest is the only id-to-file map and gained field 18 former (the old stems), E1/E2/E3 exparm rows cite only the opaque ids. The three E2 beach stills are registered (licences read from the mirrored Commons extmetadata: CC BY 2.0, CC BY-SA 3.0, CC BY 2.5, Restrictions empty), copied into their areas byte-verified against their pins, and every retired flat copy is an also_at item. licence_areas.conf: rule NC+ND -> new area cc-nc-nd (commercial=no, derivative=no) ordered ahead of NC, nx_refarea_lib rfa_has_all (a + token fires only when every part is a whole token); before this CC BY-NC-ND landed in cc-nc with derivative=yes and the ND obligation was lost. 2026-10-03 land THE PRIVACY FILTER NOW SITS AT THE PUBLISH CHOKEPOINT, NOT ONLY AT TWO FUNNELS (re-measured after the 1791044400 land): wnote and wj were filtered but the ladder writer sg_text was not, so /compare/aesthetictwin still printed the manifest path inside the AT57 acceptance text (the one store-path occurrence left on either page; a per-char leaderboard field writer bypassed both funnels too). Fix in the path: nx_compare_regen rg_emit_receipted runs every captured byte of every generator (index.html and api.json, every domain) through nx_privref_lib prvr_redact_n before the receipted write and prints private_spans=N on each published line. nx_privref_lib gained the bytes form (prvr_redact_core, prvr_redact_n, prvr_free: exactly sized output, no NUL needed), the retired-copy word private-ref:83b549b7 with or without a manifest (the retire rename keeps the store word in the leaf), and backslash and both parentheses as span delimiters (a JSON escape is never eaten into a token; the prose bracket survives). Promoted with every LOST line read: nx_compare_regen 726f4b57 389736 B (LOST 2 = the old sys_mmap ENOMEM wording, superseded by the shared syscalls arena rewrite), nx_swcompare_matrix 2afa1f7b 1448024 B and nx_swcompare_sota 9a7d7acb 823360 B (LOST 2 each = STS-LOCK wording superseded by the plane-lock rewrite), nx_refarea_gate 3cd60ae1 245808 B; all offc-installed, regen FORK-FRESH on the new matrix 2026-10-03 measure THREE METHOD CLASSES on the chokepoint filter: (1) nx_refarea_gate 71/71 GREEN via /api/gate_run on the promoted binary (85cff2c8), new teeth each paired with a control: the retired-copy word redacts with no manifest, the manifest's bare file name (derived from its path, no new literal) redacts and keeps its numbers, the bytes form returns an exactly sized buffer (26 for 26) with both JSON escapes outside the token and nothing read past n, and public text returns the same pointer with zero spans; (2) mutation: nx_gate_bite nx_refarea_gate on the filter lib killed=1 of 1 valid mutant, restore byte-identical, both twins restored; (3) independent byte scans of the served files after republish (nx_txtscan count): the store word with its slash, the retired-copy word, the manifest file name, and the stem of every one of the five stored items (searched by name, named here by count only) read 0 on graphics index.html and api.json, aesthetictwin index.html and api.json, and feed.xml; the published lines carry private_spans (aesthetictwin index 1 = the ladder acceptance text the funnels missed). Residual, named: the bare word refstill remains only as code names (graphics: 46 nx_refstill and 15 gr_refstill_referee, the matrix contract symbol and its cell anchor, plus 4 prose words; aesthetictwin api: 2 nx_refstill); a code name carries no store path and is deliberately not redacted. Generators matrix 2afa1f7b and sota 9a7d7acb carry the filter one revision behind the lib (no manifest-file-name needle); the regen net applies it to every byte they emit. Regen promoted 7024b2cc (contentdiff GREEN, lost 0) 2026-10-03 lesson THE FILTER MATCHES WHOLE NEEDLES, SO A JOURNAL THAT QUOTES A SHORTENED STEM LEAKS PAST IT: this seat's own measure row listed the searched stems in shortened form, two of them were not whole needles, and the republished page carried them until the row was rewritten to name the items by count (corrected in place minutes after landing; the served bytes were re-scanned after). Imprecision chosen and named: containment of a whole stem, id, former id or store word; a fragment of a stem is not a needle (splitting stems into words would redact common English). Rule for every seat: name a private item by its opaque id or by count, never by any part of its file name 2026-10-03 queue INTAKE CLASSES SOURCE-LANDED, NOT YET PUBLISHED (build admission refused every build for 55 minutes, load1 19 to 25 against the 16.43 ceiling; nothing promoted, served pages unchanged). Measured before the fix by served-byte count on aesthetictwin: the pictured person's surname 68 in index.html and 6 in api.json, the intake store path 28 and 4, the vault root 6 and 0, vault cids 10 prose occurrences, a third-party site name 12; graphics read 0 on name, intake path and vault root. nx_privref_lib gains four needle classes, all containment and over-redacting: (3) the intake store word and its retired leaf plus the vault root, as consts so an unreadable intake still redacts; (4) a vault cid span unless it is a public gallery url; (5) the stem of every oracle file in the intake directory LISTING (names only, no intake file opened); (6) rows of a private names conf kept inside the intake store (name rows: containment in any case and spelling; stemnum rows: a span that is the prefix followed by digits). Tokens are a class-salted hash of the folded needle, one token per name across spellings. Fixture forms (prvr_redact_in, prvr_redact_n_in) read no intake so existing teeth stay deterministic; new prvr_redact_in2 and prvr_redact_n_in2 take the conf and directory; live prvr_redact and prvr_redact_n (the regen chokepoint) read both. nx_refarea_gate gains 20 teeth (section E: each class redacted, public gallery url and non-hex dash and non-oracle stem and contract symbol and citation key left alone, one token per name, unreadable intake still redacts the words, fixture form reads no names, plus live teeth that every declared name row and every live oracle stem redacts). RECIPE when the box clears: queued rows BQ-nx_refarea_gate and BQ-nx_compare_regen fire on the first nx_orchestrate pass with headroom; then run the gate (expect GREEN), nx_contentdiff live vs staged regen and read every LOST line, /api/promote with expect_sha256, nx_offc_install, scoped regen aesthetictwin then graphics, nx_page_verify, and recount served bytes. Residual named: the bare prefix of contract and citation symbols stays (it is code, not a stem); a third-party video host named in one public acceptance row is not declared 2026-10-04 land INTAKE CLASSES PUBLISHED (the 1791063220 queue row closed). Queued rows had not fired (builds predated source by ~1.6 h), so built by hand once admission granted (load1 3.4, blocked 0). nx_refarea_gate --qualify read 87/88 RED from CWD buildroot: section E 20 of 20 PASS incl. live names 6 of 6 and live oracle stems 2 of 2; the one RED was live-conf-files-cc-by-nc-nd-in-cc-nc-nd plus 0 live manifest rows, both because the live conf and manifest are serving-root relative and --qualify anchors CWD to buildroot (the conf does carry the NC+ND row). The same candidate staged and run from the serving root: nx_behaveprobe other_rc=0 (GREEN). Content ruler, every LOST line read: gate lost 0 (gained 189), regen lost 0 (gained 29), matrix lost 0 (gained 120), sota lost 0 (gained 122), all GREEN ONE-WAY-GAIN. Promoted with expect_sha256: gate f12f25eb, regen 664f55d4, matrix 32acd2d6, sota f91c638a; nx_offc_install promoted on all four, buildroot/_offc REFRESHED for regen, matrix and sota; regen FORK-FRESH on the matrix fork. Scoped republish aesthetictwin (index private_spans=20, api 0) and graphics (0 and 0), each GREEN and each rewriting feed.xml; nx_page_verify GREEN on both. Served-byte recount (nx_txtscan count, items named here by count only) over aesthetictwin index.html and api.json, graphics index.html and api.json, and feed.xml: every private.names name row in three case forms 0 in all five files; the third-party site name 0; stemnum spans (prefix plus each digit 0-9) 0, the bare prefix surviving only as function names and citation keys; the intake store path 0; the vault root 0; both oracle stems 0; optout.names carries no rows yet (0 needles). Vault cids: 1 in each aesthetictwin file, both the declared public gallery url; feed 1 prose mention with no hex. private-ref tokens on the pages: aesthetictwin index 371, graphics index 260. Residual, named: aesthetictwin index.html carries two non-oracle intake subject stems at 2 occurrences each (exempt by design, the allow-non-oracle-intake-file-stem control) and 16 plan-plane ids sharing a prefix with intake plan file names; one gallery image path under the NAS ai image tree (not the vault). Gate binary on its serving-root run is the evidence; a NAS /api/gate_run receipt with the live manifest rows was not taken this leg |
GR63 Real-versus-render indistinguishability bench with outside liar-killers · cell WATCHING · gate NO-GATE-DECLARED
| Contract | _ABSENT_:gr_realrender_bench · executor Organ · 3 u · stands on GR62, GR61, GR42, gameengine:GE51 (UNRESOLVED: no such rung on this plan) |
|---|---|
| Definition of done | THE ACCEPTANCE RULER FOR THE OPERATOR'S GOAL (2026-09-27: benchmark a real photo, publish our frame, and not tell the difference). Pairs of licence-clean real stills and our frames of the same scene class (a synthetic human, never a pictured identity) are judged by (a) outside synthetic-image detectors run OUTSIDE the build path (NVIDIA's SIGGRAPH 2026 synthetic-video detector NIM is the named first oracle), (b) outside VLM judges through the referee panel, and (c) a human two-alternative forced choice by the operator and the team. The score is detection accuracy driven toward chance (500 permil) with the sample size declared, sequential looks corrected and an A/A calibration (gameengine GE51's arithmetic). Done-rule: a published bench page on /compare with the three arms, receipts and the receipt's date; the rung reads INCOMPLETE until all three arms exist and the operator's ACCEPT row is on the plane. [nvidia_siggraph2026_blog] |
| Matrix cell | BENCH: real-versus-render indistinguishability with outside liar-killers · runtime/nx_world_quality_grader.nx · WATCHING decl |
| Gate | NO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_world_quality_grader; declare nx_world_quality_grader TAB <gate> and the ship loop proves it and the verdict lands here. |
| Work log | 2026-09-27 lesson A SEARCH-ENGINE SUMMARY IS NOT A WITNESS: a search result asserted an NVIDIA hybrid real-time subsurface-scattering talk in the SIGGRAPH 2026 Advances course; the mirrored course page and NVIDIA's events page (1010241 bytes) carry the token subsurface ZERO times, so no bar was raised on it and the claim is recorded as unwitnessed. 2026-09-28 measure THE CROSS-BOARD PHOTOREAL CRITICAL PATH IS MECHANICAL (2026-09-28, nx_compare_rank program graphics:GR63 on promoted b5450195): 3 boards, 19 nodes, 7 done, 12 open, 0 unresolved, 0 cycles; prerequisites-first path #1 GR42 (beauty-tier first frame) #2 GR55 (skin-tone fidelity of the live render) #3 GR30 (world-quality referee) #4 GR14 (linear-light pipeline) #5 GR23 (one colour transform) #6 gameengine:GE6 (native raster lane) #7 gameengine:GE15 (frame-pacing referee) #8 GR62 (reference-still DE2000 referee) #9 GR61 (camera model as data) #10 gameengine:GE41 (browser-engine parity bars) #11 GR63 (real-versus-render bench, the acceptance ruler) #12 gameengine:GE56. Per-board: XDEP domain=graphics edges=16 done=9 open=7 unresolved=0 after the GE40 path fix and the ST6 repoint (before them: unresolved=3). Read: gameengine GE51 reads DONE on an OPEN prerequisite GE56, a done-on-open inversion the program verb does not yet name as a class (queued) |
dossiers=63 cells_joined=56 no_matrix_row=7 (sum 63) gates_recorded=0 gates_unrecorded=20 no_gate_declared=36 (sum 56 of cells_joined) evgates=UNSTAMPED
Capability comparisons
Compare the field, search individual capabilities and open their source and adoption evidence. Documented presence does not establish comparative quality.
Position map — centrality and distinctiveness
The four-quadrant map the field uses for brand strategy (Dawar and Bagga, HBR June 2015), re-derived from this matrix on every publish. Centrality is the share of the category's feature mass a player covers, each feature weighted by how many hold it; distinctiveness is the average lead over each rival on the rows the player holds; breadth is the depth-weighted share of the whole matrix (the bubble); depth is how deeply the rows held are held; momentum is the day-over-day move off the spine (green rising, red falling, grey until day two); the dashed path runs first day → previous day → today. Dividers are the category means. Axes are fitted to the field of play, so read the tick numerals, not the frame. Rival marks are documented presence, so a rival's position reads the record, never its quality. The picture grades its own readability below; the table beside it is the same data for a screen reader or a second method.
knowledge/compare/graphics.cdmap (cut|x|y, cube|x|y|z); absent = the HBR defaultsbubble area = breadth · ring = momentum (green rising, red falling, grey until day two) · dashed = category means · axes fitted to the field of play: centrality 400–1000, distinctiveness 180–340 of 0–1000 permil (the full range put every player in one corner)
readability of the 2D cut, self-graded by the layout ruler: label overlaps 0 · labels over marks 0 · off-canvas 0 · unresolved labels 0 · mark overlaps 0 (a fact of the data: two players that close are that close) · data spread 575 permil of the plot · quadrant words unseated 0
text contrast, measured with wcag2-ratio (floors from contrast.conf), light theme: labels 16.24 (floor 4.50) · notes 16.24 (floor 4.50) · quadrant words 5.89 (floor 4.50) · tick numerals 5.89 (floor 4.50) · axis titles 5.59 (floor 4.50) · dark theme: labels 13.78 (floor 4.50) · quadrant words 5.55 (floor 4.50) · axis titles 5.97 (floor 4.50) · dark classes under their floor 0 (one figure serves both themes: a dark shortfall is a token to fix, never a class to hide) · classes refused under their floor 0 (a refused class is not drawn; the scale classes are measured, never hidden) · export: SVG PNG (receipt, rendered by the estate's own rasteriser from this page)
readability of the 3D cube, self-graded by the layout ruler: label overlaps 0 · labels over marks 0 · off-canvas 0 · unresolved labels 0 · mark overlaps 3 (a fact of the data: two players that close are that close) · data spread 302 permil of the plot
10 panels over 5 registered axes, every pair once (the lower axis on x, the higher on y) · each panel fitted to its own field of play, first and last tick numerals shown · no labels in a small cut, the legend names the colours; the grade below reports the mark terms only (MM summed over the panels, spread averaged)
readability of the small multiples, self-graded by the layout ruler: label overlaps 0 · labels over marks 0 · off-canvas 0 · unresolved labels 0 · mark overlaps 18 (a fact of the data: two players that close are that close) · data spread 511 permil of the plot
| Player | Quadrant | centrality | distinctiveness | breadth | depth | momentum | Rows held | Days on spine | First seen |
|---|---|---|---|---|---|---|---|---|---|
| Nishi | Unconventional | 553 | 303 | 336 | 692 | 0 since 2026-10-04 | 51 | 15 | 2026-09-15 |
| Unreal Engine 5 | Aspirational | 921 | 291 | 593 | 831 | 0 since 2026-10-04 | 75 | 15 | 2026-09-15 |
| Godot | Peripheral | 771 | 217 | 450 | 802 | 0 since 2026-10-04 | 59 | 15 | 2026-09-15 |
| Infinigen | Peripheral | 784 | 268 | 476 | 806 | 0 since 2026-10-04 | 62 | 15 | 2026-09-15 |
| Houdini | Aspirational | 906 | 278 | 565 | 812 | 0 since 2026-10-04 | 73 | 15 | 2026-09-15 |
| Day | Matrix rows reviewed | Players recorded |
|---|---|---|
| 2026-09-16 | 98 | 5 |
| 2026-09-17 | 98 | 5 |
| 2026-09-18 | 98 | 5 |
| 2026-09-19 | 98 | 5 |
| 2026-09-20 | 98 | 5 |
| 2026-09-21 | 98 | 5 |
| 2026-09-22 | 98 | 5 |
| 2026-09-23 | 98 | 5 |
| 2026-09-27 | 105 | 5 |
| 2026-09-28 | 105 | 5 |
| 2026-10-01 | 105 | 5 |
| 2026-10-03 | 105 | 5 |
| 2026-10-04 | 105 | 5 |
| 2026-10-08 | 105 | 5 |
players 5|matrix rows 105|feature mass 320|centrality mean 787|distinctiveness mean 271|axes 5|spine days 15 (shown 14)|rows written today 0|readability defects 2D 0 cube 0|spine knowledge/status/cdmap/graphics.spine
Capability matrix — measured against source
◉ authored stronger claim; source present● present◐ partial○ absent? unassessed; excluded from relative comparison · click any capability for its evidence
Peer cell coverage: authored assessed 420, unassessed 0. Authored assessments are not verified outcomes.
No capabilities match this filter or search.
| Capability | Nishi | Unreal Engine 5 | Godot | Infinigen | Houdini |
|---|---|---|---|---|---|
| RENDER | |||||
energy-conserving GGX Cook-Torrance PBR, all integerMeasured:gp_ggx exists in runtime/_hdl_build/nx_gx_pbr_ggx_gate.nx, verified at emit. Microfacet D times F times G over 4 NoV NoL, gate-proven on a metallic roughness sweep -- bit-deterministic where float engines drift. [walter2007] Adoption: GATE:LIVE trial=GREEN — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Unreal Engine 5 (leads), Houdini (leads) | ● | ◉ | ● | ● | ◉ |
SSAO and depth-of-field in the sovereign rendererMeasured:as_ssao exists in runtime/_hdl_build/nx_anat_sov.nx, verified at emit. as_ssao with an attenuation floor that never crushes a cavity to black, plus as_dof; the same organ carries the SSS wrap, tone variation and micro-relief skin stack (sec27-30 at /world/anatomy_gpu). rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: REGISTERED-DARK — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked).Pros Implementing source symbol found; runtime acceptance separateCons behind Unreal Engine 5 (leads), Godot (leads), Houdini (leads); not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked) | ● | ◉ | ◉ | ● | ◉ |
pre-integrated SSS bound to baked normalsMeasured:sr_sss_blur exists in runtime/nx_sdfrender.nx, verified at emit. LANDED 2026-08-25 -- Penner's SECOND half. âš THE WATCH SYMBOL WAS DELIBERATELY CHANGED FROM sr_sss_preint TO sr_sss_blur: sr_sss_preint ALREADY EXISTED, already wired and already toothed BEFORE this rung (nx_sdfrender.nx:782, called in sdf_shade_ray, pinned by nx_sdfrender_gate T6), so leaving the watch on it would have flipped this row GREEN on a symbol that does not implement the contract -- the name-keyed completion defect this board has already refused once. The shipped half was the CURVATURE contribution (SDF Laplacian); the half that stops the clay read is per-channel NORMALS, and it was absent. Now each channel is lit by its own blur of DETAIL-toward-GEOMETRIC normal, so fine relief scatters (red loses it, blue keeps it) while the coarse surface stays sharp -- sdf_shade_ray banks the geometric normal BEFORE the relief perturbation and specular, rim, fill, AO and shadow all keep it, which is Penner's own split. sr_sss_preint DELEGATES to sr_sss_preint_n so there is one copy of the arithmetic and every existing caller and tooth is bit-identical by construction; the identity dot(lerp(a,b,t),L) = lerp(dot(a,L),dot(b,L),t) means no per-channel normal is built or renormalised in the hot loop, so no isqrt was added. WEIGHTS DERIVED, NONE DIALLED, and zero conf rows were needed: the profile is the existing SSS_WRAP_R0/G0/B0 already cited and pinned by two gates, the relief width is the baker's own NTB_RELIEF_UM/LAMBDA (its comment declares it a stated physical model, not an artist slider), and blur = W/(W+W_relief) = 179/138/86. nx_sdfrender_sss_gate 11/11 GREEN via /api/gate_run exit_code 0. T6 IS THE LOAD-BEARING TOOTH AND ITS BAR IS NOT A CONSTANT -- it calls the SHIPPED wrap-only path at runtime as the incumbent control (354 vs 741 permil), and a plain blur scores 758, PASSING T5 and FAILING T6, so T5 alone is measurably gameable and that is proven rather than asserted. Bite-proven by manual plant (blur ignoring its channel argument -> 7/11 RED exit 1, dying at T0/T2/T6 with the neg-control collapsing to FALSE-POSITIVE), restore proven byte-identical through the WHOLE BUILD CLOSURE rather than by byte count. Sibling nx_sdfrender_gate rebuilt against the change: 6/6 GREEN, T6 values bit-identical. OPEN AND STATED: the renderer still uses PROCEDURAL mesostructure and does not yet SAMPLE the baked map (reach: baked relief 51 vs procedural tilt 18), so SAMPLING the baked normal map is the follow-on rung; and no trustworthy before/after image exists because nx_sdfrender_gate's write_png is not landing and its T5 gates on determinism only and discards the return -- debt 1787678931, and the A/B was therefore published as pixel counts, not as a picture. [penner2011] Adoption: LIB-WIRED importers=64 nonval=43 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Unreal Engine 5 (leads), Houdini (leads) | ● | ◉ | ● | ● | ◉ |
per-frame path-traced dynamic GIMeasured:gp_bounce_gi exists in runtime/nx_gi_path.nx, verified at emit. Lumen is the bar. Ours today is cached-form-factor radiosity at 20.3x less work with cache-vs-rebuild EQUALITY provable because integer determinism -- a statement float engines cannot make. Contract: per-frame bounce plus ReSTIR-class sampling. [bitterli2020] Adoption: REGISTERED-UNAUTHORISED — PARTIAL: registered, no cap ever minted; no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked).Pros Implementing source symbol found; runtime acceptance separate; ahead of GodotCons behind Unreal Engine 5 (leads), Houdini (leads); not adopted yet: registered, no cap ever minted; no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked) | ● | ◉ | ◐ | ● | ◉ |
anti-aliased analytic rasterMeasured:aa_render exists in runtime/nx_aa_raster.nx, verified at emit. The supersampled coverage rasterizer behind fonts, bubbles and UI; nonzero-winding union in one pass. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: LIB-WIRED importers=16 nonval=10 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Unreal Engine 5 (leads), Godot (leads), Houdini (leads) | ● | ◉ | ◉ | ● | ◉ |
| MATERIAL | |||||
skin albedo from the genome via the published ITA standardMeasured:si_albedo_of_gene exists in runtime/nx_skin_ita.nx, verified at emit. CIE Lab ITA classes dark to very-light with regional offsets; the world keeps its identity in the outfit while the individual owns her skin. Replaced twelve free sRGB constants. [chardon91] Adoption: LIB-WIRED importers=42 nonval=41 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Houdini (leads) | ● | ● | ● | ● | ◉ |
anatomy-derived UV unwrapMeasured:ntx_apply exists in runtime/nx_nxa_texc_lib.nx, verified at emit. Bone-anchored unwrap written into the asset, additive and idempotent, regions-from-anatomy exact with sums checked. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: LIB-WIRED importers=11 nonval=8 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen (leads), Houdini (leads) | ● | ◉ | ◉ | ◉ | ◉ |
deterministic albedo atlas bake and samplingMeasured:ntb_bake exists in runtime/nx_nxa_texbake_lib.nx, verified at emit. Gate 11 of 11; atlas and textured views published at /world/anatomy_gpu; 1024 and 2048 proven. [kkguides] Adoption: LIB-WIRED importers=13 nonval=10 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen (leads), Houdini (leads) | ● | ◉ | ◉ | ◉ | ◉ |
full PBR map set regenerated per genomeMeasured:ntb_bake_pbr exists in runtime/nx_nxa_texbake_lib.nx, verified at emit. FLIPPED 2026-08-30 (the note below already declared it LANDED while the symbol field still spelled _ABSENT_, a cell that could never flip as written). Albedo alone reads as clay; the measured floor is 4 maps by 4 regions. Rivals ship fixed files -- ours regenerate per genome at load: ntb_bake_pbr LANDED (nx_nxa_texbake_lib: albedo, spec, gloss, relief-derived normal by 4 regions, skin-IOR F0; verified func-not-comment at admission). rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: LIB-WIRED importers=13 nonval=10 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen (leads), Houdini (leads) | ● | ◉ | ◉ | ◉ | ◉ |
| BODY | |||||
procedural body from an anthropometric canon, zero scanned assetsMeasured:bg_parse_canon exists in runtime/_hdl_build/nx_body_gen.nx, verified at emit. The canon is DATA rows (rings, relief, layers); BodyParts3D is oracle only, never product. [bodyparts3d] Up to 8.3M tris with offered-vs-emitted demand reporting, so a capped emit can never again pass as complete. Adoption: REGISTERED-DARK — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked); only referenced by nx_creatorfit -- a mention, not a run.Pros Implementing source symbol found; runtime acceptance separate; ahead of Unreal Engine 5, Godot, HoudiniCons behind Infinigen (leads); not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked); only referenced by nx_creatorfit -- a mention, not a run | ● | ◐ | ○ | ◉ | ◐ |
Catmull-Rom spline lofting through canon ringsMeasured:bg_cmr exists in runtime/_hdl_build/nx_body_gen.nx, verified at emit. C1-smooth organic curvature from the same canon; Infinigen's higher-order NURBS lofted from real-reference profiles is the named remaining smoothness edge. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: REGISTERED-DARK — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked); only referenced by nx_creatorfit -- a mention, not a run.Pros Implementing source symbol found; runtime acceptance separate; ahead of GodotCons behind Infinigen (leads), Houdini (leads); not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked); only referenced by nx_creatorfit -- a mention, not a run | ● | ● | ○ | ◉ | ◉ |
inside-out layer stack -- bone, muscle, skin, emergent formSource present; authored stronger claim:bg_parse_layers in runtime/_hdl_build/nx_body_gen.nx, verified at emit. EXCEED with the why: read from Infinigen source, their creatures are SINGLE-SURFACE; our fat parameter attenuates muscle relief through real nested layers so lean shows the abs and heavy hides them from the SAME muscle layer, A/B proven. [infinigen23] Honest: inner layers are still coarse envelopes, not named muscles. Adoption: REGISTERED-DARK — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked); only referenced by nx_creatorfit -- a mention, not a run.Pros Source symbol found; stronger capability claim is authored; ahead of Unreal Engine 5, Godot, Infinigen, HoudiniCons not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked); only referenced by nx_creatorfit -- a mention, not a run | ◉ | ○ | ○ | ○ | ◐ |
canon computed from rules -- sex, build, muscularity, seedMeasured:bp_load_dimorph exists in runtime/_hdl_build/nx_body_proc.nx, verified at emit. The Infinigen pattern (generators plus constraints plus search): dimorphism functions plus seeded noise emit the canon the body factory reads; one rule engine for male, female, anime, individuals. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: RUN-BY:fork:nx_aesthetic_canon_gate — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of Unreal Engine 5, Godot, Houdini; fully adopted on the estate ladderCons behind Infinigen (leads) | ● | ◐ | ○ | ◉ | ◐ |
face features and digits emitted by ruleMeasured:emitDigitP exists in runtime/_hdl_build/nx_body_proc.nx, verified at emit. Nose, lips, chin, brow, ear, eye, finger and toe emitters. Honest: crude against MetaHuman-class sculpts [metahuman] [metahuman-dna] -- sculpt fidelity is the headline gap, and the fix is canon-mode search under the panel, never hand-modelling. Adoption: RUN-BY:fork:nx_aesthetic_canon_gate — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of Unreal Engine 5, Godot; fully adopted on the estate ladderCons behind Infinigen (leads), Houdini (leads) | ● | ◐ | ○ | ◉ | ◉ |
a second species through the identical canon engineDARK —runtime/_hdl_build/nx_body_proc.nx EXISTS but does not declare bp_species_canon: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. Infinigen ships five genome classes; we have proven one species canon. Contract: a horse canon through the same pipeline turns the species-agnostic claim into a measurement. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Infinigen (leads), Houdini (leads); open contract; expected source declaration not established | ○ | ○ | ○ | ◉ | ◉ |
seed-parametric creatures -- quadruped, bird, fish plansMeasured:creaturegen_build_plan exists in runtime/nx_creaturegen.nx, verified at emit. Sovereign Infinigen: the layered-anatomy method generalized to animal body plans with seeded proportions; the skin emerges as the smooth-union envelope plus dermal offset. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: LIB-WIRED importers=10 nonval=3 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of Unreal Engine 5, Godot, Houdini; fully adopted on the estate ladderCons behind Infinigen (leads) | ● | ○ | ○ | ◉ | ◐ |
interpolatable part-genome across creature typesDARK —runtime/nx_creaturegen.nx EXISTS but does not declare cg_genome_parts: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. Infinigen's genome joins body, head, leg, foot, tail and horn parts and interpolates BETWEEN types; ours has fixed plans. The expansion contract is typed parts with legal interpolation -- their capability, our layers underneath. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Infinigen (leads), Houdini; open contract; expected source declaration not established | ○ | ○ | ○ | ◉ | ◐ |
| ANATOMY | |||||
eight-system internal anatomy rendered inside and outSource present; authored stronger claim:as_load_sites in runtime/_hdl_build/nx_anat_sov.nx, verified at emit. EXCEED with the why: MetaHuman and Infinigen render no internals at all; the bar here is a medical atlas. Peel, explode and stereo proven against the CUDA oracle on the same invariants. Honest: organ shapes vs atlas still open. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: REGISTERED-DARK — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked).Pros Source symbol found; stronger capability claim is authored; ahead of Unreal Engine 5, Godot, Infinigen, HoudiniCons not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked) | ◉ | ○ | ○ | ○ | ◐ |
articulated rig with deterministic physics on the real skeletonMeasured:rg_fk_y exists in runtime/_hdl_build/nx_anat_rig.nx, verified at emit. FK joint tree plus damped-pendulum integrator on the 662k-tri BodyParts3D skeleton; replay bit-identical on all frame checksums. Honest gaps: sagittal hinges only, no constrained rigid-body solver. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: SOURCE-ONLY — PARTIAL: source exists, never compiled: /api/build it.Pros Implementing source symbol found; runtime acceptance separateCons behind Unreal Engine 5 (leads), Godot (leads), Houdini (leads); not adopted yet: source exists, never compiled: /api/build it | ● | ◉ | ◉ | ● | ◉ |
| MOTION | |||||
real mocap decoded and retargetedMeasured:m_joints exists in runtime/nx_bvhmotion.nx, verified at emit. BVH and VMD decoded sovereignly, retargeted at 30fps; rivals ship authored clip libraries. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: REGISTERED-DARK — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked).Pros Implementing source symbol found; runtime acceptance separate; ahead of InfinigenCons behind Unreal Engine 5 (leads); not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked) | ● | ◉ | ● | ○ | ● |
general secondary dynamics -- jiggle, recoil, hair, clothMeasured:sd_step exists in runtime/nx_softdyn.nx, verified at emit. Spring-damper with the stability envelope ENFORCED inside the solver so an unstable config cannot be requested; gate 7 of 7 with an anti-vacuity tooth. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: LIB-WIRED importers=8 nonval=6 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of Infinigen; fully adopted on the estate ladderCons behind Unreal Engine 5 (leads), Houdini (leads) | ● | ◉ | ● | ○ | ◉ |
one damped-oscillator identifier shared by the solver AND by videoMeasured:rd_analyze exists in runtime/nx_ringdown_lib.nx, verified at emit. SHIPPED 2026-08-25. Until today every fidelity claim on /world/beach was PLANT-against-CITED-BAND: nx_dyna_bind_skin rings the XPBD cage down and compares fn/zeta to gamefeel_oracle.conf. Sound for the plant, silent about the render -- correct fn and zeta do not imply correct amplitude, phase to footfall, or mode shape. The estimator was EXTRACTED VERBATIM out of nx_dyna_bind_skin into a lib so our own plant and real footage cannot be measured by two different rulers; the ringdown consumer was rebuilt and nx_behaveprobe read IDENTICAL first_diff_off=-1. It carries a real third state (RD_UNOBS_NO_SAMPLES / NO_CROSSING / NO_PERIOD / NO_DECAY) so a zero frequency can never be confused with an unmeasurable one, and it keeps the incumbent's DECLARED log2 imprecision rather than silently improving it. nx_dynaoracle_gate 18/18 GREEN exit 0 against an ANALYTIC fixture: a mass-spring-damper the gate integrates itself, so w0=sqrt(k/m) and zeta=c/(2 sqrt(km)) are known in closed form and the teeth ask whether the estimator RECOVERS numbers it was never told -- measured zeta 71 permil against a true 79 inside the DERIVED tolerance of 10, period 100 samples against a true 99 inside the DERIVED sampling bound of 11. Both tolerances are derived from declared imprecision, never picked. Bite-proven: a RIGID signal must refuse with a named third state, and a stiffer spring must read as a higher frequency, so a constant-returning stub fails. [physrig] [softsmpl] Adoption: LIB-WIRED importers=3 nonval=2 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of Unreal Engine 5, Godot, Infinigen, Houdini; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth) | ● | ○ | ○ | ○ | ○ |
coarse-to-fine optical flow -- large-displacement tissue trackingMeasured:nx_motion_mean_flow_pyr exists in runtime/nx_motion.nx, verified at emit. SHIPPED 2026-08-25 and it closed a MEASURED blindness. nx_motion.nx has carried Lucas-Kanade since 2026-05 with exactly ONE consumer -- its own test (corpus_complete=1) -- i.e. BUILT+UNWIRED, and nx_dynaoracle is its first production consumer. Its own header declares the small-motion assumption, and the first real subject violated it: on a fixture with a 14-pixel oscillation at 2 Hz and 30 fps (peak 5.9 px/frame) it recovered about 1 px of 14 and the consumer correctly reported UNOBSERVABLE. Tuning the fixture down would have been fitting the test to the tool, so the tool was fixed with the standard remedy. The pyramid COMPOSES the incumbent rather than adding a second flow estimator, and the same edit removed a genuine hot-loop allocation: sobel, convolve, temporal-gradient and LK all gained scratch-reusing _into variants with the allocating forms DELEGATING, so every existing caller is unchanged BY CONSTRUCTION -- proven, not asserted, by nx_behaveprobe on a real consumer reading IDENTICAL/STRONG. Measured after: coverage 214 to 1000 permil on the same clip. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: LIB-WIRED importers=3 nonval=1 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of Infinigen; fully adopted on the estate ladderCons behind Unreal Engine 5 (leads), Houdini (leads) | ● | ◉ | ● | ○ | ◉ |
METRIC SCALE for video measurement -- the declared limit of the video oracleDARK —runtime/nx_dynaoracle.nx EXISTS but does not declare do_metric_scale: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. THE CONTRACT THAT CLOSES nx_dynaoracle's OWN DECLARED LIMIT, named here rather than left as a caveat in a header. The oracle states before any number that amplitude in MILLIMETRES is UNMEASURABLE from an uncalibrated clip and prints that string instead of a figure, so tissue_amp_walk_mm 32..52 and tissue_amp_run_mm 110..194 remain CITED and unmeasured against our render. Everything the oracle DOES report is dimensionless BY CONSTRUCTION -- fn, zeta, drive frequency, drive-to-response phase lag, and amplitude as a RATIO of a body landmark -- which is exactly why those survive a clip with no calibration, no known camera and no known subject size. The missing input is a metric body frame. Multi-HMR 2 recovers multi-person whole-body mesh from a single image in the ANNY representation, and ANNY is an Apache-2.0 parametric human covering infants to elders with interpretable controls [anny] [multihmr2] -- a permissive licence, unlike SMPL. THE RULE THIS ROW INHERITS: ORACLE ONLY, NEVER PRODUCT, the same rule this board already applies to BodyParts3D, and the same reason the VLM lens on the referee panel abstains rather than acquits -- a learned model may CALIBRATE our instrument and may never ship inside it. Contract: recover a metric body frame per clip, publish the mm-per-pixel it implies WITH the uncertainty that recovery carries, and convert the oracle's ratios into millimetres so the two cited amplitude bands become MEASURABLE rather than merely citable. Until it lands, amp_mm stays a named refusal and every amplitude claim on this board is a ratio.Pros none measured yetCons open contract; expected source declaration not established | ○ | ○ | ○ | ○ | ○ |
DAMPING identified from video -- the parameter the field does not fitSource present; authored stronger claim:do_crop in runtime/nx_dynaoracle.nx, verified at emit. EXCEED with the why, and the estate declared this gap itself before filling it: gamefeel_oracle.conf's own closing note records that the 2025-26 differentiable-simulation literature (MonoPhysics, PhysCoRe, EgoPhys) recovers Young's modulus, Poisson ratio and yield stress and NEVER damping, never on human soft tissue. PhysRig fits material prototypes and likewise not damping [physrig]; SoftSMPL learns soft-tissue dynamics from 4D scans, not from monocular video [softsmpl]. Damping is the parameter that governs jiggle decay. nx_dynaoracle measures fn, zeta, drive frequency and drive-to-response phase lag from a frame stack through the SAME estimator as the plant, cancelling camera and body motion by differencing a tissue ROI against a RIGID reference ROI. The noise deadband is not a tunable and there is no conf row for it: it is MEASURED per clip from the reference patch's own top and bottom halves, which share camera and body motion exactly, so their differential is measurement noise BY CONSTRUCTION -- a negative control the caller cannot forget to supply and cannot fudge. END-TO-END MEASURED on a synthetic video fixture with analytically known ground truth: coverage 1000 permil (149/149 pairs), drive frequency 500 mHz recovered EXACTLY, damped fn 2142 mHz against a true 1960 (+9.3 percent), zeta 245 permil against a true 200 (+22.5 percent). DECLARED LIMITS, stated before the numbers rather than discovered after: amplitude in MILLIMETRES is UNMEASURABLE from an uncalibrated clip and the organ prints that string rather than a figure, so tissue_amp_walk_mm stays CITED and unmeasured here; closing it needs a metric parametric-mesh recovery (Multi-HMR/Anny class) used as an ORACLE ONLY, never as product, the rule this board already applies to BodyParts3D [multihmr2] [anny]. Integrating velocity cannot recover the INITIAL displacement, so the first lobe carries an unrecoverable DC offset and is skipped before the log decrement -- frequency is timing and survives an offset, damping is amplitude and does not. ffmpeg is used as a pure DECODER and that break-glass is declared in source with the sovereign VP9/H.264 lane named as its replacement. Adoption: REGISTERED-DARK — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked).Pros Source symbol found; stronger capability claim is authored; ahead of Unreal Engine 5, Godot, Infinigen, HoudiniCons not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked) | ◉ | ○ | ○ | ○ | ○ |
dynamics bound to skin verticesMeasured:dbs_displace_free exists in runtime/nx_dyna_bind_skin.nx, verified at emit. RETARGETED 2026-09-01 by the board owner, which is exactly what the previous note on this row asked for ("Retargeting this contract is the board owner call"). The old contract runtime/nx_nxa_dyna.nx:dyna_bind_skin COULD NEVER FLIP, and it held GR13 open at ranker slot #8 for a capability that had shipped. MEASURED 2026-09-01, full population, corpus_complete=1 over buildroot/runtime: func dyna_bind_skin( is declared EXACTLY ONCE and it is at runtime/nx_softbind.nx:238, whose own comment reads "Named for the R7/GR13 watch contract: this IS the binding the boards declare" -- it ticks three asset-declared soft-tissue REGION points (belly, glute, thigh) on the one solver and is CALLED from sb_tick_fn, so it is wired, not orphaned. That is NOT this rung's done-rule, which asks for skin VERTICES with collision against the body; the per-vertex half is func dbs_displace_free( at runtime/nx_dyna_bind_skin.nx:203 (matches=1, corpus_complete=1) -- the symbol this row's own previous note already named as the shipped one, and the symbol procgen.matrix and softbody.matrix already carry LIVE. This row now watches THAT, so three boards stop disagreeing about one capability. DECLARED LIMITS, not hidden: (a) the flip is the receipt and the gate is the proof, and nx_dyna_bind_skin_gate is on the graphics.gates roster with its re-measured verdict; (b) the REGION-point half above is a separate capability that no row on this board carries. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: REGISTERED-DARK — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked); only referenced by nx_dyna_bind_skin_gate -- a mention, not a run.Pros Implementing source symbol found; runtime acceptance separateCons behind Unreal Engine 5 (leads), Godot (leads), Houdini (leads); not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked); only referenced by nx_dyna_bind_skin_gate -- a mention, not a run | ● | ◉ | ◉ | ● | ◉ |
hair secondary motion -- spring-bone chain on the ONE solver, measuredMeasured:hd_ringdown exists in runtime/nx_hairdyn_lib.nx, verified at emit. SHIPPED 2026-08-25, gate GREEN 11/11 exit 0 and bite-proven both ways (valid_mutants=1 not_reached=0 killed=1, restore byte-identical, mutant_live=0). Hair is driven as a CHAIN on the same solver as the tissue -- sd_step_axq stepped per segment, K from sb_k_sym, C from sb_c_for_zeta, plant identified by rd_analyze from nx_ringdown_lib. One solver, one derivation pair, one estimator, all COMPOSED: a second spring integrator would have been the duplicate-ruler defect. sd_step_ax now DELEGATES to a Q8-anchor form because a chain needs sub-unit anchors -- on the model-unit entry point every hop floors the parent's motion away -- and every existing caller stays bit-identical by construction. Measured plant fn 4285 mHz, zeta 22 permil; chain propagation proven by phase lag 83 permil at 2 segments rising to 916 at 12. 60 Hz tick DERIVED from SB_MHZ_TICK_SQ, never written. HONEST: requested zeta is an UPPER BOUND in this solver, not an equality -- sd_step's damping term truncates, so below a velocity threshold no damping applies; the author's first published figure of 66 permil came from linear arithmetic and the RINGDOWN said 22, so the gate asserts monotonicity rather than equality and both places were corrected. [springbones] Adoption: LIB-WIRED importers=77 nonval=73 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of Infinigen; fully adopted on the estate ladderCons behind Unreal Engine 5 (leads), Houdini (leads) | ● | ◉ | ● | ○ | ◉ |
| RENDER | |||||
scalp groom ANATOMY REFEREE -- nose-arbitered facing, density 4800 floor, length band, conformance, face-cone-clear; negative control is the REAL banked defectMeasured:gg_stats exists in runtime/nx_nxa_groom_gate.nx, verified at emit. v3 GREEN 10/10 (2026-08-27). Facing basis is the RIG CONVENTION anterior-negative cross-checked by mesh density (three geometry heuristics gave three answers; the organ REFUSES on disagreement). Surface tooth is NN-to-mesh with the bound derived per asset from median inter-vert spacing -- its first bin-based form was refuted by its own control (real roots read 764 permil interior). Two banked REAL defects are the negative fixtures (a face-side groom: face-cone 540 permil; a chord-blend interior fill: far 613 permil), and the gate refused three intermediate placements pre-install before root placement was rewritten to the PUBLISHED method (2026-08-27): area-weighted barycentric sampling of the scalp TRIANGLES with face-normal launch, per Blender particle hair and the Infinigen creature hair chain -- every root on a face by construction, no synthesis stage, far 0 permil, uniform coverage. The NN bound itself was recalibrated on measured data after refusing a correct groom (triangle interiors are ~circumradius from every vertex): legit max 941 vs interior-fill median 1,397, bound 5x spacing = 1,130. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: GATE:LIVE trial=- — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Unreal Engine 5 (leads), Houdini (leads) | ● | ◉ | ● | ● | ◉ |
| MOTION | |||||
hair GEOMETRY on the cast -- the renderer rung both hair legs needMeasured:gpe_hair_geometry exists in runtime/nx_game_page_emit.nx, verified at emit. FLIPPED 2026-08-30 (receipt: gpe_hair_geometry live in nx_game_page_emit.nx consuming HVRT+HTRI as Catmull-Rom ribbon cards plus the nxHairTick runtime spring chain, pinned by nx_gpe_hair_geometry_gate; the visible remaining gap is SHADING owned by hr_marschner_lobe -- the flat genome lock colour is why live hair reads gold-static). Historical 08-25 diagnosis, superseded: THE ASSET WAS NEVER THE BLOCKER. All four corpus FBX carry groomed hair (dark_witch 22386v/38016t, paladin 19252v/29914t, dark_knight 16583v/28416t, toon3d8 309913v) and nx_nxa_hair extracts them correctly -- it was built for the FIRST TIME today and a null-deref on its first ever run was fixed. The blocker is the CONSUMER: HVRT appears exactly 3 times in 23,221 sources (corpus_complete=1) and ALL THREE are inside nx_nxa_hair.nx itself, so NOTHING READS IT. Every site page is emitted by an organ, so no page can render hair geometry; on /world/beach hair is still a GLSL colour override at nx_game_page_emit.nx:480 over a crown-and-back-of-skull band. The dynamics leg lands on the same wall: nx_hairdyn_lib is imported only by its gate, sb_tick_fn still drives SB_HAIR as ONE point, and the renderer consumes ONE offset where a chain emits N. ★★★★★★HAIR IS ONE DEFECT WEARING TWO MASKS -- an ingest with no consumer and a chain solver with no renderer -- AND BOTH LEGS CONVERGE ON A RENDERER RUNG, NOT A PHYSICS ONE. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: REGISTRY-UNCERTAIN — PARTIAL: registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion.Pros Implementing source symbol found; runtime acceptance separate; ahead of InfinigenCons behind Unreal Engine 5 (leads), Houdini (leads); not adopted yet: registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion | ● | ◉ | ● | ○ | ◉ |
FACS action-unit expression basisDARK —runtime/nx_faceanat.nx EXISTS but does not declare fa_facs_au: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. Residue named at the 2026-08-30 flip of fa_expr_morph below, kept as a narrower watch so the flipped cell cannot overclaim: fa_expr_morph is the ONE named entry point dispatching the seven MORF channels (width, length, eyes, nose, cheek, lips, jaw) in the faceanat percent convention -- morphs an expression can drive, not a FACS action-unit basis. Contract: AU-indexed expression channels over the same blended field, graded readable by the referee panel (the GR12 done-rule). One contract, two pages: the charsim board carries the same watch. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot, Houdini (leads); open contract; expected source declaration not established | ○ | ◉ | ● | ○ | ◉ |
FACS facial expression morphsMeasured:fa_expr_morph exists in runtime/nx_faceanat.nx, verified at emit. FLIPPED 2026-08-30 (adoption seat; receipt: func fa_expr_morph( at buildroot/runtime/nx_faceanat.nx:208 -- the shared channel vocabulary over the seven morph functions, an unknown channel returns -1 and never a silent no-op; comparewatch plane graphics_fa_expr_morph LANDED; the flip is the receipt, and the FACS action-unit residue is carried by the narrower watch row above rather than hidden inside this cell). Prior note: Identity morphs exist as anthropometric axes; expression axes do not. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: LIB-WIRED importers=26 nonval=12 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of Infinigen; fully adopted on the estate ladderCons behind Unreal Engine 5 (leads), Houdini (leads) | ● | ◉ | ● | ○ | ◉ |
| SPLAT | |||||
sovereign integer 3D gaussian splatting, anisotropic, depth-sortedMeasured:gs_render_aniso exists in runtime/nx_gsplat.nx, verified at emit. The frontier photoreal representation (GaussianAvatars, HumanSplat lineage); all integer with an exp-LUT, front-to-back compositing. [kerbl3dgs] Adoption: LIB-WIRED importers=20 nonval=6 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of Unreal Engine 5, Godot, Infinigen, Houdini; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth) | ● | ◐ | ○ | ○ | ◐ |
differentiable color fit step toward trained splatsMeasured:gs_color_grad_step exists in runtime/nx_gsplat.nx, verified at emit. 3DGS at SOTA is a TRAINED fit, not just a splat renderer; the gradient step is the first tooth of that loop. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: LIB-WIRED importers=20 nonval=6 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of Unreal Engine 5, Godot, Infinigen, Houdini; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth) | ● | ◐ | ○ | ○ | ◐ |
hybrid -- outside splat, inside parametric meshMeasured:gb_bind exists in runtime/_hdl_build/nx_gsplatbind.nx, verified at emit. The mid-2026 industry architecture (SplattingAvatar class), operator-ordered 2026-07-27: splats bound barycentrically to triangles, the mesh deforms and the splats follow. [splattingavatar24] Built; the leap organ composes it. Adoption: REGISTERED-DARK — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked).Pros Implementing source symbol found; runtime acceptance separate; ahead of Unreal Engine 5, Godot, Infinigen, HoudiniCons not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked) | ● | ○ | ○ | ○ | ○ |
genome-driven splat avatar -- THE LEAPDARK —runtime/nx_gsplat.nx EXISTS but does not declare gs_avatar_genome: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. Operator-resequenced to the front. v1 ran end to end (94,416 tris to 87,036 gaussians to portrait) and the operator's verdict on v1 stands recorded; the work relocated to canon-mode search under the judge, composing the gated face pipeline, and hair volume (85 percent of the industry reference's polygon budget is hair). Same watch as the charsim board -- one contract, two pages flip. THE BAR BECAME DATA 2026-08-19 (GR2 leg 1): under the admitted referee panel on one figure-crop framing (nx_charjudge_census figcrop, list knowledge/cjc_gr2_framing.list) the ring figure reads skin 7 / percept -112 BAD, the v1 splat skin 118 / percept 123 BAD, the six banked WebGL rig crops skin 279-661 / percept 210-228 GOOD, Mackenzie 624 / 281 -- so v2 is done when its crop reads GOOD with percept above 228 and skin at or above 279 on the same conf and receipt. [gaussianavatars24] [infinihuman25]Pros none measured yetCons open contract; expected source declaration not established | ○ | ○ | ○ | ○ | ○ |
4D animated splats driven by the rigDARK —runtime/nx_gsplat4d_lib.nx EXISTS but does not declare g4_avatar_rig: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. One animation system, two render fidelities -- an architecture MetaHuman cannot have because its animation is welded to its mesh.Pros none measured yetCons open contract; expected source declaration not established | ○ | ○ | ○ | ○ | ○ |
| WORLD | |||||
five live wasm worlds, sovereign-only renderMeasured:wc_mob_mesh exists in runtime/nx_wasm_craft.nx, verified at emit. /world/craft, shore, vale, isle, crags: NishiLang compiled to wasm, CPU raycaster, no GPU shader in the path; skinned genome-distinct characters in-world, gate 71 of 71 with eyes, wrap-SSS and the hourglass rework all toothed. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: LIB-WIRED importers=41 nonval=25 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of Infinigen, Houdini; fully adopted on the estate ladderCons behind Unreal Engine 5 (leads), Godot (leads) | ● | ◉ | ◉ | ○ | ○ |
procedural worldgen with real lighting and hydrologyMeasured:wp_shade exists in runtime/nx_worldpipe.nx, verified at emit. Sun-facing diffuse off the real heightfield, gate-proven in both directions; hydrology traces downhill to the sea. Honest: coherent but thin against Minecraft-class staging. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: LIB-WIRED importers=8 nonval=3 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Infinigen (leads), Houdini (leads) | ● | ● | ● | ◉ | ◉ |
indoor rooms by constraint solve -- the Infinigen 2.0 expansionMeasured:ig_room_solve exists in runtime/nx_indoorgen_lib.nx, verified at emit. FLIPPED 2026-08-30 (adoption seat; receipt: func ig_room_solve( at buildroot/runtime/nx_indoorgen_lib.nx:193; comparewatch plane graphics_ig_room_solve LANDED; the note below already declared it SHIPPED while the symbol field still spelled _ABSENT_ -- a cell that could never flip as written; the flip is the receipt, nx_indoorgen_gate 8/8 and craft-gate T72-T74 are the proof). Prior note: SHIPPED 2026-08-17/18, their 2.0 alpha frontier through OUR engine: nx_indoorgen composes the estate part-solver (envelope, noclip, door-clearance, wall-hug rows), quantizes to voxel cells under a pre-declared accept rule and REFUSES an unsatisfiable set (neg-control toothed) rather than overlap furniture; SITING is solved too (flattest-anchor search per site cell + foundation skirt -- the rolled-anchor stamp it replaced rejected every vale cell, measured before fixed), and the engine stamps the solved layouts -- cottages on craft, driftwood huts on the shore, a 4-building stone-cot village in the vale -- proven by nx_indoorgen_gate 8/8 and craft-gate 74/74 (T72 structure, T73 seam determinism, T74 density-as-data). SHIPPED LIVE 2026-08-18: reconciliation done (the 21 AHEAD strings were superseded rewords of nx_parse diagnostics, proven by fixture), nx_compile_wat promoted at 364 pages, and nx_craft_emit shipped all 5 recipe worlds (craft/beach/vale/isle/crags) vm-gate 5/5, page_verify GREEN; the module-owned page-sizing MECHANISM is shipped in both backends (wat_mem_pages reads a declared static, guards bite-proven 0/70000/5) but the ENGINE declaration is blocked by an nxasm gap (nonzero-init writable statics unsupported both lanes -- debt filed with the unblock spec); until it lands the vm-gate FITS tooth is the drift catcher, and it is the tooth that caught the original 192 drift. [infinigen-indoors24] [infinigen-flying26] [infinigen-docs] Adoption: LIB-WIRED importers=2 nonval=1 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Infinigen (leads), Houdini (leads) | ● | ● | ● | ◉ | ◉ |
ground-truth dataset emission -- RGB, depth, normals, semanticsDARK —runtime/nx_world_snap.nx EXISTS but does not declare ws_groundtruth_emit: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. Infinigen's annotation pipeline is the reason its datasets exist; our renderer already holds depth and normals per pixel. Contract: an emit verb publishing aligned channels, making the estate a dataset factory. THE RIVAL BAR BECAME CONCRETE 2026-08-25, read from NAVER LABS Europe's rendered code index (the raw fetch carried it; an earlier claim that this page was a JS shell was MY OWN keyword-filtered grep failing, not the fetch): Syn4D is a multiview synthetic 4D dataset of dynamic scenes covering camera pose, depth, dynamic 3D reconstruction, 2D and 3D tracking, human pose and novel-view synthesis [syn4d], and Anny-One is a large-scale synthetic dataset for 3D human understanding with multi-person configurations, multi-view setups at extreme camera positions, diverse FOVs, indoor scenes and full ANNY annotations [anny]. PhysRig reached the same wall from the modelling side and had to BUILD a synthetic corpus because, in its own words, evaluating physics-based skinning is hard for lack of suitable benchmark datasets [physrig]. THE EXCEED IS STRUCTURAL, NOT A BIGGER NUMBER: every one of those ships a FIXED corpus, while ours would emit from the genome and the recipe planes -- parametrically unbounded, seed-addressable and regenerable rather than downloaded, with the ground truth exact because the generator knows it rather than estimated because a labeller guessed it. That is the Infinigen-2 procgen integration the operator asked for, arriving AFTER the output is nailed down and not before: the measure-then-generate order, with nx_dynaoracle and the referee panel as the rulers that decide when the output IS nailed down. THE 2026 BAR IS NOW CITED AND PINNED: Syn4D [syn4d] is a multiview synthetic 4D dataset whose procedural multi-camera pipeline jointly emits RGB, depth, instance segmentation, optical flow, dense long-range 3D tracks and parametric SMPL-X pose, all spatially and temporally aligned, so any pixel unprojects to 3D at any frame and reprojects into any other camera. That cross-view consistency is the part our single-camera snap does not yet claim, and it is the honest distance this contract has to close rather than a box we can tick by emitting channels alone.Pros none measured yetCons behind Unreal Engine 5, Infinigen (leads), Houdini; open contract; expected source declaration not established | ○ | ● | ○ | ◉ | ◐ |
native-resolution delivery on the visitor's GPUDARK —runtime/nx_craft_emit.nx EXISTS but does not declare cw_webgpu_compute: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. A CPU raycaster cannot reach 1200x750 in 24ms on typical hardware (measured in the 2026-08-17 delivery fixes); contract: OUR raycaster emitted as a compute shader -- the visitor's GPU executes our code, sovereignty kept. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot (leads); open contract; expected source declaration not established | ○ | ◉ | ◉ | ○ | ○ |
| REFEREE | |||||
honest body ruler with an anti-Goodhart detail judgeSource present; authored stronger claim:bb_detail_score in runtime/_hdl_build/nx_bodybench.nx, verified at emit. EXCEED with the why: headline = MIN of front, side, shape and DETAIL; the detail judge is IoU-in-variance-space, non-gameable by noise (the FBM Goodhart scores 0) and subdivision-proof (1.53M tris moved it nowhere). No rival ships an equivalent honesty instrument. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: REGISTERED-DARK — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked); only referenced by nx_bodydensity_gate -- a mention, not a run.Pros Source symbol found; stronger capability claim is authored; ahead of Unreal Engine 5, Godot, Infinigen, HoudiniCons not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked); only referenced by nx_bodydensity_gate -- a mention, not a run | ◉ | ○ | ○ | ◐ | ○ |
mechanical character-image judge with full-population censusMeasured:cjc_hw exists in runtime/_hdl_build/nx_charjudge.nx, verified at emit. corpus=169 mean=138, the number every beauty claim must beat. Proven blind in BOTH directions (clay 257, photoreal-class 0) -- which is exactly why the panel contract below gates all fidelity claims. Adoption: RUN-BY:fork:nx_charjudge_census — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of Unreal Engine 5, Godot, Infinigen, Houdini; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth) | ● | ○ | ○ | ○ | ○ |
multi-lens judge panel that can see beautyMeasured:cjc_panel exists in runtime/nx_charjudge_census.nx, verified at emit. FLIPPED 2026-08-30 (adoption seat; receipt: func cjc_panel( at buildroot/runtime/nx_charjudge_census.nx:717; comparewatch plane graphics_cjc_panel LANDED; the note below already recorded GR1 LANDED 2026-08-19 and RE-ADMITTED 2026-08-25 while the symbol field still spelled _ABSENT_ -- a cell that could never flip as written; the flip is the receipt, nx_cjc_panel_gate 19/19 and its admission receipt are the proof. Residue stays named in the note, not hidden by the flip: the VLM lens is UNOBSERVABLE and abstains, and percept is the single voting lens). Prior note: Phase 0 of past-MetaHuman. LANDED 2026-08-19 (GR1, the ranker's #1): cjc_panel in nx_charjudge_census composes the incumbents -- nx_charjudge (stats), nx_color_kovac_skin_pixel (published skin gamut over the centre box), nx_image_beauty path (symmetry), nx_percept judge (relative perceptual position vs the 10 human-confirmed commercial frames) -- forked per image, numbers always, verdicts ONLY under an admission receipt that calibrate writes at FP=0 FN=0 and binds to the exact floors (edit a floor = unadmitted by construction). CALIBRATION RUN 1 on the labelled corpus v1 (n=10: 3 BAD = the operator's green gumby crop, the recorded v1 splat verdict, a figureless world snapshot; 7 GOOD = 6 banked WebGL NXA rig captures from /world/rigview on the real GPU + the published Mackenzie render): stats good_min 70 vs bad_max 234 NOT separable (the blind ruler, measured again), sym 650 vs 1000 NOT separable -- both report-only; skin 130 vs 40 and percept 175 vs 129 SEPARABLE -- floors 85 / 152 from the margins, composite FP=0 FN=0, ADMITTED. nx_cjc_panel_gate GREEN against the deployed binary and bite-proven both ways (the green twin fails the skin lens; a moved floor un-admits; a mislabelled corpus refuses and revokes the receipt), one ruler in runtime/_hdl_build (a same-day sibling twin in runtime/ was retired reversibly). HONEST HORIZON: the VLM lens is UNOBSERVABLE -- no sovereign VLM oracle exists on the estate (tree grep + local probe 2026-08-19), the slot abstains and never acquits; n=10 is a first corpus and the second-corpus retest is owed before the floors are believed; the skin lens is framing-sensitive (central box) so verdicts compare like framings, which the GR2 done-rule already requires. SECOND-CORPUS RETEST 2026-08-19 (corpus v2 = knowledge/cjc_panel_labels.list: BAD gumby crop + v1 splat, GOOD = two PUBLISHED MetaHuman hero stills fetched sovereignly and banked, the far bar the plan names): stats 103 vs 234 and sym 488 vs 1000 again NOT separable; percept 212 vs 126 SEPARABLE with floor 152 inside the band (fp=0 fn=0, consistent with v1); but the Kovac DAYLIGHT skin rule reads the cinematically graded MetaHuman skin at 0-6 permil (centre box) and 0-4 (figure mask), so floor_skin=85 fails both GOOD refs (fn=2) and the retest was RECORDED as having REFUSED and revoked the receipt. -- CORRECTED 2026-08-25, AND BOTH HEADLINE CLAIMS OF THAT RETEST WERE WRONG. (1) IT NEVER REVOKED ANYTHING: it ran in `panel` mode and ONLY `calibrate` writes a receipt, so knowledge/status/cjc_panel_admission.txt still read admitted=1 at floors none,85,none,152 six days later and the live panel printed admission=ADMITTED -- a referee actively calling BOTH published MetaHuman references BAD while this board recorded it as safely revoked. The panel's own law that editing a floor un-admits it fires on a floor EDIT; a corpus that REFUTES the floors revokes nothing. ★★★★★★A REVOCATION NOBODY PERFORMED DID NOT HAPPEN, AND A NOTE ASSERTING A SAFETY STATE THE ARTIFACT DOES NOT CARRY IS WORSE THAN NO NOTE -- it is the stale-RED-gate class inverted, a stale GREEN nobody re-measured. (2) GRADING WAS NOT THE MECHANISM AND THE NAMED REMEDY WAS BUILT AND REFUTED: two published chroma-only rules with no luminance term (Kovac Cb/Cr with Y dropped; Chai and Ngan 1999) were measured over the FULL 3x3 rule-by-region grid on the union -- NINE OF NINE NOT SEPARABLE, good_min=0 in every cell, and dropping the luminance gate never raised good_min, it only inflated bad_max (centre 52 to 848/903). The cause is ONE CORPUS ROW: cmp_graphics_metahuman_ref0_rgb.png reads exactly 0 skin in all nine cells and nx_charjudge's own face axis reads cjp_face=0 -- no face in a GOOD row of a CHARACTER corpus. Its banked original measures identically, so the conversion is faithful, and cjc_ref_mackenzie.png (also an external published render) scores 624, so 'published external reference' is not the discriminator. RE-ADMITTED 2026-08-25 on the UNION v1 u v2 (knowledge/cjc_panel_labels_union.list, n=13 = 9 GOOD / 4 BAD, partition reconciled): the skin floor is REMOVED and skin is REPORT-ONLY, percept is the sole voting lens at floor 152 (173 vs 129, margin 44, fp=0 fn=0), and the receipt is rewritten BY THE CALIBRATE PATH now binding skin_lens and skin_rule as well as the floors, so changing region or ruler un-admits by construction. nx_cjc_panel_gate 13/13 GREEN exit 0 against the DEPLOYED binary and bite-proven both ways (T4 wrong floors refuse / matching admit with the green twin failing the skin lens; T5 swapped labels REFUSE; T10 a refused recalibration rewrites the receipt to admitted=0; new T11 matching floors on another region or ruler do NOT admit; new T12 an unrecognised skin_rule REFUSES by name). HONEST LIMITS: percept is now the ONLY voting lens -- a single-lens referee until a second earns a floor, a real reduction in strength stated rather than hidden; the VLM lens stays UNOBSERVABLE and abstains, never acquits; n=13 measures separability, not generalisation; and excluding the two MetaHuman rows the skin lens WOULD separate at region=fig (GOOD 373-833 vs BAD 6-286), so a second voting lens is recoverable if the corpus owner replaces ref0 -- the rows were deliberately NOT relabelled or dropped, because dropping the rows that refute a floor is exactly the relabelling this corpus forbids. [metahuman-ref0] [metahuman-ref1] Adoption: RUN-BY:fork:nx_cjc_panel_gate — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of Unreal Engine 5, Godot, Infinigen, Houdini; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth) | ● | ○ | ○ | ○ | ○ |
head-to-head vs-ruler -- parity is not betterMeasured:vs_verdict exists in runtime/_hdl_build/nx_vsbench.nx, verified at emit. beat_index 0 permil over 10 cross-class cells. This page therefore claims breadth and mechanisms, never perceptual wins. Adoption: REGISTERED-DARK — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked).Pros Implementing source symbol found; runtime acceptance separate; ahead of Unreal Engine 5, Godot, Infinigen, HoudiniCons not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked) | ● | ○ | ○ | ○ | ○ |
| RESEARCH | |||||
9762-paper citable graphics index, mirrored non-rottablyMeasured:gx_scan exists in runtime/nx_gfxpapers_extract.nx, verified at emit. SIGGRAPH, EG, I3D, EGSR, SCA, SGP, HPG, PG 2000-2026 from the Ke-Sen mirror in knowledge/library (partition anchors 9984 reconciles); any row here can cite a named paper instead of an adjective. [kesen-mirror] Adoption: REGISTERED-DARK — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked).Pros Implementing source symbol found; runtime acceptance separate; ahead of Unreal Engine 5, Godot, Infinigen, HoudiniCons not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked) | ● | ○ | ○ | ○ | ○ |
SIGGRAPH Advances course archive mirrored beside itMeasured:gp_collect2 exists in runtime/nx_gfxpapers_ingest.nx, verified at emit. LANDED 2026-08-17: gp_collect2 plus gp_root_of turned the ingest into a second-root mirror -- anchor-driven collection, rebuild host derived from the index url, subdir-safe names; the Advances crawl banked 25 of 26 reachable pages, partition RECONCILED, the one dead link already rotted at capture. Open tail: the extractor citing adv_ pages into the paper index. [advances-archive] [karis2021] Adoption: RUN-BY:clock — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of Unreal Engine 5, Godot, Infinigen, Houdini; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth) | ● | ○ | ○ | ○ | ○ |
| DELIVERY | |||||
GPU last mile in third-party browsers (WebGL2 tier LIVE)Measured:gpe_build exists in runtime/nx_game_page_emit.nx, verified at emit. LIVE 2026-08-18 BY OPERATOR ORDER: GPE_EMIT_WEBGL flipped on -- every /world page now renders on the visitor's GPU at device resolution (voxel world uploaded as a 3D R8UI texture, shader passes for world plus skinned characters) with the sovereign NishiLang-wasm rasteriser ALWAYS compiled in as automatic fallback and ?gl=0 lever. The 08-16 sovereign-only order still binds the charsim QUALITY ladder: quality is measured on the sovereign path; the GPU is the delivery cable. Godot leads on first-class web export; UE5 web export is community-tier. [webgpu-status26] Adoption: REGISTRY-UNCERTAIN — PARTIAL: registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion.Pros Implementing source symbol found; runtime acceptance separate; ahead of Infinigen, HoudiniCons behind Godot (leads); not adopted yet: registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion | ● | ● | ◉ | ○ | ○ |
WebGPU tier -- WGSL world pass over the same organ stateMeasured:gpe_emit_webgpu exists in runtime/nx_game_page_emit.nx, verified at emit. FLIPPED 2026-08-30 (adoption seat; receipt: func gpe_emit_webgpu( at buildroot/runtime/nx_game_page_emit.nx:125; comparewatch plane graphics_gpe_emit_webgpu LANDED; the note below already recorded STAGE A LANDED 2026-08-18 while the symbol field still spelled _ABSENT_ -- a cell that could never flip as written; the flip is the receipt, the real-GPU verification over the CDP wire is the proof. Residue stays a separate contract, not hidden here: WebGPU as the DEFAULT tier is gameengine GE1, _ABSENT_:gpe_webgpu_default). Prior note: STAGE A LANDED 2026-08-18, LEVERED ?gpu=1: the organ emits the full WGSL world pass (line-faithful FSH transliteration incl. the fragment-origin Y-flip) plus WebGPU init/draw with a BOOT-STALL CEILING -- both headless Chromiums on the dev box measured wedging INSIDE requestAdapter/createShaderModule with no rejection, so the race caps a visitor's cost at GPU_BOOT_MS and the ladder falls through to GL/CPU. Verified on REAL GPU via the CDP wire: page doors gpu:1/fail:null/px-dG, frame judged RENDERED (ink 870 permil, 103 distinct colors), pixel-faithful to the GL tier. DEFAULT stays GL until the NPC and eye passes reach parity (stage B) -- characters are GL programs today, and a default that loses the girls would be a regression wearing a newer API. WebGPU ~82 percent global; no RT extension in spec (web RT stays compute-shader); subgroups behind flags, bindless in proposal. [webgpu-status26] [webgpu-roadmap26] Adoption: REGISTRY-UNCERTAIN — PARTIAL: registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion.Pros Implementing source symbol found; runtime acceptance separate; ahead of Infinigen, HoudiniCons not adopted yet: registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion | ● | ● | ● | ○ | ○ |
shaders compiled from NishiLang (nx_wgsl backend)Measured:wgsl_emit_module exists in runtime/nx_wgsl.nx, verified at emit. FLIPPED 2026-08-30 (adoption seat; receipt: func wgsl_emit_module( at buildroot/runtime/nx_wgsl.nx:490 -- the WGSL backend entry over the module, stage, output buffer and trace; comparewatch plane graphics_wgsl_emit_module LANDED; the note below already recorded rung 2b structured control flow landed while the symbol field still spelled _ABSENT_ -- a cell that could never flip as written; the flip is the receipt, the octane-corpus regression control named in the note is the proof). Prior note: The sovereign move: our compiler emits WGSL exactly as it emits WAT (sibling backend of nx_wasm.nx), so GPU shaders are written in NishiLang and one source of truth drives the CPU rasteriser, the WebGL/WebGPU tiers, and later the native NishiOS pipe. Rivals compile HLSL/GLSL through vendor toolchains (tint/naga); nobody compiles their own language end-to-end. [wasm3-state26] Rung 2b recovers STRUCTURED CONTROL FLOW from the CFG -- if-diamonds, while loops, loop-containing-if and nested loops -- with a recursive region reducer, on the MEASURED precondition that every loop header in the shader corpus is single-entry, refusing multi-entry and irreducible graphs by name rather than mis-shaping them [relooper-zakai11] [llvm-cfgstackify] Adoption: REGISTRY-UNCERTAIN — PARTIAL: registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion.Pros Implementing source symbol found; runtime acceptance separate; ahead of Infinigen, HoudiniCons not adopted yet: registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion | ● | ● | ● | ○ | ○ |
WORLD-ROADMAP (BEAUTY TIER): film-quality environments -- the actual Houdini-exceed laneDARK —runtime/nx_gsplat.nx EXISTS but does not declare env_beauty_scene: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. EH0, THE TIER CORRECTION (operator 2026-08-18: voxel was supposed to be ONE type, not the ceiling): the voxel worlds are the GAME tier and the rungs below improve them, but Houdini-exceed is a BEAUTY tier from the same genome/recipe planes -- full-detail SDF/mesh terrain at Infinigen scale (the estate already meshes 8.3M tris in body_gen and walks SDF-to-mesh in the skull lane; the missing piece is the TERRAIN-scale SDF world lane), rendered as gaussian-splat scenes (nx_gsplat, gate-proven) under the path-GI contract above, with ground-truth emission making it a dataset factory. THE 2026 CEILING IS LEARNED: SIGGRAPH 2026's top-attention terrain work is a diffusion generator with procedural-noise properties (seed-consistent, constant-time random access, infinite) -- so the ladder's top rung is a LEARNED terrain generator on the estate's integer nofloat ML stack, graded against that class, with classical erosion+climate below it as the sim substrate the learned tier trains and grades against. One engine, many representations: voxel game tier, heightfield live tier, SDF/mesh+splat beauty tier, all one seed. [terraindiff26] [infinigen23]Pros none measured yetCons behind Unreal Engine 5, Godot, Infinigen (leads), Houdini (leads); open contract; expected source declaration not established | ○ | ● | ● | ◉ | ◉ |
WORLD-ROADMAP (GAME TIER): hydraulic+thermal erosion on the heightfieldMeasured:wp_erode exists in runtime/nx_worldpipe.nx, verified at emit. FLIPPED 2026-08-30 (adoption seat; receipt: func wp_erode( at buildroot/runtime/nx_worldpipe.nx:304; comparewatch plane graphics_wp_erode LANDED; the note below already recorded PG20 CLOSED 2026-08-25 with nx_worldpipe_erode_gate 21/21 while the symbol field still spelled _ABSENT_ -- a cell that could never flip as written; the flip is the receipt, the erode gate over the full 102400-cell population is the proof. Residue stays named in the note, not hidden: the geomorphology-band half is half delivered -- hypsometric integral proven, drainage density refused as a subject mismatch). Prior note: EH1, THE LARGEST READS-AS-REAL JUMP (gap ladder 2026-08-18): value-noise terrain reads as blobs; carved valleys, ridgelines, talus and drainage basins are what reads as terrain. Pipe-model shallow-water hydraulic erosion plus thermal talus on the worldpipe heightfield -- integer-exact, which turns MASS CONSERVATION into a provable gate tooth instead of a float hope, and judged against PUBLISHED geomorphology bands (drainage density, hypsometric curve) exactly as T53 judges trees per hectare. Houdini's signature terrain feature. CORRECTION 2026-08-24: the earlier note here read 'catalogue-proven ABSENT (capsearch: no erosion organ)' -- that was a capsearch ZERO read as absence, and capsearch indexes registered tools only, so it is structurally blind to every library. The full Mei 2007 pipe model (all seven stages, four regimes, Q14) EXISTS in runtime/nx_water_erosion.nx and has since 2026-05; it was SOURCE-ONLY and is now gated on /compare/procgen with mass conservation as a tooth. CLOSED 2026-08-25 (PG20): the INTEGRATION landed. wp_erode in runtime/nx_worldpipe.nx bakes the terrain into the grid the hydrology stage already uses, drives the full Mei pipeline plus a generalised Musgrave talus pass through ONE owner (runtime/nx_terrain_erode_lib.nx), and stores a delta the height function composes. Proven by nx_worldpipe_erode_gate 21/21 GREEN, exit_code 0, BITE-PROVEN, over the full 102400-cell population: mass exact, INERT before the bake so no existing consumer moved, 101614 cells changed after it, causality restoring 102400 of 102400 with the stage off, and two bakes byte-agreeing everywhere. EH1's own phrase 'judged against PUBLISHED geomorphology bands' is HALF DELIVERED and the half is named: Strahler's hypsometric integral is the referee and is proven non-gameable because it is dimensionless (offset-invariant EXACTLY, scale-invariant within a derived truncation bound), reading 356 to 472 permille; DRAINAGE DENSITY is refused as a subject mismatch because km per km2 needs a metric scale this world does not have, and no maturity band edge is quoted because the primary answered HTTP 403 Cf-Mitigated challenge. [mei2007] [musgrave1989] Adoption: LIB-WIRED importers=8 nonval=3 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Infinigen (leads), Houdini (leads) | ● | ● | ● | ◉ | ◉ |
WORLD-ROADMAP (GAME TIER): climate to biome to competition fieldsDARK —runtime/nx_worldpipe.nx EXISTS but does not declare wp_biome_field: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. EH2: today's scatter is hash-per-cell with genome species; the Houdini/Infinigen form derives temperature and moisture from elevation, latitude and the rain the calendar ALREADY simulates, then species compete for light, water and slope. Most inputs exist in the engine (biome banding, snow line, seasons, rain); the field layer is the gap. Turns trees-placed into an ecosystem grown, all recipe-plane data. [infinigen23]Pros none measured yetCons behind Unreal Engine 5, Godot, Infinigen (leads), Houdini (leads); open contract; expected source declaration not established | ○ | ● | ● | ◉ | ◉ |
WORLD-ROADMAP (GAME TIER): volumetric clouds and fog on the GPU tierDARK —runtime/nx_game_page_emit.nx EXISTS but does not declare gpe_cloud_volume: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. EH3: clouds are 2D hash sheets in the shader today. Real 3D noise-field volumetrics raymarched at device resolution became FEASIBLE THIS SESSION -- the WebGL2/WebGPU tiers are exactly where they run; the sovereign CPU tier keeps its sheet clouds as the floor. [webgpu-status26]Pros none measured yetCons behind Unreal Engine 5, Godot, Infinigen, Houdini (leads); open contract; expected source declaration not established | ○ | ● | ● | ● | ◉ |
WORLD-ROADMAP (GAME TIER): road and path networks between settlementsDARK —runtime/nx_wasm_craft.nx EXISTS but does not declare wc_road_net: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. EH4: settlements landed 2026-08-18 (constraint-solved siting, 74/74); roads are the missing connective tissue -- the same part-solver lane that sites buildings routes paths (slope-cost shortest paths between anchors, stamped as data). Ladder order EH1-EH4 declared: erosion first, then biome fields feeding the existing scatter, then volumetrics, then roads; GI (gp_bounce_gi), dataset factory (ws_groundtruth_emit) and GPU compute (cw_webgpu_compute) rows above already carry the render half of the same roadmap. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5, Godot, Infinigen, Houdini (leads); open contract; expected source declaration not established | ○ | ● | ● | ● | ◉ |
WORLD-ROADMAP (GAME TIER): physically-based sky and aerial perspectiveMeasured:gpe_sky_atmos exists in runtime/nx_game_page_emit.nx, verified at emit. EH5, AND THE OPERATOR NAMED IT FIRST (2026-08-28: "our sky for example is still bad compared to a real sky"). CORRECTED 2026-09-03 AND THE NOTE SAID THE OPPOSITE FOR WEEKS: the sky is NO LONGER a two-stop ramp. The emitted shader carries derived Rayleigh with wavelength-dependent tau (440/lambda)^4 giving (.123,.287,.700), Kasten-Young airmass clamped at 38, single-scatter 1-exp(-tau*am), a sun disc, the SUNT transmittance dusk tint, volumetric clouds with Beer-Lambert extinction and a dual-lobe HG phase, and aerial perspective in shade() using the SAME constants; the shader's own comment reads that the two-stop painted gradient is gone. What was genuinely ABSENT was the SYMBOL: gpe_sky_atmos sat unnamed inline inside gpe_build_s, so this watch cell was CORRECT while this prose was STALE. The symbol now exists at nx_game_page_emit.nx:1063, proven content-neutral by nx_contentdiff at 3592 runs lost 0 gained 0 symmetry IDENTICAL-CONTENT. What reads as real is single-scattering Rayleigh plus Mie with a sun disc, and aerial perspective tinting distant geometry. It is also the rung that makes the GROUND read correctly, because sand albedo is lit by sky irradiance and a flat two-stop ramp cannot deliver it -- which is why EH5 is ordered BEFORE EH6 rather than beside it. Measured on the live CPU tier 2026-08-28: sky occupies roughly half the frame at default pitch, so this is the largest single-surface visual return currently on this board. Must stay integer-deterministic like every other render rung here. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: REGISTRY-UNCERTAIN — PARTIAL: registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion.Pros Implementing source symbol found; runtime acceptance separateCons behind Unreal Engine 5 (leads), Houdini (leads); not adopted yet: registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion | ● | ◉ | ● | ● | ◉ |
WORLD-ROADMAP (GAME TIER): ground MATERIAL -- granular sand and wet-shore darkeningDARK —runtime/nx_wasm_craft.nx EXISTS but does not declare wc_mat_ground: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. EH6: EH1 fixes terrain SHAPE and EH2 fixes WHAT GROWS, but neither touches what the ground IS, so the shore reads as tan voxel paint rather than sand. The physical tells are cheap and integer: a high-frequency granular normal that survives at grazing angles, darkening plus specular gain where sand is wet near the waterline, and slope-keyed dune banding. The input already exists and needs no new plane -- recipe spec row 66 carries the ABSOLUTE shoreline coordinate, so height-above-waterline is readable today. (operator 2026-08-28: "we dont have infinigen generated sand or palm trees") rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot, Infinigen (leads), Houdini (leads); open contract; expected source declaration not established | ○ | ◉ | ● | ◉ | ◉ |
WORLD-ROADMAP (GAME TIER): palm flora reaches the SERVED beach -- the grower landed, the recipe binding did notDARK —runtime/nx_wasm_craft.nx EXISTS but does not declare wc_veg_mesh: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. RE-SCOPED AND RE-CONTRACTED 2026-09-01 by the board owner after a full re-measurement; the 2026-08-28 note below was half stale and half still true, and each half was wrong on a different board. THE GROWER LANDED: func wsp_palm_shape( is declared and CALLED at sh9 == 3 in runtime/nx_wasm_craft.nx ("GE17: shape 3 = PALM, the frond-rosette grower"), so "only 1=dome and 2=pine have growers" has been false since 2026-08-30 and this rung has been sitting at ranker slot #2 for work already done. The contract name wc_tree_palm will never ship, because the capability shipped under another name on another board -- the same two-names-one-capability defect this board just paid for twice (sr_sss_preint, dyna_bind_skin), so the contract is repointed rather than left to rot. WHAT IS ACTUALLY OPEN, and it is NOT a grower: knowledge/world_recipes.conf gives beach NO spec pairs, DELIBERATELY -- its own comment reads "BEACH VEGETATION REVERTED 2026-08-30 ... both palm densities were judged by REAL-BROWSER frames of the served page -- 5:64 = an opaque green canopy roof, 5:400 = a forest of solid green cubes at eye level (frond rosettes render as filled voxel crowns, not fronds). A gate-green contract that fails the operator's eye is a regression; palms return to the beach only when the crown renders as fronds (or the world leaves voxels)." That is a done-rule already written in DATA, and it makes GR31 (a second art style of emission) a real prerequisite this plan did not carry. So the open contract is the BINDING plus its referee: flora derived for the shore identity rather than hand-typed as a spec pair a later edit can silently delete, proven on a captured frame of the SERVED beach, not on the presence of a grower. gameengine.matrix carries the same contract name for the same capability. CONTRACT NAME ALIGNED 2026-09-01, SAME HOUR, AND THE ALIGNMENT IS THE POINT: I first repointed this row to a contract of my own naming (wsp_beach_flora_bind) and, on re-reading gameengine.matrix, found lane 3 had split the same capability onto _ABSENT_:wc_veg_mesh minutes earlier. Shipping both would have MANUFACTURED a third instance of the exact two-names-one-capability defect this row is being corrected for -- so this row now carries THEIR name, not mine. Their measurement is also stronger than the conf's revert note and supersedes the "tune the density" reading: the engine declares 1 block = 1 m twice, wsp_palm_shape paints each frond as a chain of at most cr9=4 unit cubes in 8 directions, so one crown is ~33 cubic metres of solid green built from 1 m bars while a real coconut rachis is ~0.05 m -- the drawn element is 20x thicker than the thing it depicts, so NO sparsity value can make a blob stop being a blob. The exit condition is therefore not a recipe row at all: a crown must be drawable BELOW block resolution, which is GR31's second art style, and the mixing mechanism already exists (NPC bodies are instanced triangles z-tested against the voxel depth on the same canvas) and has simply never been generalised into a declaration. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5, Godot, Infinigen (leads), Houdini (leads); open contract; expected source declaration not established | ○ | ● | ● | ◉ | ◉ |
| NATIVE | |||||
NishiOS GPU raster rung on the proven sovereign compute laneDARK —runtime/_hdl_build/_gpu_dxg_submit_gate.nx EXISTS but does not declare gpu_raster_rung: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. From the first bit up, measured not asserted: the dxg ladder R1-R6 already resolves adapter-to-context-to-resident-VA and has a REAL submission ACCEPTED by the RTX 5080 (ABI measured by recon sweep, tamper matrix, repeatable), and the bare-metal lane carries PCI/GMMU/GPFIFO/doorbell/QMD-v5 gates with golden compute and matmul binaries. The raster contract: a compute-shader rasteriser writing a framebuffer through that same lane -- the native twin of the browser tiers, no libvulkan/libd3d12 anywhere. [webgpu-status26]Pros none measured yetCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen, Houdini (leads); open contract; expected source declaration not established | ○ | ◉ | ◉ | ● | ◉ |
FOUNDATION (meta, no rival claim): first-byte-up chain for every rung on this board -- what it rests on, where it stops being sound, and which unsound link you can actually act onMeasured:fb_rank_class exists in runtime/nx_firstbyte.nx, verified at emit. META ROW: this is the board's own soundness check, not a rendering capability, so it claims no exceed and scores no rival -- a dependency analyser measured against Houdini would be a subject mismatch. Joins every row here to the estate's own import graph (18776 nodes, 51163 edges, via eg_ancestors) and asks the cap census per module by EXIT CODE (fb_cap_query; 69 forks deduped across 141 link queries), so it derives no edges of its own and adds no second ruler. Measured 2026-08-23: 48 rows partitioning to ready 0, capped 27, placeholder 21, absent 0, empty-chain 0, unknown 0, with the sum printed and checked; 114 chain links, deepest 13. ALL 124 capped links are reported per rung and RANKED, never just the nearest -- own 23, rare 47, ubiquitous 54 -- and both boundaries are DERIVED rather than picked: OWN is identity with the rung's own source, UBIQUITOUS is presence in every resolved chain on this board. A cap in nx_syscalls (fan-in 15042) is an estate-wide concern and never a rung's blocker; a cap in the rung's own module belongs to the lane that owns the rung. ready=0 is a TRUE FINDING, not a broken instrument: only 2.56 percent of estate capacity bounds are DERIVED today, so almost no full transitive chain is clean. Proven by nx_firstbyte_gate 28/28 GREEN -- fbg_rank_order_ok parses all 27 ranked lists and 124 entries and asserts class order with ascending fan-in inside a class, plus a bite-proven neg-control that an absent matrix REFUSES by name instead of reporting an empty healthy board. Validation against the five known foundation defects of 2026-08-23: nx_gsplat CAUGHT and attributed OWN, the nx_skeleton mask LIFTED after the rig-floor lane derived that bound away (its census now reads capacity_population=0 verdict=CLEAN, and a link the census calls clean is never named as unsound), and three declared MISSES that an import graph structurally cannot see -- a runtime DATA prior, a language-level gap, and a capability no row declared. Data-dependency edges are the declared next axis and are printed as NOT-DERIVED on every run. Adoption: REGISTERED-UNAUTHORISED — PARTIAL: registered, no cap ever minted; no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked); only referenced by nx_firstbyte_gate -- a mention, not a run.Pros Implementing source symbol found; runtime acceptance separate; ahead of Unreal Engine 5, Godot, Infinigen, HoudiniCons not adopted yet: registered, no cap ever minted; no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked); only referenced by nx_firstbyte_gate -- a mention, not a run | ● | ○ | ○ | ○ | ○ |
| RENDER | |||||
linear-light pipeline, display-encoded once at outputDARK —runtime/nx_game_page_emit.nx EXISTS but does not declare sr_linear_light: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. Every rival engine computes lighting in linear and tonemaps once; our page pipeline is display-encoded throughout, measured 2026-08-26 when an ACES pass double-applied the transfer and washed the frame. Prerequisite to the SSS rung. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5, Godot, Infinigen, Houdini; open contract; expected source declaration not established | ○ | ● | ● | ● | ● |
pre-integrated subsurface skinMeasured:sr_sss_preint exists in runtime/nx_sdfrender.nx, verified at emit. ORGAN PATH CORRECTED 2026-09-01: this row named runtime/nx_game_page_emit.nx and the symbol is not declared there, so the row could never flip and GR9 sat at ranker slot #4 as OPEN while charsim.matrix:27 carried the SAME symbol LANDED -- two boards, opposite states, one symbol, and the whole discriminator was the organ path. MEASURED 2026-09-01 tree-wide, corpus_complete=1 coverage_complete=1 over buildroot/runtime: func sr_sss_preint( is declared EXACTLY ONCE, at runtime/nx_sdfrender.nx:853, delegating to sr_sss_preint_n -- Penner pre-integrated scattering with the curvature half and the per-channel normal half. The served beach page carries a PORT of the same numbers and says so in its own comment ("wraps 120/60/26 + curv gains 96/40/12 from nx_sdfrender (ONE owner)"), which is why the page looked like a plausible home and was not one. Historical intent kept: Penner pre-integrated scattering LUT baked integer-side like the sun LUT, bound to the baked normal map, with the away-side equality tooth so no uniform brightener can fake it. DECLARED LIMIT: the flip is the receipt, not the proof -- the away-side equality tooth belongs to the sdfrender lane's gate, and the page-side port is a COPY of the constants rather than a call into the one owner, which is the GE22 interop-only defect wearing a shader hat. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: LIB-WIRED importers=64 nonval=43 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth) | ● | ● | ● | ● | ● |
| HAIR | |||||
sovereign strand system vs TressFX barDARK —runtime/nx_softbind.nx EXISTS but does not declare hr_strand_sys: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. TressFX mirrored as the reference implementation; our current state is ribbon cards from generated guides -- the strand-system rung is the fork-chain leg 1. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen, Houdini; open contract; expected source declaration not established | ○ | ◉ | ◉ | ● | ● |
Marschner-class shadingDARK —runtime/nx_game_page_emit.nx EXISTS but does not declare hr_marschner_lobe: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. FALSE GREEN FOUND AND CORRECTED 2026-09-01 by the board owner. THE SPELLING WAS THE ONLY THING HOLDING THIS ROW UP. Two independent measurements, both full-population: (1) nx_shelltool grep over buildroot/runtime, coverage_complete=1 corpus_complete=1, finds hr_marschner_lobe in exactly FIVE places and NOT ONE of them is a declaration -- four are literals inside nx_gpe_marschner_gate.nx and the fifth is a COMMENT inside the emitted GLSL string in nx_game_page_emit.nx. The one ruler this estate declares (nx_symdecl_lib, a top-level declaration) therefore reads ABSENT, which is why the published page has been rendering this cell as not-measured all along. (2) the referee this row cited as 13/13 GREEN was RE-RUN on 2026-09-01 and reads NX-GPE-MARSCHNER passed 12/13 verdict=RED: T9 src-ships-the-H8-rows-the-discriminator-arithmetic-uses (red [.65,.33,.18], dark [.16,.14,.18]) FAILS, because the H8 hair palette those rows came from was replaced by melanin classes on 2026-08-30 and the gate's discriminator still calibrates against the retired rows. Ten teeth including both neg-controls and both served-page teeth still PASS, so the SHADER IS REAL AND IS SERVED -- what is absent is a NishiLang symbol and a green referee. The row is therefore respelled as a watch contract rather than deleted: the capability is carried entirely inside a quoted GLSL string, which is the same defect gameengine GE22 names (page JS/shader is interop only, authored logic ratchets to zero), so shipping hr_marschner_lobe as a real declaration is BOTH the fix for this cell and a GE22 payment. Historical: Marschner 2003 via the Karis 2016 production reduction, mirrored; absorption per d'Eon 2011 eumelanin/pheomelanin; served on all six worlds at emit stamp b1788123025. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen, Houdini; open contract; expected source declaration not established | ○ | ◉ | ◉ | ● | ● |
| FORK | |||||
Infinigen subsystem map live censusDARK —runtime/_hdl_build/nx_graphics_board.nx EXISTS but does not declare if_fork_map: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. Only Infinigen itself holds column 1; the row measures OUR mapping coverage of their subsystems with sovereign twins, references and gates. Ownership corrected 2026-09-07: authenticated nx_fs found the root path ABSENT and read the surviving _hdl_build source; nx_shelltool job 1788799934 found zero if_fork_map matches across that directory (8813 files, 58599682 bytes, corpus_complete=1). This resolves the board source location only. GR18 remains OPEN: the census, subsystem twins, pinned references, gates and emit-time measurement are not implemented by this path correction. Search beyond this directory remains required before selecting an implementation owner. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5; open contract; expected source declaration not established | ○ | ● | ○ | ○ | ○ |
| MATERIAL | |||||
skin MICRO-SURFACE on the live path -- pores and areola structure, generated not paintedDARK —runtime/nx_game_page_emit.nx EXISTS but does not declare gpe_skin_microsurface: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. THE CAPABILITY EXISTS AND THE FAMILY CANNOT SEE IT. Measured 2026-08-28: `sk_pore` is a real function at nx_sdfrender.nx:215, called at :1120 -- and a corpus-complete grep (23,334 files, 126,766,976 B, corpus_complete=1) finds it in NO OTHER FILE. CORRECTED 2026-08-28 after measuring rather than assuming: the maps are NOT missing. nx_asset_floor_gate reads GREEN 14/14, capability score 7/7, TEXM PRESENT -- the shipped asset carries a 721,830 B TEXM section and the GL tier samples it, so the 4/7 in nx_nxa_texm's own header is a STALE self-description of a gap that has since been closed. CORRECTED AGAIN AND NOW GATE-PROVEN: the pore model is not missing either. nx_relief_lib owns a three-band micro-relief law with cited constants -- relief 60um at 1200um, secondary 20um at 250um, pore 30um pits of 100um diameter at 800um pitch -- shared by the baker and the SDF renderer so they cannot disagree. The defect is RESOLUTION, and nx_relief_bands_gate (shipped and run 2026-08-28, GREEN 15/15) prints it: a WHOLE-BODY atlas spans 1.7m, so even at 4096 the texel is 415um and BOTH secondary and pore read carried=0 -- the bands are computed and then clamped away, and the lib says what that leaves behind, a DIFFERENT COARSER BAND WEARING THIS ONE'S NAME. The live cast ships at npc-atlas 2048, worse still. RES NEEDED whole-body: secondary 13600. Over a 250mm FACE region: secondary 2000, pore 625. So the remedy is a PER-REGION atlas, not a bigger one -- 625 is already met by the atlas we ship. That is the operator report 2026-08-28: "her skin needs texture not painting especially like her nipples". This row is NOT a duplicate of ntb_bake_pbr: that contract is the PBR map SET, this is the GENERATOR that gives those maps content at skin scale -- pore distribution driven by body region, and the areola as an anatomically structured surface rather than a painted disc. Per the operator it must come from the procedural genome from the first byte, not from an authored source texture, which is also the only form that survives a 1001-distinct-genome cast. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot, Infinigen, Houdini (leads); open contract; expected source declaration not established | ○ | ◉ | ● | ● | ◉ |
| ROADMAP | |||||
groom style zones with pinned dynamics constantsDARK —runtime/nx_nxa_groom.nx EXISTS but does not declare gr_style_zones: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. ROW AUTHORED 2026-09-01 TO CLOSE A RANKER BLIND SPOT AND IT IS A BOARD DEFECT NOT A CAPABILITY CLAIM. graphics.plan carries rung GR17 and NO matrix row carried its symbol, so nx_compare_rank printed UNMAPPED (no matrix row carries this symbol) value=0 for it and sank it below every graded rung regardless of merit -- the rung was invisible on the published page and unrankable BY CONSTRUCTION, which is the corrupt-input case: a critical path computed over a population that silently excludes real work. Contract-symbol absence re-proven at authoring time by nx_absent over buildroot/runtime .nx, coverage_complete=1 corpus_complete=1, so this cell cannot be a stale open. Rival codes are feature-observation reads of 2026-09-01 in the convention this board declares in its own header (not benched numbers) and they exist so the ranker has a deficit to compute -- correct them from a bench, never from taste. The done-rule stays the rung's own, proven by its gate; this flip would be a receipt. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Infinigen, Houdini (leads); open contract; expected source declaration not established | ○ | ◉ | ○ | ● | ◉ |
render the sections we already generate -- groom and garment must reach the viewerDARK —runtime/nx_nxa_texbake_lib.nx EXISTS but does not declare ntb_view_sections: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. ROW AUTHORED 2026-09-01 TO CLOSE A RANKER BLIND SPOT AND IT IS A BOARD DEFECT NOT A CAPABILITY CLAIM. graphics.plan carries rung GR19 and NO matrix row carried its symbol, so nx_compare_rank printed UNMAPPED (no matrix row carries this symbol) value=0 for it and sank it below every graded rung regardless of merit -- the rung was invisible on the published page and unrankable BY CONSTRUCTION, which is the corrupt-input case: a critical path computed over a population that silently excludes real work. Contract-symbol absence re-proven at authoring time by nx_absent over buildroot/runtime .nx, coverage_complete=1 corpus_complete=1, so this cell cannot be a stale open. Rival codes are feature-observation reads of 2026-09-01 in the convention this board declares in its own header (not benched numbers) and they exist so the ranker has a deficit to compute -- correct them from a bench, never from taste. The done-rule stays the rung's own, proven by its gate; this flip would be a receipt. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot, Houdini (leads); open contract; expected source declaration not established | ○ | ◉ | ● | ○ | ◉ |
garment and hair penetration ruler on the fine axesDARK —runtime/nx_clipcheck.nx EXISTS but does not declare cc_garment_pen: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. ROW AUTHORED 2026-09-01 TO CLOSE A RANKER BLIND SPOT AND IT IS A BOARD DEFECT NOT A CAPABILITY CLAIM. graphics.plan carries rung GR20 and NO matrix row carried its symbol, so nx_compare_rank printed UNMAPPED (no matrix row carries this symbol) value=0 for it and sank it below every graded rung regardless of merit -- the rung was invisible on the published page and unrankable BY CONSTRUCTION, which is the corrupt-input case: a critical path computed over a population that silently excludes real work. Contract-symbol absence re-proven at authoring time by nx_absent over buildroot/runtime .nx, coverage_complete=1 corpus_complete=1, so this cell cannot be a stale open. Rival codes are feature-observation reads of 2026-09-01 in the convention this board declares in its own header (not benched numbers) and they exist so the ranker has a deficit to compute -- correct them from a bench, never from taste. The done-rule stays the rung's own, proven by its gate; this flip would be a receipt. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5, Houdini; open contract; expected source declaration not established | ○ | ● | ○ | ○ | ● |
genome asymmetry -- synthetic faces measure too perfect against the photographic barDARK —runtime/_hdl_build/nx_body_gen.nx EXISTS but does not declare bg_asym_permil: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. ROW AUTHORED 2026-09-01 TO CLOSE A RANKER BLIND SPOT AND IT IS A BOARD DEFECT NOT A CAPABILITY CLAIM. graphics.plan carries rung GR21 and NO matrix row carried its symbol, so nx_compare_rank printed UNMAPPED (no matrix row carries this symbol) value=0 for it and sank it below every graded rung regardless of merit -- the rung was invisible on the published page and unrankable BY CONSTRUCTION, which is the corrupt-input case: a critical path computed over a population that silently excludes real work. Contract-symbol absence re-proven at authoring time by nx_absent over buildroot/runtime .nx, coverage_complete=1 corpus_complete=1, so this cell cannot be a stale open. Rival codes are feature-observation reads of 2026-09-01 in the convention this board declares in its own header (not benched numbers) and they exist so the ranker has a deficit to compute -- correct them from a bench, never from taste. The done-rule stays the rung's own, proven by its gate; this flip would be a receipt. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5, Infinigen, Houdini; open contract; expected source declaration not established | ○ | ● | ○ | ● | ● |
camera rejection sampling -- the composition filter behind the rival figuresDARK —runtime/nx_nxa_texbake_lib.nx EXISTS but does not declare ntb_cam_reject: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. ROW AUTHORED 2026-09-01 TO CLOSE A RANKER BLIND SPOT AND IT IS A BOARD DEFECT NOT A CAPABILITY CLAIM. graphics.plan carries rung GR22 and NO matrix row carried its symbol, so nx_compare_rank printed UNMAPPED (no matrix row carries this symbol) value=0 for it and sank it below every graded rung regardless of merit -- the rung was invisible on the published page and unrankable BY CONSTRUCTION, which is the corrupt-input case: a critical path computed over a population that silently excludes real work. Contract-symbol absence re-proven at authoring time by nx_absent over buildroot/runtime .nx, coverage_complete=1 corpus_complete=1, so this cell cannot be a stale open. Rival codes are feature-observation reads of 2026-09-01 in the convention this board declares in its own header (not benched numbers) and they exist so the ranker has a deficit to compute -- correct them from a bench, never from taste. The done-rule stays the rung's own, proven by its gate; this flip would be a receipt. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Infinigen (leads); open contract; expected source declaration not established | ○ | ○ | ○ | ◉ | ○ |
one colour transform derived and generated in NishiLang, two execution surfacesDARK —runtime/nx_sdfrender.nx EXISTS but does not declare sr_view_transform: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. ROW AUTHORED 2026-09-01 TO CLOSE A RANKER BLIND SPOT AND IT IS A BOARD DEFECT NOT A CAPABILITY CLAIM. graphics.plan carries rung GR23 and NO matrix row carried its symbol, so nx_compare_rank printed UNMAPPED (no matrix row carries this symbol) value=0 for it and sank it below every graded rung regardless of merit -- the rung was invisible on the published page and unrankable BY CONSTRUCTION, which is the corrupt-input case: a critical path computed over a population that silently excludes real work. Contract-symbol absence re-proven at authoring time by nx_absent over buildroot/runtime .nx, coverage_complete=1 corpus_complete=1, so this cell cannot be a stale open. Rival codes are feature-observation reads of 2026-09-01 in the convention this board declares in its own header (not benched numbers) and they exist so the ranker has a deficit to compute -- correct them from a bench, never from taste. The done-rule stays the rung's own, proven by its gate; this flip would be a receipt. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot, Infinigen, Houdini (leads); open contract; expected source declaration not established | ○ | ◉ | ● | ● | ◉ |
identity parameter space with correlated sampling -- the axis a finite asset library cannot enterDARK —runtime/_hdl_build/nx_body_gen.nx EXISTS but does not declare bg_identity_space: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. ROW AUTHORED 2026-09-01 TO CLOSE A RANKER BLIND SPOT AND IT IS A BOARD DEFECT NOT A CAPABILITY CLAIM. graphics.plan carries rung GR24 and NO matrix row carried its symbol, so nx_compare_rank printed UNMAPPED (no matrix row carries this symbol) value=0 for it and sank it below every graded rung regardless of merit -- the rung was invisible on the published page and unrankable BY CONSTRUCTION, which is the corrupt-input case: a critical path computed over a population that silently excludes real work. Contract-symbol absence re-proven at authoring time by nx_absent over buildroot/runtime .nx, coverage_complete=1 corpus_complete=1, so this cell cannot be a stale open. Rival codes are feature-observation reads of 2026-09-01 in the convention this board declares in its own header (not benched numbers) and they exist so the ranker has a deficit to compute -- correct them from a bench, never from taste. The done-rule stays the rung's own, proven by its gate; this flip would be a receipt. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5, Infinigen (leads), Houdini; open contract; expected source declaration not established | ○ | ● | ○ | ◉ | ● |
placement solved not authored -- garments and props fitted by the constraint solverDARK —runtime/nx_part_solver_lib.nx EXISTS but does not declare pl_fit_solve: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. ROW AUTHORED 2026-09-01 TO CLOSE A RANKER BLIND SPOT AND IT IS A BOARD DEFECT NOT A CAPABILITY CLAIM. graphics.plan carries rung GR25 and NO matrix row carried its symbol, so nx_compare_rank printed UNMAPPED (no matrix row carries this symbol) value=0 for it and sank it below every graded rung regardless of merit -- the rung was invisible on the published page and unrankable BY CONSTRUCTION, which is the corrupt-input case: a critical path computed over a population that silently excludes real work. Contract-symbol absence re-proven at authoring time by nx_absent over buildroot/runtime .nx, coverage_complete=1 corpus_complete=1, so this cell cannot be a stale open. Rival codes are feature-observation reads of 2026-09-01 in the convention this board declares in its own header (not benched numbers) and they exist so the ranker has a deficit to compute -- correct them from a bench, never from taste. The done-rule stays the rung's own, proven by its gate; this flip would be a receipt. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5, Infinigen (leads), Houdini (leads); open contract; expected source declaration not established | ○ | ● | ○ | ◉ | ◉ |
| REFEREE | |||||
world-quality grader BUILT and wired -- the world half of the referee storyMeasured:wqg_build_wired exists in runtime/nx_world_quality_grader.nx, verified at emit. MEASURED 2026-09-01 AND THE BOARD HAD NO ROW FOR IT AT ALL. nx_catalog nx_world_quality_grader reads SOURCE yes 37940 B, BUILT NO, PROMOTED NO, REGISTERED NO, AUTHORISED NO, INVOKED NO, VERDICT SOURCE-ONLY (S2). The estate has an admitted CHARACTER referee (cjc_panel) and NO built WORLD referee, so every world fidelity claim on this board is graded by the author's own eye -- the exact failure the character lane closed in GR1 and never closed here. Its rubric is already depended upon in prose: nx_layer_verdict.nx carries an adapter for its 12-i64 output and cites _wqg_grade_from_8 by name, and _ft_diag.nx imports the organ, so the consumer exists and the producer has never been built. Contract: the organ builds, promotes, registers and returns a grade for a named world capture, and the grade is what a world rung's done-rule reads. Absence of a built binary is a nx_catalog verdict, not a grep. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: REGISTERED-DARK — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked).Pros Implementing source symbol found; runtime acceptance separate; ahead of GodotCons not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked) | ● | ● | ○ | ● | ● |
| HAIR | |||||
the sovereign strand organ reaches a live world -- incumbent exists and is unbuiltDARK —runtime/nx_nxa_strands.nx EXISTS but does not declare wc_strand_bind: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. MEASURED 2026-09-01 AND THE BOARD HAD NO ROW FOR THE INCUMBENT. nx_catalog nx_nxa_strands reads SOURCE yes 9565 B, BUILT NO, PROMOTED NO, REGISTERED NO, INVOKED NO, VERDICT SOURCE-ONLY (S2). Rung GR15 buys a sovereign strand system at 2 u as if from nothing while 9565 B of strand source sits unbuilt beside it, so the plan overstates its own cost and no reader can see that the first move is a build rather than a design. Contract is deliberately the BINDING and not the organ: a strand set generated by the sovereign organ must reach a live world through an engine export, because building the organ alone would leave the same JS-side groom the operator is looking at. Sibling contracts already on the boards: gameengine wc_hair_tick (guides simulated in wasm) and gameengine gpe_wasm_hair_bind (page JS is interop only). rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Infinigen, Houdini (leads); open contract; expected source declaration not established | ○ | ◉ | ○ | ● | ◉ |
| WORLD | |||||
a second art style of emission -- surface extraction beside the voxel raycast, chosen as dataDARK —runtime/nx_wasm_craft.nx EXISTS but does not declare wc_style_surface: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. OPERATOR 2026-08-30: everything is voxel and we dont need it to be that should just be one art style of emission. MEASURED 2026-09-01: the served /world/beach WASM module exports 78 functions and not one of them names a mesh, surface, rig, face, hair or skin producer -- the exports are simulation state (positions, gene, garment id and category, mood) and the world is drawn by a per-pixel integer DDA raycast over a voxel field, so voxel is not the default art style, it is the ONLY one the engine can emit. The extractor already exists and is graded on two other boards: nx_csg_extract, marching tetrahedra on the Kuhn six-tetrahedron split, watertight by construction (procgen csg-booleans-marching-tetrahedra and dcc marching-tetrahedra-extraction, both LANDED). Nothing binds it to a world. Contract: the same worldgen field emitted as a surface mesh through the incumbent extractor, style selected by a recipe row and never by an engine fork, with the voxel path unchanged and byte-identical when the row is absent. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot, Infinigen, Houdini (leads); open contract; expected source declaration not established | ○ | ◉ | ● | ● | ◉ |
| ACCEPTANCE | |||||
hair accepted as complete by operator + referee + seatDARK —runtime/nx_accept_decl.nx EXISTS but does not declare ga_accept_hair: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. OPERATOR STANDING ORDER 2026-08-30: "doing it poorly isn't a complete ... it's not complete till you and I and the nishi team agree." Mechanical rows landed (gpe_hair_geometry, hr_marschner_lobe, palette variety, wind chain) and the operator's verdict is REJECT: "hair clips the body and moves like a mop, not strand by strand" (accept- row beach-hair-colour-wind-0830). Flips ONLY through nx_accept_gate on an ACCEPT row + referee GOOD on a full-res live-page crop + seat attestation; the strand/collision rebuild is frontier F1192. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen, Houdini; open contract; expected source declaration not established | ○ | ◉ | ◉ | ● | ● |
skin micro-relief accepted as complete by operator + referee + seatDARK —runtime/nx_accept_decl.nx EXISTS but does not declare ga_accept_skin: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. Operator verdict 2026-08-30 UNJUDGED (staged in ref9d_r2.nxa, not yet served; bar = real texture, not smooth painting, researched Aug-2026) -- accept- row beach-skin-relief-0830. Flips only via nx_accept_gate on three-party agreement. ASSET HALF LANDED 2026-08-30 (symbol stays _ABSENT_ because the live path is the page half). Research mirrored and pinned FIRST, on the operator's directive that per-region means TEXTURE and not more smooth surface painting: the field keeps pore detail in a tiling texture separate from the wrinkle normal [ue-skin] with a distinct micro skin detail layer [mh-materials]; scan vendors ship displacement per frequency band at 16k [txyz-vface]; microstructure is synthesised from measured statistics when scans are not licensed [graham13], pore-first, with published depth bands of 20-100um primary and 5-40um secondary [disney23]; and it is a DISPLACEMENT that smooths under stretch and roughens under compression [nagano15]. The learned and splatted frontier reaches sub-millimetre pores from capture [rgca23] [hravatar25] [tega25] -- noted, not adopted. MEASURED BEFORE BAKING (nx_nxa_texbake_region measure: the UV-to-3D Jacobian on every triangle of the served ref9d.nxa): the span ruler says 830um at 2048, but the per-joint cylindrical charts sample the torso at a mean footprint of 3281um (worst tile 6681), the face at 1004um, limbs 937um, gens 2022um; the whole-body atlas carried the primary band on 299 permil of torso area and the secondary band on none. The per-region layout (gR 4/4/9/2, tile px 512/512/227/1024) brings the footprint to face 365um, torso 1206um, limbs 768um, gens 367um, and the area carrying primary/secondary/pore-grid becomes face 793/598/674, torso 366/129/310, limbs 578/132/441, gens 659/536/556 permil. No unique atlas this estate ships can put two texels inside a 250um line around a torso, so the three finer bands ride a PERIODIC MICRO TILE from the same nx_relief_lib law: 480 px, texel 50um (two per pore orifice), repeat 24000um (the LCM of the band lattices), exactly periodic, carried=1 for primary, secondary, pore-grid AND pore-orifice by construction on every region; the page repeats it per chart at circumference/repeat by length/repeat from the joint table TEXR carries. SHIPPED: nx_relief_lib periodic composition (additive), nx_nxa_texbake_region_lib + nx_nxa_texbake_region (measure and bake; promoted and registered), knowledge/areola_relief.conf (the cited anatomy ledger the baker refuses to run without), and nx_nxa_texbake_region_gate GREEN 21/21 with three bite pairs: geometry-not-paint (full-band tile mean slope 26 against a primary-only smooth control of 12 permil, ratio 2166 over a derived bar of 1300), the pore period measured on the height field 804um against the 800um law (889 pits of 900 cells; the smooth control has none), the tile exactly periodic, the micro-normal re-deriving byte-identical from the height, TEXM byte-identical, and the areola as anatomy: apex from the DYNA bust pair, disc L* 36.8 against a skin ring of 53.0 (a 16.2 step against the ledger's 16 [nac-review21] [nac-tattoo24] [montgomery-sd] [doucet09]), the independent referee nx_areola_pigment returning MATCH (composite 905), and height slope inside the areola 118 permil against a flat control ring outside (nipple dome plus ten Montgomery domes as GEOMETRY). STAGED sites/nishifamily/world/ref9d_r2.nxa 39478840 B sha256 e73b179d5ef045cc04d5f36c597fa48c06a25b520c85b34dfb9ee28d92401f15: 17 sections, ref9d's 16 carried byte-for-byte plus TEXR 28720576 B; ref9d.nxa untouched. RESOURCE ENVELOPE, named rather than hidden: TEXR adds 28.7 MB (face 10.2, torso 3.2, limbs 5.4, gens 7.5, tiles 2.4), the 16-bit height PNGs dominate; the page can derive the unique normal from the height in-shader and drop 7.8 MB, and a 1024 limbs set is the next cut. HONEST UNIQUE-ATLAS TRUTH: the torso and limbs unique height carries no band at their footprints (bands_unique 0); only face and gens carry primary and pore-grid uniquely -- the pore band lives on the micro tile, stated, not hidden. OWED: the page half (sample the TEXR set per chart, repeat the micro tile, blend the micro normal onto the unique one), and ntr_tension_attenuate, which needs a stretch field the asset does not yet carry. AND THE PAGE-HALF ASSET MAY ALREADY EXIST, UNWIRED AND UNNAMED: sites/nishifamily/world/ref9f.nxa, 40448728 B, sha256 3767db7c391131f59840699e0274c299f1b1000312f7e08255af6641bbca8973, sits in the SERVED docroot and is named by no organ source (ABSENT-PROVEN over all 23407 .nx in buildroot/runtime, coverage_complete=1 corpus_complete=1) and by no compare row. It is NOT junk and NOT a partial write: its section table, read directly, is the served ref9e's seventeen (VERT TRIS CLUS SKEL SKIN ANIM GVRT GTRI GPIN POSE TEXC TEXM FACE MORF DYNA HSTR GARM) with TEXR inserted before GARM, and the arithmetic closes exactly -- ref9e 11728120 + the TEXR payload 28720576 recorded on this row + 32 for one section-table entry = 40448728, the same +32 that separates ref9d 10758232 from the staged ref9d_r2 39478840. So ref9f = the SERVED garment base carrying this row's micro-relief, i.e. the union asset the page half needs, built and never wired. DO NOT RETIRE IT as an orphan on a size or a name census: a built-and-undeployed artifact is byte-indistinguishable from abandoned junk and this one is a superset of the thing being served. NEXT ACTION is to point the page's NPC loader at ref9f.nxa and sample TEXR per chart, not to build another asset -- AND THE HOOK ALREADY EXISTS, SO THIS COSTS NO BUILD AND NO DEPLOY: beach.html:1818 parses const NP9=location.search.match(/npcasset=([A-Za-z0-9_]+)/) and line 1820 fetches /world/<that>.nxa before falling back ref9e then ref9d then ref9, so https://nishifamily.com/world/beach?npcasset=ref9f loads the TEXR-carrying asset on the LIVE page today. That is the cheapest available way to decide whether ref9f is the page half or a dead end, and it is also the full-res capture the ga_accept_skin referee lens needs. Verified by reading the served bytes, not inferred: beach.html is 565325 B titled REVIEW STAGE b1788142855, and its default fetch really is /world/ref9e.nxa.Pros none measured yetCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen, Houdini (leads); open contract; expected source declaration not established | ○ | ◉ | ◉ | ● | ◉ |
| DELIVERY | |||||
the character is drawn by the ENGINE, not by hand-written page JSDARK —runtime/nx_game_page_emit.nx EXISTS but does not declare gpe_wasm_npc_bind: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. MEASURED 2026-09-01 on the SERVED page fetched whole (sites/nishifamily/world/beach.html, 566500 B; the partition sums exactly): base64 WASM 299012 B (528 permil), HAND-WRITTEN JS 263646 B (466 permil), GLSL/WGSL <=64677, CSS 611, HTML 3231. The module exports 78 functions across 282 JS call sites and NOT ONE export names a mesh, surface, rig, face, hair, vertex or skin producer -- the engine renders the VOXEL WORLD and every character is drawn in JavaScript: loadNPC 42840 B, drawNPCs 21931, animEval 8728, nxHairTick 6208, nxDressTick 4980, drawEyes 735. Those bytes are not plumbing, they are behaviour: dual-quaternion skinning, the foot-plant solver, rigid-hand joint copying, shadow-umbra derivation, an HUD/telemetry spine and a save/restore scheduler. This row carries the gameengine board OWN contract name (gpe_wasm_npc_bind, GE23) rather than a second name for one capability. Operator standing order 2026-08-30: i dont want js powering any of this, emit from the first byte up via nishi lang. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen (leads), Houdini (leads); open contract; expected source declaration not established | ○ | ◉ | ◉ | ◉ | ◉ |
| HAIR | |||||
the guide chain is simulated in the engine, and it collides with the bodyDARK —runtime/nx_wasm_craft.nx EXISTS but does not declare wc_hair_tick: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. TWO MEASURED FACTS THAT BELONG TOGETHER. (1) ZERO hair-body collision exists in any of the six shipped world pages -- the only colliders live in the standalone rigview.html/rigmesh.html, which the world engine does not use -- so the operator complaint that hair clips the body is structural, not a tuning miss. (2) The estate ALREADY OWNS the solver that fixes it: nx_wasm_craft.nx carries 32-capsule collision push-out plus Mueller-2012 Follow-The-Leader, and it has ZERO CALLERS, while the page ships a JS twin whose own comment admits it is REPLICATED VERBATIM in integer-faithful JS. A capability built, gated and never called is not a missing capability, it is an adoption gap. AND THE PAGE SHIPS A FALSE CLAIM TO EVERY VISITOR ABOUT EXACTLY THIS: nx_game_page_emit.nx:383 and the emitted page at :653 both read ONE SOLVER, NO JS TWIN ... it does NOT integrate springs, falsified by the 2026-08-26 edit and never retracted -- a retraction is owed on the served surface, not only on this board. Contract name taken from gameengine GE21: one capability, one name. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen, Houdini; open contract; expected source declaration not established | ○ | ◉ | ◉ | ● | ● |
the strand chain is ordered ROOT-TO-TIP, proven by a toothDARK —runtime/nx_gpe_hair_geometry_gate.nx EXISTS but does not declare hg_seg_axis: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. A ONE-TOKEN DEFECT WITH A WHOLE-GROOM SYMPTOM, byte-confirmed in the served page by two independent reviewers 2026-09-01: hseg[v] = Math.floor((ytop-hverts[v*3+1])/span*NXHCH_N) reads component 1, but this mesh space is Z-UP -- the served shader own vec3 yup=vec3(sp.x,sp.z,-sp.y) at beach.html:292 proves component 2 is up, and the fit pass TEN LINES ABOVE the bug already uses component 2 for the same top-of-the-groom concept. So the 7-link chain lag and its squared weight hw9 are distributed FRONT-TO-BACK across the head instead of root-to-tip down each strand: the groom shears as rigid vertical slabs (the operator moves like a mop) and a band mapped to segment 6 gets full swing whether it is a root or a tip, which drives hair through the head and face. REFUTED IN THE SAME AUDIT, so the row does not overclaim: static plus blonde plus flat shading is FALSE -- the page builds real hair-card ribbon geometry (1200 guides x 6 children, Catmull-Rom, ~202k tris per head) under a genuine Karis-2016/Marschner lobe over d Eon melanin absorption. The contract is the TOOTH, not the token: a gate assertion that the segment index is derived from the up axis the shader itself declares, so the fix cannot silently regress. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen, Houdini; open contract; expected source declaration not established | ○ | ◉ | ◉ | ● | ● |
| MATERIAL | |||||
the normal map that is already inside the asset is actually readDARK —runtime/nx_game_page_emit.nx EXISTS but does not declare gpe_texm_normal_bind: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. A THREE-LAYER CONSUMER GAP MEASURED 2026-09-01, and layer one costs nothing but a call site: the SERVED ref9e.nxa already carries a TEXM section holding FOUR maps [albedo, spec, gloss, NORMAL] -- nx_nxa_texm.nx:14-15 defines NTB_MAP_NORMAL=3 -- and the page reads ONLY record 0 at BOTH of its TEXM call sites. A normal map is inside the file the browser already downloads and nothing reads it. Layer two: the geometric per-region relief (16-bit height plus derived normal, nipple dome and Montgomery glands in the HEIGHT channel, gate 21/21) was baked into a TEXR section that exists TWICE in the live docroot and is referenced by nothing. Layer three, and this is why swapping the asset alone changes NOTHING: S.TEXR is ABSENT-PROVEN across all 23419 organ sources. This row is layer one only -- one subject per row. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen, Houdini; open contract; expected source declaration not established | ○ | ◉ | ◉ | ● | ● |
the served cast loads the asset that carries the finished surfaceDARK —runtime/nx_game_page_emit.nx EXISTS but does not declare gpe_asset_ladder: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. ORPHANED WORK, MEASURED 2026-09-01 with corpus_complete=1 in both directions: the page serves ref9e.nxa (11728120 B, 17 sections). ref9f.nxa is 40448728 B / 18 sections and is ref9e PLUS TEXR:3590072 with EVERY other section byte-identical; ref9d_r2.nxa is 39478840 B and is the TEXR micro-relief asset. nx_shelltool grep over buildroot returns matches=0 for BOTH names with coverage_complete=1 corpus_complete=1 -- roughly 80 MB of finished character assets that nothing in the estate references. Both pass nx_asset_floor_gate identically on all five floor axes (re-run 2026-09-01: passed 14/14 verdict=GREEN). The page ALREADY has the mechanism -- a ?npcasset= query override and a ref9e to ref9d to ref9 fetch chain -- so the contract is a DECLARED ladder that names the richest asset a build is willing to serve, not a new loader. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5, Godot, Infinigen, Houdini; open contract; expected source declaration not established | ○ | ● | ● | ● | ● |
| FACE | |||||
the FACE and MORF sections the asset already carries are consumedDARK —runtime/nx_game_page_emit.nx EXISTS but does not declare gpe_morf_consume: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. THE OPERATOR hideous faces COMPLAINT, ROOT-CAUSED 2026-09-01, AND IT IS A CONSUMER GAP, NOT A GENERATOR GAP. The face is PAINTED IN A FRAGMENT SHADER: eyes, brow and mouth are colour bands keyed on normalised head height, and there is NO NOSE TERM AT ALL, painted or geometric. The served ref9e.nxa carries FACE:4700 and MORF:98634, and the page three NXA consumers dereference 15 tags of which FACE and MORF are not two -- all 7 FACE and 1 MORF mentions in the 330836 B emitter are COMMENTS. nx_mhbench own board reads head-SHAPED, surface FEATURELESS ... a smooth egg carrying PAINTED rectangles (brow bars, mouth bar) and surface-mounted eyes, with NO NOSE AT ALL, detail 79 at head scale. The producer nx_nxa_morf is LIVE and ITS CONSUMER WAS NEVER WRITTEN; 7 of 9 face organs are REGISTERED-DARK. Prior art carried honestly: FACE RELIEF v1 was reverted 2026-08-03 for goblin noses because this mesh normals are majority-wrong, and v2 preconditions sit unshipped in a comment -- so the done-rule must include the normal precondition, or this rung reproduces its own history. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen, Houdini; open contract; expected source declaration not established | ○ | ◉ | ◉ | ● | ● |
| REFEREE | |||||
the JS-logic ratchet cannot be laundered by editing a confDARK —runtime/nx_jslogic_census.nx EXISTS but does not declare jl_class_pin: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. THE RULER THAT ENFORCES THE ESTATE LOUDEST OPERATOR ORDER IS EDITABLE BY THE PARTY IT MEASURES. knowledge/jslogic_class.conf (7817 B) maps each JS literal fnv64 hash to a CLASS token (LOGIC / INTEROP / SHADER); the census reports logic_unknown = tL + tU, so moving a literal from LOGIC to INTEROP or SHADER removes its bytes from the counted bucket WITH NO CODE CHANGE AND NO RE-BASELINE. The conf self-invalidates when a literal is EDITED (the hash changes, the row reads UNKNOWN) and does NOT self-invalidate when only the CLASS TOKEN is edited -- so the one mutation that launders the number is the one mutation the file cannot see. Measured history: the ratchet read RED on 25 consecutive runs since 2026-08-13, total JS literal bytes GREW 107209 to 229911 (+114 percent) since the 2026-08-12 order, and the headline then improved 228377 to 128559 on 2026-08-31 while ZERO bytes of JS were removed -- the two journal rows total 229911 IDENTICALLY, with 99818 B moved into buckets the ratchet does not count. Re-run live 2026-09-01: logic=128559 interop=58656 shader=42696 unknown=0. A sibling lane wrote nx_jslogic_census_gate.nx on 2026-09-01 with 8 teeth for a DIFFERENT axis (uncharged JS lying outside the script region) and it is SOURCE-ONLY (BUILT NO, PROMOTED NO), so it neither covers this nor could yet run. CONTRACT: the class conf is PINNED -- the census reads a declared digest of the class column and REFUSES, loudly and by name, when a class changed without the baseline being re-declared in the same commit, with a neg-control that an ordinary literal edit still self-invalidates exactly the way it does today.Pros none measured yetCons open contract; expected source declaration not established | ○ | ○ | ○ | ○ | ○ |
the character judge survives the frame it is pointed atMeasured:cjc_frame_guard exists in runtime/_hdl_build/nx_charjudge.nx, verified at emit. THE WORLD LANE HAS NO WORKING REFEREE FOR ITS OWN FLAGSHIP FRAME, measured 2026-09-01. nx_charjudge takes SIGSEGV, signal 11 at 0x41510f touching 0x0, on the beach capture; nx_addr2line then REFUSES because nx_charjudge.elf was built WITHOUT -g, so the crash cannot even be located. Separately, nx_percept anchor corpus is ALL NXFH1 400x240, so a 1920x1200 capture is not comparable to it at all -- two independent reasons the panel cannot grade the surface the operator is actually looking at. The baseline it CAN produce is the honest headline of this whole board: nx_frame_score on the native beach capture reads pal=153 det1=2 det4=9 grad=7 cov=999 cohere=993 quality=408 verdict=BASIC, against the ruler OWN published calibration that a random-squares control scores 628 RICH -- the live beach grades BELOW A RANDOM-NOISE CONTROL, and fine detail is essentially absent. Contract: the judge either grades a full-resolution frame or ABSTAINS by name; it never dies, and a debug build exists so the next crash is locatable. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: RUN-BY:fork:nx_charjudge_census — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth) | ● | ● | ● | ● | ● |
the character crop actually contains a characterDARK —runtime/nx_world_snap.nx EXISTS but does not declare ws_char_crop_nonempty: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. MEASURED 2026-09-01 by a native capture with no browser: nx_world_snap char reports CHAR-FRAME framed=1 ground_y=11 and then CHAR-CROP pixels=0 -- the framing succeeds and the crop returns no subject image, so every downstream character judgement on a world capture is being made about an empty picture or not at all. The same run reports style_applied=0 gen_applied=0 w=1920 h=1200 res_class=FULL png_bytes=6913793 and writes a 13825216 B .nxfh beside the PNG, which is the artifact nx_frame_score actually eats. Second, smaller finding recorded so it is not lost: nx_world_snap HAND-ROLLS its NXFH1 writer at lines ~515-555 while _hdl_build/nx_nxfh_lib.nx:141 nxl_dump(path,fb,w,h) is the shared codec -- a duplicate ruler under rule 15. Contract: the crop refuses by name rather than emitting an empty image, and a tooth asserts pixels>0 whenever framed=1, so a zero crop can never again read as a successful capture. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5, Godot, Infinigen, Houdini; open contract; expected source declaration not established | ○ | ● | ● | ● | ● |
FOUNDATION (meta, no rival claim): the board ranker MEASURES every row instead of trusting its spellingDARK —runtime/nx_compare_rank.nx EXISTS but does not declare rk_measure_all: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. THE DONE-RULE CORRUPTION, READ OUT OF THE RANKER OWN SOURCE 2026-09-01 (buildroot/runtime/nx_compare_rank.nx lines 551, 552 and 619): mpres[mrows]=1 ; if rk_starts(fld[2],_ABSENT_)==1 { mpres[mrows]=0 } ... if mpres[mi]==1 { done=1 } else { done=rk_measure(morg,sym) }. THE RANKER MEASURES ONLY THE ROWS ALREADY SPELLED ABSENT AND TRUSTS EVERY ROW SPELLED PRESENT -- the measurement is applied exclusively to the rows that cannot benefit from it and never to the rows whose claim it would test, and it is asymmetric in the flattering direction: a rung can be promoted open to done by measurement and can NEVER be demoted done to open. A false green is therefore invisible BY CONSTRUCTION and the critical path is computed on corrupt input. THE PAGE DOES NOT HAVE THIS DEFECT: nx_swcompare_matrix calls sd_present on BOTH branches with the same ruler, which is why graphics.matrix row 85 has been rendering not-measured on the served page while the ranker printed it as the board #1 FINISH row -- two instruments over one row, and only one of them was looking. CONTRACT: one line plus its bite -- the ranker measures both branches; and because a fleet-wide re-baseline is the real risk, the done-rule requires the before/after done-count published for every domain in the same run, never a silent flip.Pros none measured yetCons open contract; expected source declaration not established | ○ | ○ | ○ | ○ | ○ |
REVIEW VIEWERS: the rig, view and stream pages gain the sovereign wasm tierDARK —runtime/nx_nxa_rig_emit.nx EXISTS but does not declare nre_sov_tier: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. The GR41 watch cell (operator 2026-09-02: rigview printed NXA REFUSED no webgl2). Today the three NXA viewers have exactly one renderer, WebGL2; gameengine GE28 makes a refusal name its cause and remedy, this cell removes the dependency by composing nx_trimesh (the sovereign textured rasteriser ntb_view already uses server-side) into the wasm memory model nx_meshview_wasm proved. No rival ships a software tier for its web viewer, coded No for all four; the exceed is that a review instrument keeps working on a device whose browser refuses GL. Done-rule in graphics.plan GR41: a wasm trimesh build sized for a 14k-vertex rig renders the same NXA the GL tier renders from the same SKEL and SKIN sections, the page falls to it with the badge naming the tier, and a headless capture's figure crop grades on cjc_panel inside the GL tier's band.Pros none measured yetCons open contract; expected source declaration not established | ○ | ○ | ○ | ○ | ○ |
| BEAUTY | |||||
a ray-triangle BVH so the path tracer hits the cast meshes (GR43)DARK —runtime/nx_gi_path_lib.nx EXISTS but does not declare gp_bvh_ray_tri: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. MATH GAP MEASURED 2026-09-03: no ray acceleration structure exists in the estate by name (nx_bvh_lib is BioVision mocap). The rival engines and DCCs ship BVH-accelerated tracing natively; Infinigen inherits Cycles. Index-level, UNPINNED. Contract in graphics.plan GR43. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen (leads), Houdini; open contract; expected source declaration not established | ○ | ◉ | ◉ | ◉ | ● |
Sobol with Owen scrambling drives every bounce, the LCG kept as the control (GR44)DARK —runtime/nx_gi_path_lib.nx EXISTS but does not declare gp_sobol_owen: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. MATH GAP MEASURED 2026-09-03: the tracer samples from a hashed LCG; no low-discrepancy sampler exists by name. Index-level, UNPINNED. Contract in graphics.plan GR44. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen (leads), Houdini; open contract; expected source declaration not established | ○ | ◉ | ◉ | ◉ | ● |
| LIGHT | |||||
spherical-harmonics irradiance projected from the physically-based sky for the real-time tier (GR45)Open — watchingruntime/nx_sh_irradiance.nx : sr_sh_irradiance, re-measured on every compare beat. Ship that symbol and this mark flips itself; the comparewatch- plane row flips with it. MATH GAP MEASURED 2026-09-03: no spherical-harmonics organ exists by name; the live tier lights from a two-stop ramp. Index-level, UNPINNED. Contract in graphics.plan GR45. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen (leads), Houdini; open contract; expected source declaration not established | ○ | ◉ | ◉ | ◉ | ● |
cascaded shadow maps on the kernel and WebGPU tiers, the cast shadowing the ground (GR46)DARK —runtime/nx_wgsl.nx EXISTS but does not declare rk_shadow_cascade: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. MATH GAP MEASURED 2026-09-03: no shadow-map organ exists by name; the served beach draws dark blobs under feet. Index-level, UNPINNED. Contract in graphics.plan GR46. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind themPros none measured yetCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen (leads), Houdini; open contract; expected source declaration not established | ○ | ◉ | ◉ | ◉ | ● |
| IDENTITY | |||||
integer Cholesky so the correlated identity draw has its primitive (GR47)Measured:la_cholesky_q30 exists in runtime/nx_linalg.nx, verified at emit. MATH GAP MEASURED 2026-09-03: nx_matrix stops at 3x3 determinant and inverse; no Cholesky by name. Index-level, UNPINNED. Contract in graphics.plan GR47. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: LIB-WIRED importers=6 nonval=4 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Infinigen (leads) | ● | ● | ● | ◉ | ● |
| HAIR | |||||
strand budget and frame cost against the published 2026 real-time barDARK —runtime/nx_game_page_emit.nx EXISTS but does not declare gpe_hair_lod: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. THE BAR IS NOW A NUMBER WITH A METHOD, NOT AN ADJECTIVE. Published real-time cost, each measured on named hardware: a deferred software rasterizer renders a 127k-strand groom at 1920x1080 on an NVIDIA RTX 5080 in 3.1 ms against 13.6 ms for the mesh-shader baseline at MSAA 1 without LOD, a 4.4x speedup, and with LOD the far view falls from 4.5 ms to 0.4 ms as the strand count drops from 127k to 21k, with multiple grooms totalling 483k strands rendered at 4K in 9.4 ms against 37.6 ms [hairswrast26]; an independent EGSR 2025 level-of-detail framework on an RTX 3090 measures a 0.99x overhead at 50 percent screen occupancy, about 2x at 10 percent and 13x at 0.3 percent where it reads 0.4 ms against 5.3 ms, plus a constant 0.1 ms per frame to reorganise for the next level [hairlod25]. The shipping product bar is a per-LOD table rather than a slogan: MetaHuman carries 50000 hair strands at LOD0 and 25000 at LOD1 and has NO strand tier below that, falling to 30000 hair card vertices at LOD0, and is cards only on Mac and on iOS and Android where the best LOD is 3 [mhlodspec]. WHERE WE ACTUALLY STAND, stated precisely rather than kindly: gpe_hair_geometry ships Catmull-Rom ribbon CARDS driven by a runtime spring chain, which is MetaHuman's LOD2-and-below representation, so our hair is not a degraded strand groom but a different and cheaper tier -- and no strand count, card count or per-frame hair cost has EVER been measured for our own renderer, so none of the three bars above can currently be compared against us at all. THIS ROW CONTRACTS THE MISSING MEASUREMENT, NOT A NEW RENDERER: gpe_hair_lod must publish our own strand or card count and per-frame hair cost at three named camera distances on named hardware, in the same units the bars use, so the next hair claim is comparable instead of arguable. The published 13x and 11x results also name what to build first if cost is the blocker -- distance-adaptive detail, which this board has not got in any form.Pros none measured yetCons behind Unreal Engine 5 (leads), Godot (leads), Infinigen, Houdini; open contract; expected source declaration not established | ○ | ◉ | ◉ | ● | ● |
| SKIN | |||||
skin-tone fidelity as CIELAB error and ITA-class preservationDARK —runtime/nx_skin_ita.nx EXISTS but does not declare skin_tone_fidelity_dE: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. THE BAR IS A MEASURED ERROR, NOT AN ADJECTIVE. Over 19848 rendered MetaHuman images in Unreal Engine scored against Chicago Face Database references, skin-tone fidelity is Delta E CIE76 and Individual Typology Angle error: extraction-method medians 12.69 (cheek) 9.67 4.90 3.77, lighting medians Delta E 2.2 to 4.5 under the reference light but 5.8 to 14.6 under Paramount and 7.6 to 18.7 under frontal light, and ITA error flat for classes I and II then strongly non-linear below -- class V median 4.14 under reference light rising to 31.32 and 39.51, class VI 11.37 to 46.61 to 53.85, with darker phenotypes rarely preserved and compressed toward lighter classes [truetotone26]. WHERE WE STAND, measured from source: nx_skin_ita GENERATES genome-driven albedo across the published ITA classification (its gate's T4 tooth refuses a locus that spans none of the classes), so the generation side exists -- but NO organ measures the fidelity of OUR RENDERED skin against a reference set in Delta E or ITA error under named lighting, so our number on this row is UNMEASURED and the bar cannot be compared against us at all. The contract this row watches: render the ITA ladder through the live path under three named lights, score Delta E and ITA error per class against the references, and publish the per-class medians in the same units. RIVAL CELLS on this row: Unreal reads Part because the mirrored paper itself renders MetaHumans and measures the bias; the other three are NOT WITNESSED by the mirrored source and are coded 0 for that reason, not as a claim they lack it.Pros none measured yetCons behind Unreal Engine 5; open contract; expected source declaration not established | ○ | ◐ | ○ | ○ | ○ |
measured ground-truth BSSRDF as the calibration oracleDARK —runtime/nx_skin_pbr_gate.nx EXISTS but does not declare skin_bssrdf_calibration: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. THE ORACLE IS A CAPTURE RIG, AND IT STAYS OUTSIDE OUR BUILD PATH. A polarimetric multispectral rig -- two polarization machine-vision cameras at 20 fps behind different multispectral filters, 40 LEDs in eight modules of five near-coaxial, stereo modules for dynamic geometry -- fits spatially varying single scattering and refractive index and validates against a hyperspectral camera and ten objects of known refractive index at a mean reconstruction error of 0.028; its polarimetric loss gives the minimum RMSE on the specular-dominant polarization observation and the smallest RMSE on specularity and full rendering averaged over 200 different novel camera poses of 11 participants [polbssrdf25]. WHERE WE STAND: nx_skin_pbr_gate proves our skin shading is physically shaped, and nothing in the estate is CALIBRATED against a measured BSSRDF -- our subsurface term has no ground truth behind it, so its error is UNMEASURED. The contract: adopt one published measured face BSSRDF as data (license beside it), fit our skin model's parameters to it, and publish the RMSE against the measurement as the row's number. RIVAL CELLS: none of the four rivals is witnessed by the mirrored source; coded 0 for that reason only.Pros none measured yetCons open contract; expected source declaration not established | ○ | ○ | ○ | ○ | ○ |
| FABRIC | |||||
sheer and stretched fabric over skin renders as a thin layered mediumDARK —runtime/nx_gi_path.nx EXISTS but does not declare gr_sheer_bsdf: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. GR57 (2026-09-27, the beach-lounger reference class). The garment is a THIN LAYERED MEDIUM over skin: reflectance from the skintwin Kubelka-Munk layering (km_layer_over), transmission rising with weave strain from the drape solver and with wetness, the skin term visible through the fabric in the beauty tier and the live tier. The field this month renders fabric on its own (Eurographics 2026's texture-free woven model and garment fold inter-reflection [kesen_eg2026]); no mirrored source renders sheer fabric OVER skin with strain-driven transmission. Watch contract: gr_sheer_bsdf in the path tracer, a swatch-over-skin KAT within a declared DE2000 band at three strains. Rival cells are coded 0 because no mirrored source witnesses them, not as a claim they lack it.Pros none measured yetCons open contract; expected source declaration not established | ○ | ○ | ○ | ○ | ○ |
| WET | |||||
one wetness field darkens skin, clumps hair and darkens sandDARK —runtime/nx_gi_path.nx EXISTS but does not declare gr_wet_state: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. GR58. One wetness FIELD drives three materials: skin darkens with a sharpened specular (skintwin ST7), hair clumps by capillary adhesion with an index-matched cuticle (skintwin ST9), sand darkens by the wet-porous term (GR27). The estate's wet-appearance scan read 2026-09-27 holds no 2026 paper on the subject; the standing references are the micro-ellipsoid wet-porous model and Jensen's 1999 model, so this row has no current-month bar and says so. Watch contract: gr_wet_state consumed by the beauty tier with dry-to-wet DE2000 and lobe shift published per material. Rival cells coded 0: not witnessed by a mirrored source.Pros none measured yetCons open contract; expected source declaration not established | ○ | ○ | ○ | ○ | ○ |
| SKIN | |||||
sun-tan and tan lines are a derived melanin field, never paintMeasured:gr_melanin_field exists in runtime/nx_skin_ita.nx, verified at emit. GR59. Melanin concentration is a per-vertex field DERIVED from garment coverage history and declared sun exposure, read by the skintwin chain and relit correctly by construction; ISO/DTR 26478 patches and NIST's traceable skin spectra are the calibration oracles. Watch contract: a tan line follows a garment edge and its inside-outside ITA difference lands in the band the referee reports. MetaHuman 5.8 applies palm and sole tone automatically and exposes nail parameters, a painted-parameter route [nvidia_siggraph2026_blog] is not the witness for it; rival cells coded 0 because no mirrored source witnesses a derived field. Adoption: LIB-WIRED importers=42 nonval=41 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of Unreal Engine 5, Godot, Infinigen, Houdini; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth) | ● | ○ | ○ | ○ | ○ |
| CONTACT | |||||
the lounger cushion compresses under the body and its cover drapesDARK —runtime/nx_gm_cloth_lib.nx EXISTS but does not declare gr_support_cover: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. GR60. The cushion COMPRESSES (softbody SB15) and its cotton cover DRAPES and wrinkles (PG29 cloth XPBD, dcc DC10), with body-cushion contact free of interpenetration (softbody SB10; S4R 2609.20524 is the September reference for procedurally assembled scenes [arxiv_co_siga2026]). Watch contract: contact area and sink depth against the stiffness rows, a non-trivial wrinkle field, zero interpenetration at the declared contact distance. Rival cells coded 0: not witnessed by a mirrored source.Pros none measured yetCons open contract; expected source declaration not established | ○ | ○ | ○ | ○ | ○ |
| CAMERA | |||||
exposure, sensor response, lens and depth of field are ONE camera row both tiers readDARK —runtime/nx_gi_path.nx EXISTS but does not declare gr_camera_row: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. GR61. A photograph is the scene times a camera: exposure, sensor response and tone curve, focal length, aperture and depth of field, vignetting, flare and chromatic aberration as one data row on the linear-light pipeline through the one colour transform. Watch contract: a rendered grey card and a photographed one agree within a declared DE2000 band under the declared exposure, and depth of field matches a measured circle of confusion at two distances. Rival cells coded 0: not witnessed by a mirrored source.Pros none measured yetCons open contract; expected source declaration not established | ○ | ○ | ○ | ○ | ○ |
| REFEREE | |||||
per-region DE2000 against a licence-clean reference photograph classMeasured:gr_refstill_referee exists in runtime/nx_colorsci_lib.nx, verified at emit. GR62. A reference photograph CLASS (licence-clean stills, never an identity) is measured per region in CIELAB (skin ITA, sand, fabric, cushion, sky) and every beach frame publishes its per-region DE2000 beside the world referee grade through the skintwin CIEDE2000 ruler (cs_de2000, built and promoted, zero adopters outside the garment lane). Watch contract: the number on the beach page with bands from measured references. Rival cells coded 0: not witnessed by a mirrored source. Adoption: LIB-WIRED importers=6 nonval=3 — fully adopted (top of its ladder).Pros Implementing source symbol found; runtime acceptance separate; ahead of Unreal Engine 5, Godot, Infinigen, Houdini; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth) | ● | ○ | ○ | ○ | ○ |
| BENCH | |||||
real-versus-render indistinguishability with outside liar-killersDARK —runtime/nx_world_quality_grader.nx EXISTS but does not declare gr_realrender_bench: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. GR63, the acceptance ruler for the operator's 2026-09-27 goal (benchmark a real photo, publish our frame, no visible difference). Pairs of licence-clean real stills and our frames of the same scene class with a synthetic human are judged by outside synthetic-image detectors run outside the build path (NVIDIA's SIGGRAPH 2026 synthetic-video detector NIM is the named first oracle [nvidia_siggraph2026_blog]), by outside VLM judges through the referee panel, and by a human two-alternative forced choice; the score is detection accuracy toward chance with sample size declared and an A/A calibration. Watch contract: a published bench page with the three arms and receipts; INCOMPLETE until all three arms and the operator's ACCEPT row exist. Rival cells coded 0: no rival publishes such a bench in a mirrored source.Pros none measured yetCons open contract; expected source declaration not established | ○ | ○ | ○ | ○ | ○ |
| Class | Count | Note |
|---|---|---|
| rows with rival marks | 82 | rows carrying a Yes, Best or Part in a rival column |
| cited | 23 | a reference mark resolves to a pinned mirror |
| uncited | 59 | an observation read, badged until a reference lands |
| rows, summed | 82 | RECONCILED: the parts sum to the total 82 |
Delivery and evidence
Inspect risks, technical debt, rendered observations, experiments and references. Read scope and limitations alongside every result.
Release history and work log
| When | Rung | Kind | Entry |
|---|---|---|---|
| 2026-09-06 | GR11 | retract | OPERATOR 2026-09-06: the WebGPU lane that delivers https://nishifamily.com/world/beach?gpu=1&review=0 is named as a rolling disaster (bad output, random coding) despite every benchmark, oracle and industry document being available -- a heavy-investment climb that stalled short of SOTA, one of the key pillars this work addresses from September 2026 research. On this board GR11 stays HALF LANDED as written and its LIVE LEVERED half reads as PRESENCE only until the same frame bit-compares against the CPU path, the CTS and WPT webgpu suites pass on the real GPU, and the operator ACCEPTs; the ecosystem stall detector (EC40) reads this board's investment against its rank velocity |
| 2026-09-06 | GR11 | queue | SOTA RE-BASELINE, September 2026: pin the current WebGPU and WGSL specifications, the WebGPU CTS and WPT webgpu suites and the reference implementations as refs with years; name the oracles this board already holds and did not consume (nx_wgsl_pixel_gate, nx_wpt_runner, the CDP real-GPU wire, the referee panel, the CPU-path bit-compare); split GR11 REMAINING into gated rungs each carrying an oracle receipt; a rung without a receipt never reads done, and a one-step landing without a measured outcome ranks below a longer rigorous one (engineshift ES27) |
Debt register
| Id | Sev | What it is | Unblock |
|---|---|---|---|
| charjudge-blindnot in this scope | 5 | The single referee is hue- and semantics-blind in both directions: green clay scored 257 while a photoreal-class splat render scored 0, its smoothness read as no-detail. Every fidelity claim is ungradeable through it alone. | CLOSED 2026-08-19: cjc_panel landed and ADMITTED at FP=0 FN=0 on labelled corpus v1 -- fidelity claims now ride the panel's verdicts; residual inside GR1: the VLM lens is unobservable until a sovereign VLM oracle exists, and the floors need the second-corpus retest. |
| avatar-v1-qualitynot in this scope | 5 | The operator verdict on the v1 splat portrait stands recorded; causes measured and mechanical: un-searched body knobs (the 5-knob search is FLAT -- the form lives in the 16 canon ring rows), no face composite, no hair, and a splat radius that melts features. | GR2: canon-mode search plus composing the gated face pipeline plus hair volume; promote and register the organ out of the build lane. |
| 1786646407not in this scope | 5 | Arena overrun in the shared image-decode path on webp and one jpg: a write past the allocation while pixels still render -- the harmless-while-written, fatal-when-read class. | Fix the decode lib, then re-run the compositor smoke until the canary lines stop appearing. |
| magic-anatomynot in this scope | 4 | Rule-11 honest counts at threshold 2: the body factory holds 510 sites over 107 distinct values and the anatomy renderer 461 over 88. Most are structural; the anthropometric band edges 600 and 764 per-mille are POLICY hiding as arithmetic. | Hoist the band edges into the canon table as data, in the body lane with its own gates; prove neutrality by byte-identical rebuild. |
| advances-403-checknot in this scope | 2 | www.realtimerendering.com serves 403 to our client while the kesen and advances subdomains serve; a mirror that does not verify per-root reachability would publish false coverage. | GR7 prints reachable, mirrored and dead per root with the partition reconciled, exactly like the banked Ke-Sen run (236 reachable, 235 mirrored, 1 dead). |
Risk register
| Risk | Likelihood x impact | Mitigation |
|---|---|---|
| The referee gap gates everything: shipping fidelity rungs before GR1 reproduces the gumby-through-a-green-board failure | certain x high | GR1 first; flip bars pre-declared before experiments; the operator eyeball stays a named lens on the panel. |
| The box is saturated -- swap high, edge 503s under load -- and render sweeps and regens contend with it | likely x medium | Ask the spend gate before sweeps, batch off-peak, publish box receipts beside results, and stop writing when the edge 503s. |
| A watch contract flips on symbol existence, so a stub could read green | possible x high | The done-rule proven by a gate is the bar; the flip is the receipt -- standing law on every rung above. |
| Infinigen is a moving bar: 2.0 alpha is shipping the indoor frontier right now | likely x medium | Bars carry their sources and dates; the frontier watch topics aimed at this arc on 2026-08-15 feed the library on the beat. |
| Estimates rest on one graphics calibration point (R21) | certain x medium | Recalibrate on every landed rung and publish actual against estimate; the mangagen corroboration is banked. |
Release history continued
| When | Rung | Kind | Entry |
|---|---|---|---|
| 2026-09-06 | GR27 | land | SOURCE: nx_world_shader_src.nx 77,418 -> 94,324 B (7 CAS edits, size ladder exact): fn sand(hp,c,rd,t)->vec2f in the ONE shader source -- the field normal tilted by hash grain facets and wind ripples, both Nyquist-faded against the pixel footprint stretched by the grazing angle; the wet band from an 8-direction x 3-radius water probe of the voxel volume at the column top and one below, submerged when water sits above; wet albedo x0.55 with a Blinn film lobe and a Fresnel sky sheen; dry quartz glints. shade() takes it on the field for classes 1, 2 and 5 (nx_wasm_craft gencol: 5 is sand at and below the waterline, SHORE repaints slot 1 as shore sand), suppresses the per-cell hash relief there, and the film loses the sun inside a block shadow. Build of nx_wgsl in flight; gate, promote and the beach re-emit owed. Every constant is a named data hook: the ripple heading is the wind row, the grain scale the recipe's sand grade |
| 2026-09-06 | GR27 | lesson | The shoreline coordinate the rung note points at (recipe spec row 66) is NOT needed: the volume already carries water as class 4, and probing the column's neighbours derives wetness from the actual water, which also answers lakes at other heights where a single waterline uniform would not. The per-class ground material word is the real data hook still owed; today the sand classes are keyed by id under the field |
| 2026-09-06 | GR27 | measure | nx_wgsl rebuilt on the NAS (second admission try, first refused QUEUE at load 18.09): 1,201,520 -> 1,223,496 B sha 0bc81fa0; nx_contentdiff lost_from_live=0 gained=19 ONE-WAY-GAIN GREEN 3/3; nx_behaveprobe world live 24,891 B vs staged 29,136 B both rc=0 DIFFERS STRONG (the sand function in both dialects); /api/promote PROMOTED with expect_sha256, prev banked; nx_offc_install refreshed the runner twin; nx_wgsl_gate 288/288 GREEN on the promoted binary |
| 2026-09-06 | GR27 | land | SERVED: nx_craft_emit re-emitted craft, beach, vale and isle (native gate 95/95, feel oracle 17/17, wasm-vm 19/19, page JS lexes and scopes resolve); the world fragment captured from nx_wgsl world grew to fragment_bytes=14508 and sites/nishifamily/world/beach.html 714,422 -> 716,600 B now carries fn sand(hp:vec3f,c:vec3i,rd:vec3f,t:f32)->vec2f in its WGSL. Shipped, referee grade OWED (no browser capture was launched: the no-GUI-without-asking law), operator ACCEPT owed -- the cell stays incomplete until the three signatures exist |
| 2026-09-02 | GR34 | land | gpe_texm_normal_bind SOURCE-LANDED in buildroot/runtime/nx_game_page_emit.nx (337,538 -> 339,9xx B, eight anchored CAS edits, every intermediate state a compiling NishiLang string): the body-pass fragment shader gains uniform sampler2D uNrm + int uNrmOn and perturbs n with the asset's TEXM normal map in a screen-space cotangent frame (dFdx/dFdy of world position and UV, no tangent attribute); the resident-cast decode reads the TEXM record whose map_id is 3 by scanning the record table (never assuming slot 3), decodes it once through nxDecodePng and binds it on unit 4 (unit 3 is the dress map uGD); GL.npc carries nrm; the per-frame bind sets uNrmOn=1 only when both the map and the uniform exist; the visiting-cast draw resets uNrmOn=0 so it never inherits the resident binding. uNrmOn=0 is byte-identical to the incumbent shading. Build is QUEUED (BQ-nx_craft_emit, admission-refused all day); the served page still runs the record-0-only build until it fires and the worlds re-emit. |
| 2026-09-02 | GR34 | queue | PROOF OWED before the cell may flip: (1) nx_nxa_texm_gate stays 12/12; (2) the new tooth -- a headless capture of /world/beach?review=1 with window.__npc_nmap===1 must DIFFER from the same frame with the map forced off, and the diff must sit inside the figure crop (a bind that changes no pixel is not a bind); (3) cjc_panel grade on the lit crop beside the record-0-only crop, because a normal map that reads worse is a regression on the operator's surface, not progress. |
| 2026-09-02 | GR34 | lesson | the ranker's own text was the spec: this rung cost 0.5 because the map was already inside the file the browser downloads. Layer two (per-region relief) and layer three (spec/gloss records 1 and 2 into the BRDF) stay open rungs; do not fold them in here. |
| 2026-09-02 | GR2 | measure | OPERATOR 2026-09-02, looking at /world/beach?review=1&cast=1&hud=1: that looks like our junk cpu stuff. SEEN BY THE SEAT THROUGH THE GECKO WIRE (readPixels off the page's own GL context, 2235x1117, WebGPU tier gpu=1): a flat sand plane rendered as high-frequency speckle noise with no albedo structure, no cast shadows except dark blobs under feet, no sky in frame at default pitch, no fog or depth cue, four figures with hard-edged painted garments, ribbon hair and flat cartoon shading, a hotbar palette in the corner. The native voxel frame from nx_world_snap (1920x1200 seed 20260728) is a block world: cubic trees, block snow, block clouds. The heightfield renderer (nx_worldgen, sovereign Infinigen rung 1, built and run on the laptop in 29 s for 6 worlds) is continuous terrain with fBm, lake, sky gradient and sun, but the march step (t += 55 + t/48) terraces every slope into visible stairs. THE VERDICT IS STRUCTURAL, NOT COSMETIC: the live beach is the GAME-tier voxel engine, and no amount of tuning makes a voxel raymarcher photoreal. |
| 2026-09-02 | GR2 | measure | REFEREES, honestly: nx_uiq_visual says RENDERED (its own env line: guardrails-not-perceptual); nx_frame_score gives the voxel frame composite 542 BASIC and its own calibration line says values above about 600 order by statistics, not perceived quality; nx_percept's ladder is 12/13 REFUSED (the judge does not reproduce the human ordering). THE ESTATE HAS NO CALIBRATED ENVIRONMENT-REALISM REFEREE, so a photoreal claim cannot be graded today by anything but eyes; that referee is a prerequisite rung, not an afterthought. |
| 2026-09-02 | GR42 | queue | first byte: nx_gi_path.nx composing nx_worldgen's heightfield and nx_perlin, primary ray + shadow ray + N-bounce diffuse with a hashed sampler, sky as the light source, PNG out through nx_png_write; the laptop builds and runs it (the NAS build lane is admission-refused all day). Reference frame: infinigen.org mirrored 2026-09-02 (cmp_graphics_infinigen_org.html, pin fc604ea3); its media URLs are the next fetch so the comparison has a pinned image, never a remembered one. |
| 2026-09-02 | GR42 | measure | FIRST FRAME 2026-09-02, laptop toolchain: nx_gi_path.nx compiled sovereign (asm 228,072 B, ELF 51,371 B) and rendered seed 77 clear-day at 480x288, 4 spp, 2 bounces in 21,959 ms: 751,158 marches, 362,814 sky rays, 188,945 water rays, 199,399 terrain hits, PNG 157,474 B, deterministic by construction (seeded integer LCG). SEEN: continuous slopes with NO terracing (bisection-refined hits), sky-lit shade in the valleys, mirrored lake -- and the frame is PALE: the sky model and palette are display-encoded values that were transported as linear and encoded again, the double-transfer defect GR14 already banked. Fixed in source the same hour (gp_lin decode before transport, one encode at output); second frame owed. Reference pinned the same hour: infinigen.org/img/random_sample.jpeg (447,956 B, pin 134fe4cf, a 3x2 mosaic of Infinigen scenes: birch forest, fern creek, coral reef, beach rocks, cactus desert, canyon). UNGRADED by rule: nx_uiq_visual reports RENDERED for both ours and the live capture (guardrails only), nx_percept's ladder is RED, so the comparison is eyes-only until the referee rung lands. |
| 2026-09-02 | GR42 | land | nx_gi_path BUILT ON THE NAS the moment the queue found headroom (59,288 B, sha a2fd860c, src 5b0c6b22) and PROMOTED + REGISTERED (live nishihost/nx_gi_path.elf). Source then split into nx_gi_path_lib.nx (the tracer, 13 functions), nx_gi_path.nx (CLI + the gp_bounce_gi watch symbol) and nx_gi_path_gate.nx (inherits nx_gate_verdict: 15 teeth, 4 named neg-controls, in-process, under a second). Gate GREEN 15/15 on the laptop; BITE-PROVEN against the subject: GP_REFINE 8 -> 0 in the lib turns it RED 13/15 on exactly the refinement teeth, restore verified by hash (513c7f4c) and back to GREEN. Two defects the gate found in its own first run and that are now fixed in source: an 8-bit linear decode could not round-trip display values (c=102 -> 40 -> 100), so radiance now carries 10 fractional bits; and the seed neg-control was VACUOUS because its probe ray landed on the lake (a deterministic mirror event) -- the fixture now searches pitches until a terrain hit is reached and asserts it. All three sources landed on the NAS; NAS rebuilds of nx_gi_path (from the split) and nx_gi_path_gate requested; the gate's promote, registration and organ_gate.conf pairing follow when they build. |
| 2026-09-02 | GR34 | measure | LIVE ON THE SERVED PAGE 2026-09-02 (b1788381661, read through the Gecko wire off the page's own GL context): window.__npc_nmap=1 -- the TEXM normal record was located by map_id, decoded once and bound on unit 4 for the body pass; nmapfail empty; tier WebGL2 + npc-atlas 2048; frame p50 75 to 84 ms, p95 136 to 138, p99 160 against a 133 ms budget at 2235x1117. The bind is PRESENT; the pixel-diff tooth (same frame with the map forced off must differ inside the figure crop) and the referee grade are still owed before the cell may read GREEN, and until then this rung reads shipped, unproven. |
| 2026-09-03 | GR32 | measure | 2026-09-03 the strand delta the operator sees between /world/rigview and /world/beach is a RENDERER delta, not an asset delta: both pages read HSTR, and rigview loads the older ref9d while the beach loads ref9e. nx_nxa_rig_emit draws every strand as a sub-mm fibre with analytic coverage (its own comments name the trichology width, the length spread and the flyaway minority); the world emitter samples guides out of HSTR into 1200 opaque ribbon cards. The closure this board owns is GR15 plus this rung: nx_nxa_strands (still SOURCE-ONLY, 9565 B, never built) reaches a live world through an engine export and the fibre-coverage shader is emitted ONCE by nx_wgsl from NishiLang for both the viewer and the world -- extract rigview's coverage shading into that emitter, never copy its page JS onto the beach (gameengine GE21 GE22 GE38 ratchet page JS to zero). Until then ecosystem EC22 reads the beach UNBOUND on the groom capability the viewer proves. |
| 2026-09-04 | GR47 | land | INTEGER CHOLESKY LANDED IN THE ORGAN THE MATRIX ROW NAMES: la_cholesky_q30 plus la_is_symmetric, la_chol_max_resid_q30 and la_chol_status_name added to runtime/nx_linalg.nx, byte-identical both trees at 12378 B, with nx_linalg_chol_gate.nx 11037 B at 26/26 GREEN. Q30 Cholesky-Banachiewicz over nx_fixq30_lib (fq_mul, fq_div, fq_sqrt) as the rung required, composing that arithmetic rather than adding a second one. CLAUSES 1 TO 3 OF THE DONE-RULE ARE PROVEN AND CLAUSE 4 IS NOT CLAIMED: GR24 composing it for the correlated draw is GR24 own rung and remains open, so this is the primitive landing and not the draw. |
| 2026-09-04 | GR47 | measure | THE BAND IS MEASURED AND THE FIRST ONE I WROTE WAS DECORATIVE. Worst absolute entry of A minus L L-transposed, in Q30 units: identity 0, textbook 3x3 0, irrational-root 3x3 3, dimension-6 4, and dimension-32 4 at the declared cap. THE RESIDUAL IS FLAT IN DIMENSION, so the error is dominated by per-operation truncation rather than by accumulation -- a result worth having, because it is the opposite of what the header first asserted and the header was corrected to say could grow rather than grows. I first declared the band at 4096 and the gate own ratchet tooth REFUSED it: a bar three orders above the worst observation would never catch a regression. Tightened to 64, which is 16x the worst measured and about 6e-8 relative to FQ_ONE, and pinned in BOTH directions so it can be neither loosened into decoration nor tightened into a permanent red. THE LOAD-BEARING TOOTH IS NOT THE RESIDUAL: a residual check passes for any L whose product happens to reconstruct A, so the factor is compared against a HAND-DERIVED L for the textbook 3x3 and for the 2x2 positive control, and both match EXACTLY at 0 Q30 units. |
| 2026-09-04 | GR24 | land | THE CORRELATED IDENTITY DRAW LANDED AS A GATED CORE: nx_identity_space_lib.nx 8264 B plus nx_identity_space_gate.nx 11080 B, gate 22/22 GREEN. x = mu + L z, where L is GR47 la_cholesky_q30 and is NOT recomputed here -- the rung own words, one ruler never a second decomposition. CLAUSE 1 PROVEN AGAINST THE DEFECT IT REPLACES, which is the only way to prove it: over a population of 2000 the empirical off-diagonal reads 1.974 against a target of 2.000 (1.3 percent) with axis variances 3.984 against 4 and 5.150 against 5, while THE SAME MACHINERY DRIVEN BY A DIAGONAL FACTOR decorrelates to 0.018 -- a 110x separation, so a covariance check that only read the diagonal would have passed the independent per-channel sampling this rung exists to replace. CLAUSE 2 PROVEN BY WIDENING: drawing 5 axes instead of 2 from the same seed leaves axis 0 and axis 1 BIT-IDENTICAL, because every axis draws from its own splitmix-derived stream, so appending a sixth master parameter can never reshuffle the existing back catalogue. CLAUSE 3 is bit-identical regeneration with a different-seed neg-control so reproducible does not degenerate into constant. CLAUSE 4 IS NOT CLAIMED: the referee panel showing population spread above the tier floor needs a panel whose only voting lens is currently self-refusing, and borrowing another rung evidence would be the whole defect this board exists to prevent. |
| 2026-09-04 | GR38 | retract | THE RECORDED MECHANISM FOR THIS RUNG IS WRONG AND THE CORRECTION MATTERS MORE THAN THE FIX. The matrix row and this rung both state that nx_charjudge takes SIGSEGV signal 11 touching 0x0 ON THE BEACH CAPTURE, and pair it with the anchor-corpus size mismatch as two independent reasons the panel cannot grade the surface the operator is looking at. RE-MEASURED 2026-09-03 against a live capture taken this session (2868x1604, IHDR depth 8 colour type 2 non-interlaced, 3297194 B): IT GRADES IT, rc=0, CHARJUDGE=533 composition=1000 palette=835 contour=533 face=1000. It also grades 4227x2099 and 1024x8000 at rc=0. THE FRAME WAS NEVER THE TRIGGER AND FRAME SIZE NEVER BLOCKED THIS LANE. THE TRUE MECHANISM, REPRODUCED THREE TIMES: argv is NULL-TERMINATED, so calling nx_charjudge with the image and NO LABEL makes argv[2] the terminator and printing that label dereferences 0x0 -- exit 139 which is 128+11. Two arguments grade; one argument crashes; zero arguments took a DIFFERENT wrong path and reported a missing PATH as a READ failure. A CRASH ATTRIBUTED TO THE DATA WHEN IT BELONGS TO THE CALL SENDS EVERY READER INTO THE WRONG ORGAN -- here it sent them into the renderer and the anchor corpus. The anchor-corpus 400x240 mismatch is a SEPARATE and still-open fact; only the crash half is retracted. |
| 2026-09-04 | GR38 | land | cjc_frame_guard SHIPPED into runtime/_hdl_build/nx_charjudge.nx with gate nx_charjudge_frame_gate.nx at 18/18 GREEN. The old main() dereferenced FOUR pointers unchecked (argv entries, the decoded header, the pixel buffer, and two mmap results) and validated WIDTH while never checking HEIGHT at all. The guard is a PURE function taking pointer values as integers, so every refusal path is reachable in-process without manufacturing a corrupt PNG -- a gate that could only test what a real file produces would leave the null-pointer branches, which are the ones that actually crashed, permanently unexercised. TEN NAMED STATES: OK, NO-LABEL-ARGUMENT, NO-PATH-ARGUMENT, DECODER-RETURNED-NO-HEADER, DECODER-RETURNED-NO-PIXELS, NON-POSITIVE-DIMENSIONS, UNSUPPORTED-CHANNEL-COUNT, PIXEL-BUFFER-SHORTER-THAN-DIMENSIONS, FRAME-AREA-OVERFLOWS-THE-BUFFER-SIZE, FRAME-BUFFER-ALLOCATION-REFUSED. DONE-RULE MEASURED CLAUSE BY CLAUSE: it grades a full-resolution frame (byte-identical output to before the change on the documented two-argument call, so no consumer regresses); the one-argument call that was exit 139 now answers abstain NO-LABEL-ARGUMENT graded false at exit 4; the abstention is a THIRD state distinct from graded 0, read 2 and decode 3, and it prints NO axes so nothing can read a score out of a refusal; and --debug now emits debug-line rows=1813 files=15 so the next crash is locatable. The overflow fence is DERIVED from one i64 per pixel rather than picked, and a tooth asserts the largest safe area is still ACCEPTED so the fence cannot be one step too tight and read as robustness. |
| 2026-09-04 | GR38 | retract | CORRECTING MY OWN RETRACTION FROM MINUTES AGO, WHICH WAS HALF WRONG AND I PUBLISHED IT TOO FAST. I wrote that the recorded SIGSEGV-on-the-beach-capture mechanism is wrong because a live 2868x1604 capture grades at rc=0. THAT MEASUREMENT WAS TAKEN AGAINST A STALE LAPTOP TREE: nx_charjudge.nx is 2723 B on the laptop against 3328 B on the NAS, and nx_charjudge_lib.nx is 15879 B dated Aug 7 against 29466 B on the NAS -- the lib is nearly DOUBLE and the laptop build could not even compile the NAS CLI because CHJ_NOUT does not exist locally. WHAT STANDS: the argc mechanism is REAL AND PRESENT IN THE NAS SOURCE TOO -- I read it, and argv[2] is dereferenced with no argc check in both copies, so a call with no label dereferences the argv terminator at 0x0 in either. WHAT DOES NOT STAND: my claim that the frame was never the trigger. I proved that only for a lib fourteen kilobytes behind the shipping one, so a frame-related crash in the CURRENT judge is NOT excluded and the row's frame half is UNPROVEN rather than refuted. A CROSS-TREE CONTROL IS NOT A CONTROL: this is the same error as comparing two build lanes and attributing the difference to the code, which this estate recorded earlier the same day, and I repeated it within hours. The anchor-corpus 400x240 mismatch remains separate and open. |
| 2026-09-04 | GR26 | land | gpe_sky_atmos NOW EXISTS AS A SYMBOL IN runtime/nx_game_page_emit.nx, AND THE INTERESTING PART IS THAT THE CAPABILITY WAS ALREADY SHIPPING. This rung is ranked 2 on this board at value 22000 and the operator named it FIRST of seven, and the matrix note still says the sky is a two-stop vertical colour ramp today. IT IS NOT. The emitted shader already carries derived Rayleigh with wavelength-dependent tau (440/lambda)^4 giving (.123,.287,.700), Kasten-Young airmass clamped at 38, single-scatter 1-exp(-tau*am), a SUN DISC at d>.99996 with a mix skirt, the SUNT transmittance dusk tint, and raymarched volumetric clouds with Beer-Lambert extinction and a dual-lobe HG phase; AERIAL PERSPECTIVE is in shade() using the SAME constants, mix(col,sky(rd),fw9). The shader's own comment says the two-stop painted gradient is gone. WHAT WAS GENUINELY ABSENT WAS THE NAME: gpe_sky_atmos occurred ZERO times, because the emission sat inline inside gpe_build_s, a thousand-line function. So the WATCH CELL was correct and the MATRIX NOTE was stale, which is a distinction worth keeping -- the instrument was right and the prose around it had rotted. A CAPABILITY THAT SHIPS WITHOUT ITS CONTRACT NAME IS INVISIBLE TO EVERY INSTRUMENT WATCHING FOR THE NAME, AND IT ROTS IN THE PESSIMISTIC DIRECTION: THE BOARD UNDERSTATES WHAT IS BUILT, WHICH IS THE HALF NOBODY AUDITS. FIXED BY NAMING THE REAL CODE, NOT BY WRITING A STUB: gpe_sky_atmos owns the sky emission point and every sky byte the page emits passes through it. Also removed a genuine duplicate import of nx_atmosphere.nx on consecutive lines. PROVEN NEUTRAL BY THE ESTATE'S OWN PROMOTE RULER rather than asserted: nx_contentdiff between the pre-change and post-change binaries reads runs_scanned=3592 lost_from_live=0 gained_by_candidate=0 symmetry=IDENTICAL-CONTENT, 3/3 GREEN. Both trees byte-identical at 358580. SCOPE DECLARED: the shader SOURCE literal still sits at the call site rather than inside the function. Hoisting it was built and proven content-neutral locally under the same ruler, and deliberately NOT shipped, because it means retyping four kilobytes of shader across a transport that has 503'd repeatedly today and one wrong byte is a corrupted sky nobody would notice until a human looked at the page. |
| 2026-09-04 | GR25 | measure | Live /world/rigview via Marionette (Waterfox 153.1, 1600x1000): __nxa reports clip=187 then clip=192 across two spaced reads, nv=14164 nt=28092 nj=104, fps 74-100. clip is NOT a defect count -- the source comment says __nxa.clip = contacts and the loop increments it ON THE PUSH-OUT, so it counts contacts RESOLVED and a WORKING collider RAISES it. Driving it to zero would delete the collision response and make clipping worse. |
| 2026-09-04 | GR25 | measure | ROOT CAUSE of garment-through-body on rigview, arithmetic not opinion: the vertex shader displaces the RENDERED surface by aS.x times (aP.x below zero ? uJ1 : uJ1R) plus aS.y times uJ2, applied AFTER dual-quaternion skinning, while the CPU collision cloud sets bsx from joint skinning ONLY and never sees j1o, j1oR or j2o. Constants measured from source: BRAD=.006*H shell, SRCH=.030*H search radius, soft-tissue clamp plus/minus .032*H. The breast travels 5.3x the shell and 1.07x the WHOLE search radius, so at peak excursion the cloth vertex riding the true surface is farther from the nearest UNDISPLACED sample than SRCH, hits the no-sample-in-range guard and registers NO CONTACT AT ALL. The collider goes blind exactly when the excursion is largest, which is when the clipping is most visible. |
| 2026-09-04 | GR25 | land | nx_nxa_rig_emit.nx 50326 to 51710 B on the NAS publishing tree, four CAS-verified edits, each size delta PREDICTED then ASSERTED: +948 soft-tissue term in the collision proxy, +8 missN declared, +10 missN incremented, +418 missN published. The injected term is DERIVED from the shader line rather than invented, so proxy and render cannot drift apart: a 200000-trial random-input equivalence check against the shader expression gives worst absolute deviation 7.105e-15 against a ~10-unit collision shell, with 197402 firing trials and a 2598-trial all-zero control proving the non-soft-vertex path is unchanged. Sovereign build GRANTed and GREEN (asm 292893 B, elf 65409 B). SOURCE-LANDED AND BUILD-PROVEN, NOT YET SERVED. |
| 2026-09-04 | GR25 | lesson | A COUNTER NAMED FOR WHAT THE READER ASSUMES RATHER THAN WHAT IT MEASURES SENDS THE READER AT THE WRONG REMEDY: __nxa.clip reads as clipping-defects and counts contacts-resolved, and acting on the name would have removed the collision response. Read the increment site before believing a counter name. SECOND, measured twice today: a REFUSED /api/build DESTROYS the _build fossil -- the 65409 B artifact from a GRANTed sovereign build was gone after nx_organ_ship's /api/build was refused on the I/O storm, so any staleness comparison taken after a refusal measures against an absence. THIRD: check-before-build stopped a duplicate rung -- GR25 already owns garment-through-body and hair-through-body penetration with GR20 as its ruler, so this is a CONTRIBUTION to GR25 and closes nothing on its own. |
| 2026-09-04 | GR25 | queue | OWED, blocked only on CAPABILITY not on knowledge: this seat holds jobrun, fswrite and stage_alias but NOT mgmt, so /api/promote is out of reach and nx_organ_ship routes its build through /api/build which admission refused. The build IS queued as BQ-nx_nxa_rig_emit with queue_launch_rc=0 and fires on the next nx_orchestrate pass with headroom. RECIPE TO FINISH, in order: (1) _offc/nx_sov_build_run.elf nx_nxa_rig_emit --build-only, which bypasses mgmt admission and was already proven GRANT-GREEN on this exact source; (2) nx_organ_ship nx_nxa_rig_emit once the I/O storm clears, so the content ruler, behaviour probe and gate all run before the promote; (3) re-read __nxa.clip and __nxa.miss on /world/rigview. miss is the NEW blind-set axis, a SEPARATE axis never folded into clip, and its DELTA across the fix is the reading -- the target is NOT zero, because a hem swinging clear of the body is legitimately a miss. |
| 2026-09-04 | GR1 | measure | THE DARK REFEREE FIRED, 2026-09-04, and its first full census SEPARATES THE TWO PIPELINES CLEANLY. nx_charjudge_census panel over 18 acorp captures: judged=18 failed=0 truncated=0, partition RECONCILES, admission=ADMITTED on cjc_panel_labels_union_tiered.list with one voting lens (percept) at floor 152. EVERY WORLD RENDER FAILS: ac_beach_char_r1 percept=68, ac_beach_world_r1=94, ac_craft_world_r1=131, ac_craft_char_r1=151 (one point short of the floor). EVERY CHARACTER CAPTURE PASSES: 171 181 177 195 196 199 214 219 226 228 316 328 337 350. Means 111 world versus 238 character -- the character pipeline scores 2.15x the world pipeline ON THE SAME REFEREE, and the fail/pass boundary falls exactly on the pipeline boundary. This is the operator's judgement ("/world/beach looks still below ps2 level graphics", "beach wasnt hit yet because the pipeline is broken") reproduced as a NUMBER by an instrument that had never been run. |
| 2026-09-04 | GR1 | lesson | A DARK REFEREE STARVES EVERY STATISTICAL ORGAN DOWNSTREAM, AND THE STARVATION READS AS A STATISTICS PROBLEM. Before this run the referee- plane held ONE ROW in its life (ref-face-1788160423), so nothing in the estate was contrastable and /compare/acceptance could not answer "are we improving" for any subject. The A/B core is NOT the blocker: nx_abstat_lib is 24/24 GREEN and is imported by exactly two organs (its own gate and nx_frameab, corpus_complete=1), i.e. LIB-GATE-ONLY -- the same shape the ranker reports for GR47 la_cholesky_q30. ONE RUN OF THE PANEL TOOK THE PLANE FROM 1 GRADE TO 19. THE FIRST REAL A/B PAIR NOW EXISTS IN THE DATA: op_before_0827t2019 versus op_after_0828t1034, head 171 to 214 and body 181 to 195, both IMPROVED -- two arms, so PAIRED-NOT-YET-POWERED against nx_abstat's NX_MC_MIN_N of 30, exactly as the acceptance page's own arm-state chip says. The remedy for an unanswerable improvement question was never a better test; it was running the instrument that produces the arms. |
| 2026-09-04 | GR1 | retract | I PUBLISHED TWO CLAIMS AN HOUR AGO THAT THIS CONF REFUTES, AND THE REFUTATION WAS ALREADY WRITTEN DOWN. (1) I read verdict=TIER-UNADMITTED on all 18 rows as a defect in the admission receipt. IT IS NOT: knowledge/cjc_panel.conf admits the tier set {textured} DELIBERATELY -- clay has 0 GOOD rows and photoreal 0 BAD rows, so both abstain BY DESIGN -- and I supplied a list of BARE PATHS with no tier field, so every row took tier_default=clay, the unadmitted tier. The panel behaved correctly and my input caused the abstention. (2) I reported beach percept=68 and 94 against floor 152 as a measurement of the served page. IT IS NOT: the conf names this exact trap, TWO RENDERERS ONE WORD -- clay rows are nx_world_snap NATIVE voxel captures while the SERVED page is TEXTURED because its NPC fragment shader samples the atlas, and the conf further records that the only clay rows are 240x150 captures predating the 2026-08-28 resolution fix, re-scoring 102 to 131 at full resolution. A tier whose rows predate a resolution fix cannot be compared with anything captured after it. WHAT SURVIVES: the census itself, 18/18 judged, failed=0, partition reconciles, and the clean separation of world-path from character-path captures. WHAT DOES NOT: any claim that 68 grades the beach page. |
| 2026-09-04 | GR1 | measure | THE REAL BLOCKER ON "IS OUR ACCEPTANCE SOTA", AND IT IS DOCUMENTED IN THE CONF RATHER THAN HIDDEN: THE REFEREE'S ONLY VOTING LENS IS SELF-REFUSING AND THE ADMISSION RECEIPT CANNOT SEE IT. floor_skin was REMOVED on 2026-08-25 (nine of nine rule/region combinations not separable), floor_stats and floor_sym were never separable, so percept is the SOLE voting lens -- stated in the conf as a real reduction in strength. And nx_percept ladder, the judge's own leave-one-out self-check and the very condition its header says percept is admitted ONLY WHILE it holds, returns 12/13 verdict=RED. The receipt binds floors, skin_lens, skin_rule and the tier set but NOT the ladder state, so the panel prints admission=ADMITTED while the one lens carrying every vote is failing its own ordering test. A SINGLE-LENS REFEREE WHOSE LENS IS SELF-REFUSING CANNOT MAKE ACCEPTANCE SOTA, AND NO AMOUNT OF A/B MACHINERY DOWNSTREAM REPAIRS IT. This is the rung to fix before any improvement verdict published from this panel means anything, and it must be fixed by earning a second lens a floor or repairing the ladder -- never by widening the ladder or dropping its check, which is loosening a gate to flatter a number. |
| 2026-09-04 | GR1 | measure | VERIFIED LIVE 2026-09-04, not quoted from the conf header, because a header is not a measurement and that claim was dated 2026-08-28: nx_percept ladder returns 12/13 REFUSED verdict=RED TODAY. The numbers: flat=-342 squares=-167 m2d=283 craft=196. L1 flat<squares GREEN, L2 squares<m2d GREEN, and all TEN held-out commercial references pass their leave-one-out (loo 100 to 372, every one above squares). THE SINGLE FAILING CONJUNCT IS L3, m2d<craft, WHICH IS THE OPERATOR ORDERING: the judge scores m2d 283 ABOVE craft 196 and the operator ranks craft higher. So the sole voting lens of the sole referee fails exactly the one check that encodes the human judgement it exists to reproduce, while the admission receipt -- which binds floors, skin_lens, skin_rule and the tier set but NOT the ladder -- keeps printing ADMITTED. The gap is 87 permil on one ordered pair, it is NOT a broad disagreement with humans, and that is why widening the ladder would be the wrong fix: eleven of twelve orderings are already right. |
| 2026-09-04 | GR1 | measure | L3 ROOT-CAUSED BY BUILDING THE MISSING INSTRUMENT, AND IT REFUTED MY OWN HYPOTHESIS. nx_percept gained an additive read-only `anchors <frame>` verb printing cosine to all 12 banked anchors with their class; shipped through the guarded lane (content=GREEN, behave=IDENTICAL, live_sha==staged_sha, _offc twin byte-identical, 31896 to 38600 B). I EXPECTED A SUBJECT MISMATCH -- 9 of 10 GOOD anchors are commercial CHARACTER renders and craft is a world frame -- AND THAT IS WRONG. MEASURED: craft GOOD-mean 866 versus m2d 894, only 28 permil apart, and veloren1 (the one voxel-world anchor) is craft's SECOND-HIGHEST GOOD match at 926, so the GOOD side is working exactly as designed. THE GAP IS ON THE JUNK SIDE. Of the 66-permil score difference, 28 comes from GOOD and 38 from JUNK, and within JUNK it is flat_control ALONE: craft-to-flat 617 versus m2d-to-flat 476, a 141-permil gap, while on squares_control m2d is actually the CLOSER of the two (672 versus 608). THE FLAT-COLOUR CARICATURE CONTROL IS PENALISING craft FOR ITS SHADING MODEL RATHER THAN ITS QUALITY: a flat-shaded voxel world is legitimately close to flat colour in hue-histogram and run-length statistics, but that anchor was banked to represent NO CONTENT, not A FLAT-SHADED RENDERING STYLE, and the ruler cannot tell those two apart. |
| 2026-09-04 | GR1 | measure | L3's CAUSE IS ONE FEATURE FAMILY, MEASURED BY A NEW `families` VERB, NOT ARGUED. nx_percept gained `anchors` and `families` (both additive, read-only, shipped through the guarded lane: content=GREEN both times, live_sha==staged_sha, 31896 to 38600 to 41216 B) plus an EXTRACTED pj_centroids so the scorer and the decomposition cannot disagree about the centroids. NEUTRALITY PROVEN ON REAL DATA, not asserted: after the extraction `ladder` reprints flat=-342 squares=-167 m2d=283 craft=196 and all TEN leave-one-out values byte-identical (294 163 346 100 347 321 246 336 372 341). THE DECOMPOSITION, diff = cos_good minus cos_junk per family: hue 7 dims craft -161 versus m2d +490, a 651-permil swing to m2d; edge 4 dims craft +178 versus m2d -34, a 212-permil swing to CRAFT; blockdelta 24 dims +59 vs +68; runlength 8 dims +872 vs +907. ONE FAMILY CARRIES THE ENTIRE DISAGREEMENT. craft's hue vector is CLOSER TO THE JUNK CENTROID THAN TO THE COMMERCIAL ONE (junk 812, good 651) because a voxel world's palette IS a handful of flat block colours, which is statistically what the flat-colour control is. And the family that AGREES with the operator is edge, where craft leads by 212 -- but edge is 4 dims of 43, so it is outvoted 39-to-4 BY CONSTRUCTION. |
| 2026-09-04 | GR1 | lesson | THE REMEDY L3 NEEDS IS NEITHER OF THE TWO THAT LOOK OBVIOUS FROM A SCORE, AND THE DECOMPOSITION IS WHAT SEPARATES THEM. NOT a floor to retune -- the floor is not in the disagreement. NOT an anchor to re-bank -- I hypothesised a SUBJECT MISMATCH (9 of 10 GOOD anchors are commercial CHARACTER renders, craft is a world frame) and the per-anchor verb REFUTED IT: craft GOOD-mean 866 versus m2d 894, and veloren1, the one voxel-world anchor, is craft's second-highest match at 926. The real shape is TWO STRUCTURAL FACTS: (1) HUE IS CONFOUNDED WITH THE JUNK CONTROL -- flat_control exists to represent NO CONTENT, and a flat-shaded rendering STYLE is statistically indistinguishable from it on a 7-bin hue-order histogram, so every flat-shaded render is scored as partly junk for its shading model rather than its quality; (2) EDGE IS UNDER-DIMENSIONED FOR ITS DISCRIMINATIVE VALUE -- it is the only family that reproduces the operator ordering and it carries 4 of 43 dims, so a family that is RIGHT is outvoted by three that are confounded or near-tied. ★★★★★★A COSINE OVER CONCATENATED HISTOGRAM FAMILIES WEIGHTS EACH FAMILY BY ITS DIMENSION COUNT, WHICH IS AN ARTEFACT OF HOW THE FEATURE WAS WRITTEN AND NOT A STATEMENT ABOUT WHAT DISCRIMINATES -- so the family carrying the human judgement can be correct and still lose every vote. Fixing this by widening the ladder or dropping its check would be loosening a gate to flatter a number; the honest rungs are a junk control that is degenerate WITHOUT being flat-shaded, and a per-family weighting earned on a labelled corpus rather than inherited from dimension counts. |
| 2026-09-04 | GR25 | measure | THE ACCEPTANCE PLANE IS THE GOAL'S REAL WORKLIST AND IT ALREADY HOLDS THIS SESSION'S COMPLAINT VERBATIM, FILED 2026-08-30. knowledge/status/accept_queue.status reads awaiting_operator=1 rejected_awaiting_seat_rework=4 awaiting_seat_or_referee=0 verdict=GREEN, page live and 3465 s fresh. The five rows: beach-garment-objects REJECT (garments as mesh objects), beach-skin-relief UNJUDGED (micro-relief staged in TEXR, not yet on the served page, must be real texture not smooth painting), beach-hair-colour-wind REJECT (HAIR CLIPS THE BODY and moves like a mop, not strand by strand), beach-feet-planted REJECT, and beach-photorealism-overall REJECT whose operator words are the answer to surpass-MetaHuman: "is this truly getting to photorealism -- no: flat light, no shadows/SSS, doll faces painted not modelled, wig-shell hair, painted nipples, cone skirt, voxel sand; PS2-era look, far from MetaHuman/UE5". SO THE OPERATOR'S 2026-09-03 JUDGEMENT WAS NOT NEW INFORMATION -- IT WAS A STANDING REJECT ROW NOBODY HAD CLOSED, WITH SEVEN DEFECTS ALREADY ENUMERATED AND THEIR SYMBOLS ALREADY NAMED. Today's rigview fix closes the CLOTH half of the hair row's clipping (the collision proxy now carries the soft-tissue term); the HAIR half is untouched, and nx_hairdyn_lib hd_tick has NO COLLIDER AT ALL -- it passes only a maxd radial clamp. |
| 2026-09-04 | GR25 | land | THE HAIR HALF OF THE OPERATOR'S CLIPPING REJECT IS SOURCE-LANDED AS ONE SHARED COLLIDER, NOT A FOURTH COPY. nx_vecmath gains vm_capsule_push (17806 to 20258 B): point-versus-capsule with the surface point written out and 1 returned only when the point was inside, so a caller pushes N points against M capsules without a sentinel. IT LIVES THERE BECAUSE THE ARITHMETIC ALREADY EXISTED THREE TIMES -- inline in nx_softjiggle's sj_step at Q256 (the only one that actually pushes), and as ss_capsule in BOTH nx_skullsdf trees at Q1024 (distance only) -- while nx_vecmath, which owns the vm_isqrt all three call, had none. Q256 is ADOPTED from sj_step rather than re-derived so the migration is byte-comparable instead of a second opinion about the same geometry. nx_hairdyn_lib gains hd_tick_collide (13520 to 16382 B) taking a flat 7-i64-per-capsule array. THE DEFECT IT ANSWERS, STATED PRECISELY: hd_tick's ONLY spatial bound is maxd, a RADIAL CLAMP about the anchor -- a clamp limits how far a segment travels and says NOTHING about what it travels through, and the two are easy to confuse because both bound a position. hd_tick IS UNCHANGED AND STILL SHIPS: this is a SIBLING because the bake path and nx_hairdyn_gate's ringdown and phase-lag teeth assert hd_tick's exact output, so folding a collider in would have moved every baked frame and every asserted number -- a behaviour change wearing a bug fix's clothes. Its return is CONTACTS RESOLVED, not defects found, and the header says so with the __nxa.clip precedent named, because that exact counter was misread as a defect count this same day and driving it to zero would delete the collision response. BUILD-UNPROVEN: three paced build attempts refused by admission (run-queue saturation then procs_blocked=8 of 8, load 19.31); queued as BQ-nx_percept and re-issuable, and hammering a storming box is itself a defect by standing order. |
| 2026-09-04 | GR25 | measure | CORRECTING MY OWN ROW FROM MINUTES AGO: THE HAIR COLLIDER IS NO LONGER BUILD-UNPROVEN. The previous land row said three paced attempts had been refused by admission; the fourth, paced ~8 minutes later, returned VERDICT=GRANT (load1 1524 under a 1643 ceiling, procs_running=1, procs_blocked=5 of 8) and nx_hairdyn_gate COMPILED: 46824 B, sha 0196dacea4338838fb0e422744fcb8b094f9b23f039cec74e5d1a9796e888c4d. That gate's closure carries BOTH new pieces, so vm_capsule_push in nx_vecmath and hd_tick_collide in nx_hairdyn_lib are compile-proven together. STILL HONESTLY OPEN: the gate BINARY is built but not promoted, so its teeth have not been RUN against the new lib -- and its teeth measure hd_tick, which is deliberately unchanged, so a GREEN from it would say nothing about the collider anyway. The collider owes its OWN tooth: a strand seeded inside a capsule must end OUTSIDE it, with a neg-control proving the tooth fires when the push is removed. Until that exists this is a compiling capability, not a proven one, and the acceptance row it serves stays REJECT. THE SAME ADMISSION STORM IS WHY nx_percept's ladderx is still unshipped: four ship attempts, the last two never reaching their build stage, so the leave-one-family-out measurement that would confirm or refute the hue-confound hypothesis is still owed. |
| 2026-09-04 | GR25 | measure | THE COLLIDER'S OWN TOOTH EXISTS, RAN, AND CAUGHT A REAL DEFECT IN THE CODE I SHIPPED AN HOUR EARLIER. nx_haircollide_gate (7450 to 8413 B, built 37223 B, run-exit=1) reports passed 11/12 verdict=RED. THE EIGHT GEOMETRY TEETH ALL PASS AT EXACT VALUES: a point inside reports a contact and lands ON the surface (100 against r=100, tol 2), the push is purely radial so the axial coordinate is untouched (500 unchanged), a point OUTSIDE reports no contact AND leaves the out-buffer unwritten, the on-axis degenerate case takes the documented +x, and a point 500 beyond the endpoint is OUTSIDE the capsule rather than inside the infinite line (the commonest way this arithmetic is written wrong). BOTH CONTROLS PASS: ncap=0 resolves nothing, and without the collider at least one segment IS left inside, so the tooth can fail. THE FAILING TOOTH IS THE END-TO-END ONE: WITH-the-collider-NO-segment-is-left-inside actual=4 expected=0, identical to the no-collider control. THE DIAGNOSTIC VALUES NAME THE SHAPE: seg0 after the collided tick is x=0 y=978 dist_to_axis=0, against x=0 y=880 without it, with contacts_RESOLVED=4 and SD_STRIDE=6. So the collider IS changing the system (y moved) and IS firing (4 contacts), but the radial correction is NOT SURVIVING to the end of the tick -- a push of (0,880,0) against a y-axis capsule must land at (100,880,0), x moving and y not, and the measurement shows the opposite. THREE CAUSES REMAIN CONSISTENT WITH THAT AND THEY HAVE DIFFERENT REMEDIES: the pushed position being overwritten by the next segment's sd_step_axq, a slot-convention mismatch inside the 6-wide stride, or the spring re-clamping after the push. NOT GUESSED -- the next cycle prints the per-iteration positions and separates them. THE ACCEPTANCE ROW beach-hair-colour-wind STAYS REJECT, AND THAT IS THE GATE WORKING: a compiling collider that leaves every segment inside the body is exactly the state a green-by-default tooth would have shipped as done. |
| 2026-09-04 | GR1 | land | FIVE EXTERNAL LIAR-KILLERS MIRRORED AND CONTENT-PINNED, 2026-09-04, on the operator's framing: "i dont want these built into our build pipeline but they should all stand as liar killers and benchmarks and oracles ... to help us avoid long costly unproductive building by giving us frames of refrence". PINS, all status=200 with provenance rows: webgpu_inspector 363369 B h3b560de731a812469f4d22b1f1017327fe9ec581d6753130a4a839f7f94fe441 -- binaryen 519176 B h6aa2d7836ec42efa13c301e5ca617417cc1c2fcd38dc7112e170431b1e808c51 -- emscripten 382705 B he883dd0ee08c7482e37e73991ef50f798238ab0ba26cd1784a008ce9e38542be -- meshoptimizer 689176 B h1d3e5434a035ee0b4f397cdaa8bfe95f58462acea8bd33c653ef23032377719d -- dawn 14923 B h44aaee866414aba755aa1646b193af8f8eecd75544e694c5ca6d549b676f7a77. WARNING ON THE LAST ONE: 14923 B is small enough to be the stub signature (login wall, JS shell, bot challenge) and it is NOT yet verified as a real index; it is recorded UNVERIFIED rather than cited, because a wrong mirror is worse than none. THE CLASSIFICATION MATTERS MORE THAN THE LIST, and it turns on a fact the industry survey assumed wrongly: OUR CORE ENGINE IS NISHILANG, not C++, Rust or TypeScript. So emscripten (compiles C/C++), wasm-bindgen (Rust) and esbuild/Vite (a JS/TS host layer we deliberately do not have) CANNOT BE TOOLCHAIN HERE -- they are BARS ONLY. What transfers as an executable liar-killer is narrower and better: binaryen wasm-validate can FALSIFY "our emitted wasm is spec-valid" and wasm-opt can falsify "our output is tight"; webgpu_inspector can falsify "the GPU receives the commands we claim"; Dawn plus RenderDoc or PIX can falsify a frame-time claim at hardware level. RANKED BY WASTED BUILDING PREVENTED, binaryen is FIRST: it answers whether nx_compile_wat's output is valid at all, in one call, against an emitter I edited today for LN35, and craft_mt.wasm (263990 B, already served) is a subject sitting right there. ★AND THIS IS THE SAME DEFECT THE WHOLE SESSION HAS BEEN FINDING FROM THE INSIDE: the referee has ONE voting lens, it disagrees with the operator, and there is NO SECOND METHOD CLASS -- external oracles ARE that second class, which is why they belong on the board as data rather than in anyone's head. |
| 2026-09-04 | GR25 | land | RED TO GREEN ON THE PREDICTED VALUES: nx_haircollide_gate 12/12 verdict=GREEN, run-exit=0, and the hair half of the operator's clipping REJECT is now a PROVEN capability rather than a compiling one. THE DEFECT WAS THE Q256 PARAMETER QUANTISATION IN vm_capsule_push, AND THE GATE'S OWN EMITTED VALUES SOLVED IT BY HAND BEFORE ANY GUESS: for seg0 at (0,880,0) against the axis (0,0,0)-(0,1000,0), t = 880*1000*256/1000000 = 225.28 TRUNCATED TO 225, giving a closest point of (0,878,0) instead of (0,880,0), an offset of (0,2,0) whose direction is +Y rather than radial, and a push to (0,978,0) -- which is the EXACT value the failing run measured. On an axis of length L a Q256 parameter resolves the closest point only to L/256, here 3.9 units, so ANY point nearer the axis than that error gets a direction dominated by axial truncation noise and is pushed ALONG the capsule instead of out of it. FIXED AT THE ROOT, NOT WITH A TOLERANCE: exact integer num/den with the numerator clamped to [0,den], one divide, no quantisation step at all, plus a named den==0 branch for the degenerate sphere case. AFTER: seg0 x 0 to 100, y 978 to 880 (axial coordinate now untouched, as a radial push requires), dist_to_axis 0 to 100 = exactly r, segments_left_inside 4 to 0, and the no-collider control still reports 4 so the tooth can still fail. ★★★★★★A QUANTISED PARAMETER DOES NOT DEGRADE GRACEFULLY WHEN THE QUANTITY IT MEASURES FALLS BELOW ITS STEP -- IT POINTS THE WRONG WAY WHILE STILL REPORTING SUCCESS. ★AND THE CONTACT COUNTER READ 4-OF-4 RESOLVED THROUGHOUT, i.e. the collider reported complete success on every segment while leaving every segment inside the body: the same counter-name trap as __nxa.clip, seen from the other side -- a SUCCESS count that is true and useless. sj_step never exposed this because its points are skin-surface against limb axes and never sit within a few units of one; hair hanging straight down the capsule is precisely the pathological case, which is why the incumbent's Q256 was safe to adopt and unsafe to keep. |
| 2026-09-04 | GR25 | land | GE21 PREPARATION CAUGHT A DEFECT IN MY OWN HOUR-OLD CODE, AND IT WAS INVISIBLE ON THE NATIVE TARGET. Checking whether nx_wasm_craft could simply import nx_hairdyn_lib (it imports nx_softbind but NOT the solver; its only hair symbols are rig_hair_y and rig_hair_r, i.e. GEOMETRY, so the chain integration genuinely lives only in the emitted JS and GE21/GE22 is real) surfaced this: hd_tick_collide called sys_mmap(24) PER INVOCATION. The emitted $nx_wasm_mmap is a STATELESS GROW-PER-ALLOCATION helper -- pages = ceil(n/64K), base = memory.grow(pages)*64K -- so on the wasm lane that is roughly 720 memory.grow calls per second at 60 Hz across a 12-girl cast, unbounded, on the exact path GE21 exists to move into wasm. IT WOULD NOT HAVE APPEARED IN THE NATIVE BUILD AT ALL. ★★★★★★NEVER ALLOCATE IN A HOT LOOP -- AND "HOT" IS A PROPERTY OF THE TARGET, NOT OF THE SOURCE: the same function is harmless natively and catastrophic in the browser, so a review that only ever reads the source cannot see it. FIXED by taking a caller-owned scratch buffer as a parameter rather than hiding an allocator inside, with the reasoning written beside it; the gate allocates it ONCE, because a gate that allocated per call would hide the very property the parameter exists to guarantee. NEUTRALITY PROVEN ON REAL DATA: 12/12 GREEN again with EVERY emitted value byte-identical to the pre-change run (seg0 100/880/0, dist 100, inside 0 with the collider and 4 without, contacts 4). ⚠DECLARED, NOT SILENT: this run also printed DEPLOYED to serving root -- because the gate now HAS a live twin from its promote, the default nx_sov_build_run refreshed it. That is the documented twin-refresh behaviour and the deployed artifact is the GREEN one matching source, but an unannounced deploy is exactly what the record warns about, so it is named here rather than left to a later census to discover. |
| 2026-09-04 | GR25 | retract | I BUILT A FIFTH COPY OF THE CAPSULE COLLIDER WHILE TELLING THE OPERATOR I WAS CONSOLIDATING THREE, AND THE ONE I MISSED IS THE ONE THAT SHIPS. nx_wasm_craft.nx carries hg_capd2 / hg_inside / hg_push -- a full capsule collider with HG_CAP_MAX=32 slots (head, neck, 2 shoulders, 2 arms, torso, DYNA anchors) in HG_CAP_W=10 format (ax ay az bx by bz r kind sx sy) -- and it references vm_capsule_push exactly ZERO times. WHY THE SEARCH MISSED IT: I grepped for `func .*capsule`, and NOT ONE of the shipping implementation's identifiers contains the word. That is this estate's own law firing on me -- A MARKER LIST KEYED ON THE VOCABULARY OF THE THING YOU ARE DETECTING CANNOT SEE INSTANCES EXPRESSED IN OTHER VOCABULARY, AND THE MISS READS AS "THIS DOES NOT EXIST" RATHER THAN AS "MY DEFINITION IS TOO NARROW". The check-before-build rule was followed in form and failed in substance. ★AND IT REFRAMES THE RUNG, WHICH MATTERS MORE THAN THE DUPLICATE: the operator's REJECT says "hair clips the body", and I read that as A COLLIDER IS MISSING. IT IS NOT. A collider is present, wired, and shipping to the browser; the defect is that IT IS NOT GOOD ENOUGH. Those are different rungs with different work. WHAT THE COMPARISON IS WORTH KEEPING FOR: hg_capd2 quantises its segment parameter exactly as my first cut did, at HG_Q12=4096 rather than Q256 -- the SAME defect class, 16x finer, so the near-axis direction error is L/4096 instead of L/256. And hg_push's degenerate branch (d < 1) pushes along +z, THE STATURE AXIS: correct for the arm capsules, which are roughly horizontal, but PARALLEL TO THE AXIS for the head, neck and torso capsules, which are the ones hair actually lies against -- so an exactly-on-axis point there is pushed ALONG the collider and stays inside, the same failure nx_haircollide_gate caught in mine. ⚠NOT CLAIMED AS THE OPERATOR'S DEFECT: d<1 needs a guide almost exactly on a collider axis and hair roots sit on the scalp SURFACE, so the branch is latent rather than shown to be reached. Measuring it is the next step, not asserting it. ⇒ THE REMEDY IS NOW CLEAR AND SMALLER THAN GE21: have hg_capd2/hg_push COMPOSE vm_capsule_push so there is one ruler, which also puts nx_haircollide_gate's twelve teeth over the SHIPPING path instead of only over the native solver. |
| 2026-09-04 | GR25 | measure | BOTH OF MY CONCERNS ABOUT THE SHIPPING COLLIDER ARE REFUTED BY ONE CONSTANT, AND THE REAL MECHANISM IS SOMEWHERE ELSE. HG_UNITS_PER_178CM=171530 means ONE UNIT IS ABOUT 0.01 mm. So (a) hg_capd2's HG_Q12=4096 segment-parameter quantisation costs L/4096 = 42 units = 0.44 mm on a stature-length capsule -- against the 670 units = 7 mm my own Q256 cut cost, which is why mine broke and theirs does not; Q12 was chosen fine enough ON PURPOSE and I was about to "fix" working code. And (b) hg_push's +z degenerate branch needs a guide within d<1 = 0.01 mm of a collider axis, which is never reached by scalp-rooted hair. THE ENGINE'S COLLIDER IS SOUND ON BOTH AXES I QUESTIONED. ★SO WHERE THE CLIPPING ACTUALLY COMES FROM, READ OFF THE RIBBON BUILDER: collision runs on the 128 SIMULATED GUIDES, but the page renders 1200 strands as SIX INTERPOLATED CHILDREN PER GUIDE, and each child is displaced laterally by a hash offset -- ox9 = hsh9(gidx9,k5)*w09*1.2 and oz9 = hsh9(gidx9+7,k5)*w09*1.2 -- APPLIED AFTER THE GUIDE WAS COLLIDED, WITH NO RE-CHECK. With w09 = max(900, span*0.045) that is roughly 900 to 1600 units before the 1.2 factor, i.e. about 9 to 17 mm of UN-COLLIDED lateral displacement on every rendered strand. That is far more than any plausible body clearance and it is enough on its own to push hair through the body. ★★★★★★A COLLISION PASS THAT RUNS ON THE SIMULATED SET AND NOT ON THE RENDERED SET IS NOT A COLLIDER FOR ANYTHING THE OPERATOR CAN SEE -- AND IT LOOKS PERFECT IN EVERY TEST THAT MEASURES THE SIMULATED SET, WHICH IS EVERY TEST ANYONE WOULD NATURALLY WRITE. My own nx_haircollide_gate has exactly this blind spot: it measures chain segments, so it would stay 12/12 GREEN with the render path clipping visibly. ⇒ THE REMEDY IS ORDERING, NOT ARITHMETIC: apply the child offset BEFORE the collision pass, or re-collide each child after offsetting, or bound the offset by the guide's measured clearance. Cheapest correct form is to collide the child, since the offset is deterministic from the hash and needs no new state. NOT YET IMPLEMENTED -- named with its evidence so the next cycle starts from the mechanism instead of from the symptom. |
| 2026-09-04 | GR25 | land | THE CHILD-OFFSET FIX IS SOURCE-LANDED, BUILT, ADJUDICATED AND PROMOTED; THE PAGE RE-EMIT IS THE ONLY STEP LEFT. nx_game_page_emit.nx 358904 to 360131 B: the child offset is now clamped so it cannot move a strand INWARD -- rq9/nrx9/nrz9 from the strand root's own xz give the radial direction about the head axis, and only the negative radial component is removed, so all lateral variety survives while a child can never end nearer the body than its collided guide. Built 430126 B; shipped and PROMOTED, live 425241 to 451160 B. THE ADOPTGATE REFUSED FIRST AND EVERY ONE OF ITS FOUR LOST RUNS WAS ADJUDICATED BEFORE ANY FLAG WAS PASSED, because three of them were NOT MINE and that is exactly the AHEAD-class shape where promoting DESTROYS capability: (1) the const-ox9 line, mine, const to let; (2) the __nx_bootfail stall banner, mine, the webgpu branch added earlier today; (3) `then(async r=>{const ex=` -- the SOURCE reads `let ex`, because GE30 threading reassigns ex to the threaded instance's exports, so this is a SIBLING SEAT'S reword; (4) the _skipdraw render/blit line, same threading work. ALL FOUR ARE REWORDS WITH SOURCE AHEAD OF LIVE, i.e. the live emitter was simply STALE -- it predated the GE30 threading, my stall-banner edit, the sky extraction and this clamp. So this is BEHIND, not AHEAD, and promoting UPDATES rather than destroys. Only then was allow_loss=4 passed, with all four named. ★THE HABIT THAT MATTERS: THREE OF THE FOUR LOST RUNS WERE NOT MINE, AND A BLANKET allow_loss WOULD HAVE DELETED A BOOT-FAILURE HANDLER, A STALL BANNER AND TWO THREADING LINES WITHOUT ANYONE NOTICING. Enumerate before allowing, every time. OWED: nx_craft_emit re-emits the six world pages and was launched; its stage-1 native gate passed 60+ teeth and the 3b wasm-vm gate passed FITS and PAINTS, then output stopped advancing at the PAINTS-REAL precondition under host load. beach.html is still 55562 s old, so THE FIX IS NOT YET ON THE SERVED PAGE and beach-hair-colour-wind stays REJECT until a re-emit lands and a frame shows otherwise. |
| 2026-09-04 | GR48 | measure | THE WASM HAIR ENGINE ALREADY CARRIES THE FOLLOW-STRAND SCHEME AND THE PAGE REIMPLEMENTS ITS SELECTION IN JAVASCRIPT. Seven measurements, each with its instrument, taken 2026-09-04 under the operator order that evidence accompanies every claim everywhere. ONE: nx_wasm_craft.nx:298 const HG_REND_MAX 1200 with the comment render strands the page's shipped NXH_GUIDES budget PRESERVED and moved here. TWO: :299-302 const HG_SIM 128 commented PICKED named as picked, 1200 over 128 is 9.4 render strands per guide, inside the TressFX 4 to 16 follow-hairs band. THREE: :306 const HG_TAB_W 8 with the render-strand table row spelled rest-idx nb1 nb2 w1 w2 lockw spanz 0, so two neighbour guides and two weights per rendered strand already exist as data. FOUR: :5326-5328 hg_x hg_p hg_d are each based at k times HG_SIM times HG_W times 8, so simulated positions are stored for 128 guides and never for 1200. FIVE: HG_TEX_SR is DEFINED at :303 and referenced only inside the comment at :302, matches=2 corpus_complete=1, i.e. the constant that sizes the guide texture has ZERO code uses -- the same defined-and-never-referenced class GR37 and the layout-token finding already name. SIX: the wasm copies render strands with src = r*nstr/nrend at :5532 while the emitted JavaScript computes s5 = Math.floor(q5*nstr/nsel) at nx_game_page_emit.nx:889 -- ONE selection rule implemented twice in two languages, and the wasm comment says so itself, the page's own selection rule. SEVEN: hverts, the hair vertex buffer name, returns matches=8 corpus_complete=1 and every hit is in the emitter. DELIBERATELY NOT CONCLUDED: whether some wasm export feeds hair positions to the page under a name not yet searched. Until that is measured this row does NOT claim the collision fails to reach the screen; it claims only the seven facts above. THE NEXT MEASUREMENT IS NAMED SO THE NEXT SEAT DOES NOT REDERIVE IT: enumerate the wasm exports nx_compile_wat emits for nx_wasm_craft and check every one against the emitted page's ex. call sites, then read hg_tick at :6023 to see what it writes. CONSEQUENCE FOR THE RUNG: if the follow-strand table is unconsumed then GR48 is far smaller than scoped -- delete the JS duplicate and consume hg_tab, rather than port a pipeline that already exists in NishiLang. |
| 2026-09-04 | GR48 | retract | I REBUILT THE WRONG BINARY, AND THE BUILD STAMP I CITED AS PROOF WAS EVIDENCE ABOUT A BINARY THAT NEVER RAN. Published earlier today: the clamp IS live, proven by the served title carrying build stamp b1788529014 matching beach.html mtime. FALSE, and the discriminating experiment is structural rather than textual: sites/nishifamily/world/beach.html:2149 reads const ox9=hsh9(...) with line 2150 the segment loop ADJACENT, while the source at buildroot/runtime/nx_game_page_emit.nx:889 reads let ox9= followed by the clamp, and the clamp MUTATES ox9 with ox9-=rad9*nrx9 which is impossible under const. The same pre-clamp const form sits in craft.html:2153 vale.html:2149 isle.html:2151. ROOT CAUSE, found by a cross-lane critic and by no single lane: nx_game_page_emit IS A LIBRARY imported in-process, NOT the organ that writes the page. The writer is nx_craft_emit, which declares CE_OUT2 as sites/nishifamily/world/beach.html and calls gpe_build from that lib. I edited the LIB and promoted the LIB's standalone binary, which emits nothing for the worlds; nx_craft_emit.elf carries mtime 1788472251, built 56164 s or 15.6 HOURS before the clamp existed, and beach.html was then re-emitted at 1788529014, 599 s AFTER my edit, BY THAT SIXTEEN-HOUR-OLD BINARY. That is why the page reads freshly written and pre-clamp at the same time, and it is why the mtime ordering source then elf then page looked like it refuted a stale-artifact story: the ordering was true of the wrong artifact. LESSON, and it is the general one: A BUILD STAMP PROVES THE PAGE WAS REGENERATED, NEVER THAT THE CODE YOU EDITED WAS IN THE THING THAT REGENERATED IT -- and before rebuilding, ASK WHICH ORGAN DECLARES THE OUTPUT PATH, because an emitter named for the page is not necessarily the writer of the page. All six worlds come from that one organ, craft beach vale via CE_OUT CE_OUT2 CE_OUT3 and isle crags ridge via the world_recipes.conf loop at nx_craft_emit.nx:618, so ONE rebuild and promote covers them. SECOND FINDING WHICH CHANGES GR48'S PREMISE: hg_tick, the sole entry point of the entire hg_ collision system, occurs EXACTLY ONCE across 23622 files with corpus_complete=1 -- its own definition, with NO CALLER. hg_solve_guide is reachable only from inside it. So HG_CAP_MAX 32 capsules, the 128 simulated guides, hg_push, hg_inside and hg_capd2 are UNREACHABLE DEAD CODE, and the groom baker nx_nxa_groom.nx contains no collide or capsule token either. The live per-frame hair is nxHairTick in the emitted page: spring, viscous damping, wind, length clamp, and NO BODY TERM OF ANY KIND. CONSEQUENCE: GR48 was scoped as move the geometry to where the collider can see it; the measurement says there is no collision running to see anything, so the rung must FIRST wire hg_tick and prove it fires, and the radial clamp is currently the ONLY body-awareness the rendered hair would have once the writer is rebuilt. THIRD FINDING, the operator's testing complaint made mechanical: beach.html:2809 const REV=location.search.indexOf(review=0)<0 means REVIEW IS ON BY POLARITY -- absent token gives -1 which is less than 0 -- and beach.html:2954 if(REV)setInterval(...,16) installs a 60 Hz look-at servo that writes through ex.look, the same door the mouse uses, sending dd/6 against a LOOKS gain of 6 so the division EXACTLY CANCELS and the full remaining angular error is applied every tick, clamped to plus or minus 1800 it-units of a 25736 it-unit turn, i.e. 25.2 degrees. Any mouse-look under about 25 degrees off-target is erased within ONE sim tick and larger deflections are slewed back, while summon re-forms the ring every 900 ticks. Position is yours, ORIENTATION IS TAKEN. review=0 is the only true escape and it also unpins the render scale that REV forces to full. The header comment claiming review=1 is a lever is STALE: no code tests for it. |
| 2026-09-04 | GR50 | retract | MY OWN DONE-RULE COMMITTED THE ERROR IT WAS WRITTEN TO STOP. GR50 as landed centres on a SERVED BYTE DELTA, and the operator named it immediately: believing byte changes equals functionality shipping. The direction I wrote is sound -- bytes did not move therefore FAILED -- but building the rung around the delta invites its converse, and the converse is exactly the fallacy I committed this morning when I read 425241 to 451160 B as delivery of a fix that was never in the emitted page. A BYTE DELTA IS A NECESSARY PRECONDITION AND NEVER A PROOF. CORRECTED DONE-RULE for os_served_delta: the loop resolves the target's served artifacts from the organ's own output declarations; an EMPTY set means the target is inert by construction, so it NAMES every importer and reports FAILED; and where the set is non-empty the ship passes only on a BEHAVIOURAL assertion about the served artifact that the pre-ship artifact demonstrably fails -- the artifact's own gate re-run against the new bytes, not the size of the change. The byte delta stays as a CHEAP PRECONDITION that can only ever refuse, never as the thing that grants. |
| 2026-09-04 | GR46 | measure | I LOOKED AT THE SERVED FRAME FOR THE FIRST TIME TODAY AND IT INDICTS THE WHOLE SESSION'S PRIORITY. METHOD, reproducible by any seat with the laptop wire and zero NAS load beyond one page fetch: python nishi-ops/ac_cdp.py https://nishifamily.com/world/beach?review=0 35 document.getElementById(c).toDataURL(image/png) piped to a file, base64 decoded locally -- 302621 B PNG, magic verified, canvas 519x260, npc_up=1, gl2=1, door webgl2, mt n=8 shared=true. WHAT THE FRAME SHOWS, and every item is visible in the capture rather than inferred: ZERO shadows and ZERO ambient occlusion, twelve figures none of which is grounded by a contact shadow; voxel terrain built from large cubes carrying one repeating tile, with stacked-block walls; skin flat and plastic with no subsurface and no specular variation; hair reading as a SOLID SHELL with no strand separation at this scale; no sky, only a flat beige horizon band. The cast pipeline itself WORKS -- twelve mobs, textured swimwear reading teal pink and red, plausible proportions -- so the failure is entirely SHADING AND ENVIRONMENT, not geometry plumbing. THE OPERATOR'S RECORDED REJECT WORDS ARE CONFIRMED VERBATIM AGAINST PIXELS: flat light, no shadows/SSS, wig-shell hair, voxel sand, PS2-era look far from MetaHuman UE5. ★★★★★★AND THE FINDING THAT MATTERS MOST: AT THIS QUALITY THE HAIR CLIPPING IS NOT VISIBLE. The hair does not read as strands, so the GR25 child-offset clamp -- a 9 to 17 mm inward correction on interpolated children -- corrects something no viewer can perceive. I spent a full session on a sub-millimetre geometric detail while the render fails at the level of lighting, materials and terrain, after the operator had already said below PS2 level in the first message. RANKED BY WHAT THE FRAME ARGUES, not by what I was working on: 1 shadows and AO, GR46 rk_shadow_cascade, because nothing is grounded and flat light reads as fake instantly and its own rung text already says the served beach draws dark blobs under feet; 2 the voxel terrain and its tiling texture, which reads as a 2003 block game regardless of character quality; 3 skin subsurface, GR29, because the plastic mannequin look survives any resolution increase; 4 a hair BSDF, GR16, anisotropic highlight and strand separation, which must precede any further hair GEOMETRY work; 5 sky and atmosphere, GR26. OWED AND NAMED RATHER THAN GUESSED: this capture is exactly the artifact the acceptance ledger lacks for its four ref_rows=0 subjects, and it CANNOT be delivered by this seat -- nx_accept referee takes a NAS-ABSOLUTE png path, nx_fs_write is text-only by design, and this seat holds no contentput.cap. THE UNBLOCK IS ONE CAPABILITY, NOT A DESIGN: grant contentput.cap or have a NAS-side beat take the browser capture itself. Until then the referee rows stay honestly empty rather than filled with a tier I cannot prove. |
| 2026-09-04 | GR1 | measure | THE ACCEPTANCE REFEREE IS A SINGLE-LENS JUDGE WHOSE ONLY LENS IS SELF-REFUSING, AND ADMISSION DOES NOT CHECK IT -- so every acceptance verdict today is unsound at the root and the operator's ask, validating whether we are improving in A B, is structurally unanswerable for a reason that is NOT missing captures. READ VERBATIM FROM knowledge/cjc_panel.conf, the panel's own data file, not inferred: nx_percept ladder, the judge's OWN leave-one-out self-check and the very condition the organ header says percept is admitted only while it holds, currently returns 12 of 13 verdict=RED with L3 failing m2d 283 greater than craft 196 against the operator ordering; and the admission receipt binds floors, skin_lens, skin_rule and the tier set but NOT the ladder state, so the panel reports admission=ADMITTED while its only voting lens is self-refusing. THE SINGLE-LENS PART IS ALSO MEASURED IN THAT FILE: floor_stats and floor_sym were never separable and are report-only, and floor_skin was REMOVED on 2026-08-25 after a nine-cell rule-by-region grid over the union corpus came back NINE OF NINE NOT SEPARABLE. So percept is the sole vote by elimination, and the file states the reduction honestly rather than hiding it. CONSEQUENCE FOR THIS BOARD: the /compare/acceptance ledger reads subjects=5 agreed=0 not_agreed=5 with four subjects at ref_rows=0, and the natural reading is that it is starved of captures. THAT READING IS WRONG AND WOULD SEND THE NEXT SEAT AT THE WRONG WORK: even a perfect capture graded today is graded by a judge failing its own ladder. TWO THINGS I ALSO SETTLED WHILE CHECKING THE TIER, both from the same file. ONE, THE TIER TRAP IS REAL AND ALREADY NAMED THERE: TWO RENDERERS, ONE WORD -- nx_world_snap's NATIVE path is flat lambert voxel with no atlas and is CLAY, while the PRODUCT is the served WebGL page which samples an npc atlas and is TEXTURED with floor 151 admitted at FP=0 FN=0 over 8 rows. Submitting a native snap as textured would grade the wrong renderer, so I did not. TWO, THE CLAY TIER'S NAMED BLOCKER IS NOW UNBLOCKED BY AN ARTIFACT ALREADY ON DISK: the file says clay is unadmitted with 2 BAD and 0 GOOD, that its only rows are 240x150 captures predating the 2026-08-28 resolution fix, and that GROWING CLAY MEANS RE-CAPTURING IT AT FULL RESOLUTION FIRST. This session's nx_world_snap char beach run returned w=1920 h=1200 q=1 res_permil=1000 res_class=FULL and wrote beach.crop.png with CHAR-CROP pixels=29857 fill_permil=671 -- a full-resolution clay capture, which is exactly the artifact named as missing. IT IS DELIBERATELY NOT LABELLED HERE: the corpus admits a row only on a named human witness and forbids relabelling to make a tier separable, so the GOOD or BAD call is the operator's and this row records the artifact, never a verdict. ORDERED CONSEQUENCE: fixing the ladder outranks collecting captures, and binding the ladder state INTO the admission receipt outranks both, because an admission that cannot see its lens self-refusing will re-admit the same broken judge on the next calibration. |
| 2026-09-04 | GR1 | measure | THE LADDER RED IS LIVE TODAY, NOT A STALE 08-28 CLAIM, AND ITS FAILURE IS ONE COMPARISON. Re-measured 2026-09-04 against the promoted binary rather than inherited from cjc_panel.conf, because a rung inherited without re-measurement cost this session a full investigation earlier today when GR39 turned out to be already fixed. VERBATIM: flat=-342 squares=-167 m2d=283 craft=196 with L1 flat<squares GREEN, L2 squares<m2d GREEN, L3 m2d<craft THE OPERATOR ORDERING RED, and all ten held-out commercial references passing their leave-one-out at loo=100..372 above squares. So the judge is internally consistent on every reference it holds and disagrees with the operator on exactly ONE pair: it ranks the 2D render ABOVE craft where the operator ranks craft above it. HYPOTHESIS, LABELLED AS ONE AND NOT PUBLISHED AS A FINDING, but newly available because this seat finally LOOKED at the served frame today: the capture is roughly four fifths repeating yellow voxel sand, and nx_percept's 43 dims are whole-frame statistics -- hue 0 to 6, blockdelta 7 to 30, edge 31 to 34, runlength 35 to 42. A score dominated by frame-wide hue and block statistics would rank a monochromatic tiled terrain low however good the characters in it are, while the operator judges the CAST. That is the SAME SUBJECT-MISMATCH CLASS as the other three defects measured today -- the collider that runs on simulated guides while the page renders interpolated children, the hair gate whose every tooth tests wiring while the operator judges pixels, and the ship loop that verifies a binary while the question is the served artifact. THE DECISIVE TEST IS NAMED AND CHEAP so the next seat does not re-derive it: score the CHARACTER CROP against the FULL FRAME and see whether the L3 ordering flips. The crop already exists at full resolution from this session, beach.crop.png, 122x526, fill_permil=671, CHAR-CROP pixels=29857. If the ordering flips on the crop the judge's subject is the frame and the remedy is a region, not a retune; if it does not flip the hue hypothesis is refuted and the fault is in the feature set. I COULD NOT RUN IT: the families decomposition verb was added to nx_percept source earlier this session and is NOT in the promoted binary, because build admission has refused every lane all afternoon on procs_blocked=8 of 8. ★AND THE ATTEMPT EXPOSED A SEPARATE DEFECT WORTH ITS OWN FIX: nx_percept SILENTLY ACCEPTS AN UNRECOGNISED VERB AND RUNS ladder INSTEAD, returning byte-identical output, so a caller cannot distinguish a verb that does not exist from a verb that agrees with the default. That is the estate's own permissive-default law -- an unknown mode must REFUSE or fail CLOSED, never fall through -- and it means every future probe of this organ can silently measure the wrong thing. |
| 2026-09-04 | GR46 | measure | THE FIRST NUMERIC A B BASELINE FOR THE BEACH, AND IT CORROBORATES THE PIXELS INDEPENDENTLY. nx_frame_score on beach.nxfh, the FULL-RESOLUTION native capture this session produced: w=1920 h=1200 pal=599 det1=11 det4=33 grad=35 cov=998 cohere=974 quality=500 verdict=BASIC. THE DELTA: GR38's log records the same ruler on the same subject class at quality=408 BASIC, so this reads +92. ⚠THE CAVEAT TRAVELS WITH THE NUMBER AND IS NOT OPTIONAL: cjc_panel.conf states that rows predating the 2026-08-28 resolution fix CANNOT be compared with anything captured after it, and GR38's capture resolution is unrecorded here, so +92 is a CANDIDATE improvement and not a confirmed one until that resolution is known. Whoever confirms it should record both resolutions in the same row. ⛔AND THE COMPARISON EVERYONE WILL REACH FOR IS FORBIDDEN BY THE RULER ITSELF: the random-squares control scores 628 against the beach's 500, and quoting that as quality is explicitly refused in the organ's own calibration line -- composite is a STRUCTURE floor, not a quality rank, and values order by statistics rather than perceived quality without a same-class perceptual comparison. A seat that publishes our world scores below random noise has committed the session's recurring defect, measuring a narrower thing than the claim. ★★★★★★THE COMPONENTS ARE THE VALUABLE PART AND THEY AGREE WITH THE EYE, WHICH IS THE INDEPENDENT CONSENSUS THE OPERATOR REQUIRES RATHER THAN A SEAT'S OPINION: det1=11 and det4=33 say there is almost no fine or coarse detail; grad=35 says the gradient field is nearly flat, which is SHADOWS AND FALLOFF BEING ABSENT expressed as a number; cov=998 and cohere=974 say the frame is near-totally covered and highly uniform. Low detail plus low gradient plus near-perfect coherence is the numeric signature of flat unshadowed untextured surfaces, and it was written down BEFORE this run from a browser capture of the served page -- twelve figures, none grounded by a contact shadow, voxel terrain carrying one repeating tile, hair reading as a solid shell. TWO INSTRUMENTS, ONE CONCLUSION, NEITHER DERIVED FROM THE OTHER. CONSEQUENCE FOR THE RANKING: grad=35 IS GR46 rk_shadow_cascade's case stated numerically, and it outranks every geometry rung on this board including the hair work this session spent itself on. THE BASELINE IS NOW REPRODUCIBLE BY ANY SEAT IN TWO CALLS: nx_world_snap char beach writes beach.nxfh at res_class=FULL, then nx_frame_score beach.nxfh prints the vector above -- so the next capture after a shadow landing is a direct A B against these exact numbers, on the estate's own ruler, with no eye involved. |
| 2026-09-04 | GR1 | land | A REFEREE VERDICT IS PUBLISHED TO /compare/acceptance FOR THE FIRST TIME THIS SESSION, AND THE LANE I DECLARED CLOSED WAS NEVER CLOSED. THE CORRECTION FIRST, BECAUSE IT IS THE REUSABLE PART: I published that this seat holds no contentput.cap and therefore could not deliver acceptance captures. FALSE. I had enumerated nishi-core/nxc2/*.cap, THIRTEEN files, and the real capability store is ~/.nishi/ with FOUR HUNDRED AND SEVENTY THREE caps including contentput.cap and a live admin.cap for nx_mgmt valid to epoch 1790000000. The estate's own memory note states the lesson verbatim -- BEFORE DECLARING A LANE CLOSED, ENUMERATE WHAT THE CAPS YOU HOLD CAN REACH -- and I quoted that note while reading the wrong directory. ★★★★★★A CAPABILITY INVENTORY TAKEN FROM ONE DIRECTORY IS A SAMPLE, AND A SAMPLE THAT COMES BACK EMPTY READS EXACTLY LIKE A DENIAL. THE WORKING LANE, END TO END, SO NO SEAT REDERIVES IT: capture the SERVED page in a real browser with nishi-ops/ac_cdp.py evaluating document.getElementById(c).toDataURL(image/png), redirect to a file and base64-decode locally so no image data crosses the context; upload with nx_content_put_client.py <local> <dest> ~/.nishi/contentput.cap; then nx_accept referee <subject> <NAS-ABSOLUTE-path> <tier> <seat>. ⚠THE DESTINATION IS THE ONLY TRAP AND BOTH REFUSALS NAME THEIR OWN REMEDY: content-put allows world/ knowledge/rigcorpus/ knowledge/fetched/ knowledge/library/ and /tmp/ while nx_mkdirp allows knowledge/ sites/ and /tmp/, so the intersection that ALREADY EXISTS and is DURABLE is knowledge/fetched/ -- choosing it avoids creating a directory merely to satisfy a path picked arbitrarily, and /tmp/ would be reaped by the 3600 s tmpstorereap beat. WHAT LANDED, VERIFIED FROM THE LEDGER AND NOT THE RECEIPT: ACCEPT-REFEREE subject=locomotion percept=339 floor=151 tier=textured composite=GOOD frame=sha256 619e543a222bcc573fb3b5d0c4830dff5d34ce9802d3440c3f043352f19755d8 capture_bytes=302621 id=ref-locomotion-1788533606, and a re-run of nx_accept list now reads referee=OK ref_rows=1 for that subject with the estate ledger moving ref_rows 1 to 2. Full lens vector recorded: cjp_stats=350 cjp_comp=360 cjp_pal=554 cjp_cont=350 cjp_face=1000 cjp_skin_center=856 cjp_fig_permil=874 cjp_sym=525 cjp_percept=339 cjp_vlm=-1 vlm=UNOBSERVABLE. ★SUBJECT FRAMING IS PART OF THE MEASUREMENT AND WAS CHOSEN, NOT DEFAULTED: the wide capture shows twelve figures mid-stride and is the RIGHT framing for locomotion and the WRONG framing for hair or skin, where each girl is roughly forty pixels tall -- grading strands from a wide shot is the subject-mismatch defect this whole session has been finding, so hair garments and skin wait for the review-framed close capture instead. ⚠TWO CAVEATS TRAVEL WITH THE GOOD AND ARE NOT FOOTNOTES: percept clears its floor but the judge producing it FAILS ITS OWN LADDER at 12/13 REFUSED measured today, so the row is honest evidence whose validity is bounded by that; and cjp_vlm=-1 means the VLM lens did not run, so this is a single-lens verdict exactly as cjc_panel.conf states. |
| 2026-09-04 | GR1 | measure | THE ACCEPTANCE REFEREE SCORES THE FRAME AND NOT THE SUBJECT, PROVEN BY THREE IDENTICAL NUMBERS, AND THE OPERATOR VERDICTS IT IS COMPARED AGAINST MAY NOT BE THE OPERATOR'S. TWO FINDINGS, both measured today, and together they say the acceptance surface cannot answer the question it exists for. FINDING ONE, THE STRUCTURAL ONE: four referee rows were landed from two browser captures of the SERVED textured page -- ref-locomotion-1788533606 percept=339 from the wide frame, then ref-hair-1788533863, ref-skin-1788533878 and ref-garments-1788533886 ALL percept=337 from the SAME review-framed capture sha256 52b5246d8a95a44138507b5c2f35827cc84210bf225816b888f19937904fb230. THREE SUBJECTS, ONE CAPTURE, IDENTICAL SCORES TO THE UNIT. nx_accept referee takes a subject argument but the lens measures the WHOLE IMAGE, so the subject selects which ROW IS WRITTEN and never what is MEASURED. ★★★★★★CONSEQUENCE: how is the hair, how is the skin and how are the garments are ONE MEASUREMENT WEARING THREE LABELS -- improve hair alone and all three move together, improve nothing and they still agree, so per-subject A B is impossible BY CONSTRUCTION and no amount of capture collection fixes it. This is the SAME CLASS as every other defect measured today and now the fifth instance: the collider runs on simulated guides while the page renders children; the hair gate's every tooth tests wiring while the operator judges pixels; the ship loop verifies a binary while the question is the served artifact; the percept ladder scores whole frames while the operator ranks casts; and now the referee scores a frame while the row claims a subject. IN EVERY CASE THE THING MEASURED IS NARROWER OR SIMPLY OTHER THAN THE THING CLAIMED. FINDING TWO, THE PROVENANCE ONE, AND IT IS WORSE: the operator stated plainly today that he did NOT reject hair. The accept- plane carries beach-hair-colour-wind-0830 REJECT 2026-08-30 actor=operator with the words poorly done; hair clips the body and moves like a mop, not strand by strand. The operator's ACTUAL recorded words that day, in CLAUDE.md, are ONE general statement about FOUR things -- these are all poorly done, doing it poorly isn't a complete, it's not complete till you and I and the nishi team agree. Two of the four rows echo that sentence almost verbatim; the HAIR row carries a SPECIFIC TECHNICAL DIAGNOSIS that appears nowhere in it. ★★★★★ONE UTTERANCE WAS SPLIT INTO FOUR VERDICTS AND EACH WAS GIVEN PER-SUBJECT WORDS ATTRIBUTED TO THE OPERATOR. nx_accept deliberately has NO verb that writes this plane, precisely because the operator owns it, but anything that can write the store can stamp actor=operator -- so the operator signature of the three-party agreement carries NO PROVENANCE beyond a self-declared field, and on this evidence it was synthesised. COST, MEASURED ON THIS SEAT: that row is why this entire session was spent on a hair child-offset clamp. A requirement no operator stated drove a day of work through his name. WITHDRAWN IN THE SAME BREATH: I published earlier today that the referee grading hair GOOD while the operator rejects it PROVES a judge-operator disagreement. It does not, because one side of that comparison may not be the operator's verdict. OWED, and neither needs a build: the referee must measure a SUBJECT REGION rather than a frame, and the accept- plane needs provenance a seat cannot forge. |
| 2026-09-04 | GR1 | land | PER-SUBJECT ACCEPTANCE MEASUREMENT IS ACHIEVED AND NEEDS NO BUILD -- AND THE SAME EXPERIMENT INVALIDATES THE FOUR ROWS I PUBLISHED AN HOUR AGO. THE FIX: nx_accept referee scores WHATEVER IMAGE IT IS GIVEN, so per-subject measurement is per-subject CROPPING, and the convention already existed -- the only pre-existing referee row, face, was graded from a HEAD CROP at ac_op_after_0828t1034_head.png. PROVEN, not proposed: the same review capture graded whole returns cjp_percept=337 for hair, skin AND garments identically, while a crop of the red-head's head and hair region -- box 348,100 to 440,215 of the 735x368 capture, upscaled 4x, uploaded as ac_beach_hair_crop_0904.png sha aa2a8a22 -- returns cjp_percept=226. THE WHOLE VECTOR MOVES, not just the composite: stats 166 to 16, cont 166 to 16, pal 615 to 289, sym 550 to 288. So the referee CAN discriminate subjects and the earlier finding stands in corrected form: it scores the frame it is HANDED, and handing it a frame for a subject question is the caller's defect rather than the organ's. ★★★★★★AND THE SAME NUMBER CONVICTS MY OWN ROWS. face, graded from a CROP, scored 214. This hair crop scores 226. Both close crops, same band. The four rows I landed earlier -- locomotion 339, hair 337, skin 337, garments 337 -- were graded from WHOLE FRAMES and sit roughly 110 points higher PURELY FROM FRAMING. cjc_panel.conf warns of exactly this in its own words, framing matters for the skin lens, COMPARE LIKE FRAMINGS, and the floor of 151 was earned on a crop-based corpus. ⇒THOSE FOUR GOOD VERDICTS ARE INFLATED BY FRAMING AND ARE NOT EVIDENCE THAT ANY SUBJECT CLEARS ITS BAR. They are honest captures measured the wrong way, and they must be re-graded from subject crops before any of them is read as a result. I am recording that against my own work rather than leaving four flattering numbers on the board. ★AND THE CROP SHOWED WHAT THE WIDE SHOT STRUCTURALLY COULD NOT: at 4x the hair mass passes THROUGH the shoulder and chest silhouette rather than draping over it, it reads as a painted shell with vertical banding rather than strands, the torso carries no form shading at all, and there is no shadow anywhere in the crop. The clipping is REAL and VISIBLE at close framing and INVISIBLE at scene framing -- which is why the review camera the operator dislikes for testing is the CORRECT camera for acceptance capture, and why a wide capture can never referee a strand-level subject. OWED NEXT, no build required: re-grade locomotion skin and garments from their own subject crops, and record the framing class ON every referee row so a crop score and a frame score can never again be compared as though they were the same measurement. |
| 2026-09-04 | GR1 | land | NINE REFEREE ROWS ARE IN THE PLANE AND THE PER-SUBJECT LIMIT IS NAMED HONESTLY RATHER THAN CLAIMED AWAY. STATE, read from knowledge/store/referee- which is the AUTHORITY, never from nx_accept list which lags it: face 214 head-crop 2026-08-31 by seat-accept-lane; then this session, locomotion 339 and hair 337 and skin 337 and garments 337 all FRAME-graded, then hair 226 head-crop, skin 229 and garments 229 and locomotion 229 all FIGURE-crop. WHAT IS PROVEN: the capture-upload-referee lane works end to end with content CAS at every hop, and DISTINCT CROPS PRODUCE DISTINCT SCORES -- 226 for the hair region, 229 for the figure region, 337 for the whole frame, with the entire lens vector moving and not merely the composite (stats 166 to 16, cont 166 to 16, pal 615 to 289, sym 550 to 288 between frame and hair crop). WHAT IS NOT PROVEN AND MUST NOT BE READ AS PROVEN: skin, garments and locomotion ALL read 229 BECAUSE THEY SHARE ONE FIGURE CROP. I made TWO distinct crops, not five, so I demonstrated discrimination between REGIONS and NOT between SUBJECTS -- three subjects still sit on one measurement wearing three labels, which is the very defect this work set out to fix, surviving one level up. TRUE PER-SUBJECT A B REQUIRES A CROP PER SUBJECT: torso skin, garment area, legs and stride. ★★★★★AND THE FRAMING CLASS MUST BE RECORDED ON THE ROW OR THE PLANE IS A TRAP: the floor of 151 was earned on a crop corpus, so the four FRAME rows at 337 to 339 are inflated by roughly 110 points of framing and are NOT evidence any subject clears its bar. They stay in the plane as honest captures measured the wrong way, superseded by the crop rows, and a framing field on every row is owed so a crop score and a frame score can never be compared as though they were one measurement. ★SIXTH INSTANCE OF THE SESSION'S ONE CLASS, AND IT NEARLY COST A FALSE CLAIM: nx_accept list reported ref_rows=6 and skin percept=337 while the referee- plane held the skin crop row at 229. TWO SPACED READS OF THE LEDGER AGREED WITH EACH OTHER AND BOTH WERE WRONG, exactly as the law predicts -- N agreeing reads from the SAME instrument are not independent evidence. I was one message from publishing that the crop rows did not land; the plane refuted it. ⇒/compare/acceptance CAN PUBLISH A SUPERSEDED REFEREE VERDICT while the current one sits in the plane, and that is a defect in the reader, not the writer. ⚠PROVENANCE, UNCHANGED AND STILL OPEN: ref-face-1788160423 carries actor seat-accept-lane, so even the pre-existing referee row was written by a lane rather than by a human witness, consistent with the accept- plane's actor=operator field being self-declared and unforgeable-by-nothing. |
| 2026-09-04 | GR46 | measure | THE FEET-PLANTED REJECTION IS A SHADING DEFECT MISFILED AS A LOCOMOTION DEFECT, AND THAT IS A THIRD INDEPENDENT LINE OF EVIDENCE FOR SHADOWS BEING THE TOP RUNG. MEASURED by cropping the served-page capture to the front figure's legs and feet, box 322,292 to 432,368 of the 735x368 review capture upscaled 5x: THE FEET ARE PLANTED. Toes are in contact with the sand, the geometry is correct, the ground intersection is right. AND THERE IS NO CONTACT SHADOW OF ANY KIND, so the figure reads as FLOATING despite being geometrically grounded; the legs carry a single flat tan with almost no form modelling. ★★★★★★THE ROW beach-feet-planted-0830 IS FILED AGAINST ga_accept_locomotion with targets charsim:ap_foot_ground and gameengine:mob_step_gait -- BOTH GEOMETRY ORGANS -- so a seat reading that row is sent to fix a foot-placement solver that is already correct, while the actual cause is an absent shadow pass. A MISFILED SUBJECT SENDS EVERY FUTURE SEAT AT THE WRONG ORGAN, and the row's own targets field is what does the sending. THREE INDEPENDENT WITNESSES NOW AGREE ON GR46 AND NONE WAS DERIVED FROM ANOTHER: nx_frame_score on the full native capture returns grad=35, a near-flat gradient field, which IS the absence of shadow and falloff expressed as a number; a browser capture of the served page read by eye shows twelve figures none of which is grounded by a contact shadow; and this leg crop shows correct foot geometry with no shadow beneath it. ⇒GR46 rk_shadow_cascade outranks every geometry rung on this board, including the hair clamp this session spent itself on and including the foot-placement work the locomotion row points at. ⚠AND IT PREDICTS A CHEAP FALSIFICATION: if shadows land and the feet still read as floating, this diagnosis is wrong and the locomotion row was right -- so the next capture of this same crop box against these same numbers settles it either way, with no eye involved. |
| 2026-09-04 | GR1 | land | PER-SUBJECT ACCEPTANCE A B IS OPERATIONAL AND IT PRODUCED ITS FIRST FAILING VERDICT, WHICH AGREES WITH THE OPERATOR. Five subjects graded from FIVE DISTINCT SUBJECT CROPS of one browser capture of the SERVED textured page: garments 143 BAD from the swimsuit region, skin 201 GOOD from shoulders and upper torso, face 214 GOOD from the pre-existing head crop, hair 226 GOOD from the head and hair region, locomotion 230 GOOD from legs and feet. AN 87-POINT SPREAD ACROSS SUBJECTS THAT PREVIOUSLY ALL READ 337 IDENTICALLY, and each row carries its own multivariate lens vector so the method is multivariate per subject rather than one composite. ★★★★★★THE RESULT THAT MATTERS: garments graded from its OWN region returns 143 against floor 151, composite=BAD -- THE FIRST FAILING REFEREE VERDICT ON THIS BOARD -- and the accept- plane carries garments as REJECT. THE REFEREE AND THE OPERATOR NOW AGREE ON THAT SUBJECT. Three framings of one subject settle the mechanism: whole frame 337 GOOD, figure crop 229 GOOD, garment-only crop 143 BAD. ⇒THE JUDGE WAS NOT BROKEN FOR THIS SUBJECT; IT WAS BEING HANDED THE WRONG PICTURE, and every earlier GOOD was a measurement of a scene that happened to contain the subject. This is the direct remedy for the frame-versus-subject defect logged earlier today and it needed NO BUILD -- only a crop. ⚠WHAT THIS DOES AND DOES NOT SETTLE, stated so nobody over-reads it: it settles that the referee CAN discriminate subjects and CAN disagree with a flattering answer, which is what makes it usable as an A B instrument at all. It does NOT settle the percept ladder, which still reads 12/13 REFUSED against the operator ordering, so a GOOD from this panel remains bounded by that; it does NOT settle provenance, since actor on every referee row here is a seat and the accept- plane's actor=operator is still self-declared; and the four FRAME-graded rows from earlier today remain in the plane at 337 to 339, superseded and inflated by roughly 110 points of framing, kept as honest captures measured the wrong way rather than deleted. ★THE REPRODUCIBLE METHOD, four steps, no build, any seat: ac_cdp.py evaluates canvas toDataURL on the served page and the base64 is decoded locally; PIL crops the subject region and upscales it; nx_content_put_client.py uploads to knowledge/fetched/ under contentput.cap with a content-CAS receipt; nx_accept referee <subject> <NAS-absolute> textured <seat> grades it. Every hop verifies by hash, and the referee- plane is the AUTHORITY -- nx_accept list lagged it by two rows during this very sequence and reported a superseded score. |
| 2026-09-04 | GR49 | measure | THE SOVEREIGN THIRD WITNESS IS COMPOSITION, NOT CONSTRUCTION -- THE WEBSOCKET STACK ALREADY SHIPS IN NISHILANG. OPERATOR 2026-09-04 asked, of the Python browser driver this seat used to read the served page: is it a tool we need to duplicate into nishi lang and use as a 3rd witness. MEASURED ANSWER, by listing buildroot/runtime rather than by estimating: nx_websocket.nx, nx_websocket_client_upgrade.nx WITH nx_websocket_client_upgrade_test.nx, nx_websocket_frame.nx WITH nx_websocket_frame_test.nx, plus nx_json.nx, nx_json_emit.nx, nx_base64.nx, nx_net.nx and nx_hmac_sha1.nx. THE RFC 6455 HANDSHAKE AND THE FRAME CODEC -- which ac_cdp.py's own header calls the roughly forty lines below and treats as the hard part -- ARE ALREADY WRITTEN AND ALREADY TESTED HERE. CDP to a local browser is plain ws with no TLS, so nothing else is needed underneath. ⇒WHAT IS ACTUALLY MISSING IS ONLY THE CDP GLUE: open the DevTools endpoint, send Runtime.evaluate, read the result frame. That is a small organ composed from tested parts, not a project, and it is buildable on the LAPTOP toolchain where a brand-new file has no laptop-versus-NAS fork to converge first -- which matters today because NAS build admission has refused every lane for hours. ★CHECK-BEFORE-BUILD PAID OFF EXACTLY AS THE LAW PREDICTS: I was one decision from writing a WebSocket client from scratch, in a tree that already ships one with tests. THE WITNESS STRUCTURE THIS ENABLES, and each rung is independent of the others by construction: ONE the third-party browser over CDP, which stays outside and exists to falsify us; TWO a NishiLang CDP client -- OUR wire, THEIR engine -- which removes Python from the EVIDENCE path even though it was never in the BUILD path; THREE nx_browser rendering the same page natively -- OUR wire, OUR engine -- which is the zero-translation rung of the operator's own ladder for NishiOS and Nishi browser. ⚠WHY THIS MATTERS BEYOND TIDINESS, measured today: the estate's own searcher is structurally blind to the lines it ships, nx_shelltool skips any line longer than its buffer and every emitted world page is a handful of 200 KB lines, so THE ONLY INSTRUMENT THAT COULD READ OUR OWN SERVED PAGE TODAY WAS A THIRD-PARTY BROWSER DRIVEN BY PYTHON. A sovereign estate whose PROOF depends on someone else's engine has outsourced its witness, and that is the half nobody was counting while congratulating the build path on being clean. |
| 2026-09-04 | GR35 | measure | RIGVIEW AND BEACH SHOW DIFFERENT MODELS BECAUSE THEY LOAD DIFFERENT ASSETS, AND THE RICHEST ASSET IS SERVED BY NEITHER. OPERATOR 2026-09-04: the rigview still shows a totally different better in someways model where the beach has a better model in other ways. MEASURED, from the SERVED pages and from source, not inferred. RIGVIEW loads /world/ref9d.nxa -- read out of the live page through the browser wire. BEACH walks a hardcoded fallback chain in nx_game_page_emit.nx that prefers ref9e.nxa, with ref9d.nxa and ref9.nxa behind it, and nx_craft_emit.nx names ref9d.nxa separately. ★★★★★★SO THE TWO PAGES RENDER TWO DIFFERENT CHARACTER ASSETS AND NOTHING RECONCILES WHICH PAGE GETS WHICH: nx_nxa_rig_emit.nx names NO asset at all in its source -- rigview's NXA arrives as an emit-time URL argument -- while beach hardcodes its chain, so the selection lives in two incompatible places and neither can see the other. THE OPERATOR'S OBSERVATION IS THEREFORE NOT A RENDERING-QUALITY QUESTION AT ALL: each page is better in different ways because each is showing a DIFFERENT MODEL, and a comparison between them measures asset choice rather than renderer quality. ⚠AND GR35 ALREADY RECORDED THE WORST PART, UNBUILT: ref9f.nxa is ref9e PLUS TEXR:3590072 of micro-relief and is referenced by NOTHING, so the RICHEST finished asset is served by neither page while roughly 80 MB of finished character work sits orphaned. THREE TIERS EXIST, THE BEST ONE IS UNUSED, AND THE SELECTION IS A HARDCODED FALLBACK RATHER THAN A DECLARED LADDER -- which is exactly what GR35 gpe_asset_ladder was costed at 0.5 to fix and has been sitting unbuilt. ★SECOND MEASUREMENT FROM THE SAME READ, and it settles which of today's two fixes actually shipped: rigview's probe returns ok=1 nv=14164 nt=28092 nj=104 hair=2600 clip=202 MISS=83. miss is the blind-set counter added to nx_nxa_rig_emit TODAY -- 83 cloth vertices with NO body sample inside the search radius -- and it is LIVE on the served page. So the rigview soft-tissue proxy fix and its new instrument DID ship while the beach clamp did not, which is consistent rather than surprising: different emitters, different binaries, and only nx_craft_emit was stale by 16.6 hours. ⇒ORDERING CONSEQUENCE: comparing beach against rigview to judge renderer progress is measuring the wrong variable until the asset ladder lands, and any A B that changes the served asset without recording WHICH asset was served is uninterpretable. |
| 2026-09-04 | GR46 | measure | THE SHADOWS ARE NOT MISSING, THEY ARE VERTICAL -- THE BEACH RENDERS AT LOCAL NOON AND CLIMBS TO EXACTLY OVERHEAD. This corrects a diagnosis I published twice today and it changes the remedy completely. WHAT EXISTS, read from source: nx_game_page_emit carries a full cast-shadow pass at uShd=1 described in its own comment as Blinn 1988 planar projection, the published no-shadow-map answer, plus sun shadows at uSun=1 via a voxel DDA march from the hit point toward the sun, plus contact ambient occlusion over the last .008 H, with the shadow depth DERIVED from sun height and the sky-fill chroma rather than tuned. It is not missing and it is not crude. WHAT GATES IT: uShd and window.__sund and sund[1] greater than NXSHD_MINY and no npcshd=0 in the query. MEASURED LIVE on the served page: NXSHD_MINY=.1 and sund[1]=0.9955, so the gate PASSES BY A FACTOR OF TEN and the pass runs on every frame. THE REASON NOTHING IS VISIBLE: sund[1] is the SINE OF SUN ELEVATION, so 0.9955 is 84.55 degrees -- almost straight down -- and a vertical sun puts every cast shadow DIRECTLY BENEATH ITS OWN CASTER where the caster hides it. Twelve figures stand on sand with nothing under them while the shadow pass works perfectly. THREE SAMPLES FROM ONE PAGE SESSION PROVE IT IS ALSO GETTING WORSE: 83.37 degrees at 20 s page age, 84.55 at 25 s, 86.33 at 75 s -- roughly 3.2 degrees per minute, CLIMBING TOWARD 90. So a visitor's first minutes, and every capture anyone has ever taken including this seat's and the referee's, land in the WORST AVAILABLE LIGHTING. THE SOURCE: sel = ex.sunel() over 4096 from the wasm engine, and THE FALLBACK IS sel = 1 WHICH IS EXACTLY VERTICAL -- an absent export would pin the sun overhead forever, so the default failure mode is the worst possible frame. There is NO time-of-day lever in the query string at all: greps for tod, time, sun, hour, noon and dusk return nothing. THE CORRECTION TO MY OWN NUMBER: nx_frame_score returned grad=35 on this world and I published that a near-flat gradient field IS the absence of shadow. It is not. It is the absence of RAKING LIGHT. Same measurement, opposite remedy -- one calls for a shadow-mapping project, the other for a starting angle. ONE VARIABLE EXPLAINS THE WHOLE OPERATOR REJECT: no visible cast shadows because they hide under their casters; flat plastic skin because vertical light strikes every body surface at a similar angle so there is no form falloff; and voxel sand reading as one repeating tile because no raking light means no dune relief. CONSEQUENCE FOR THIS RUNG: GR46 cascaded shadow maps remain worth having but they are NOT the highest-leverage change. Starting the day at 25 to 35 degrees instead of 90 costs no renderer work, uses the pass that already ships, and is bounded below only by NXSHD_MINY=.1 which is 5.7 degrees. THE A B IS RUNNING AND NEEDS NO CODE: the cycle advances on its own, so the SAME page at roughly 15 minutes of age gives the same scene under raking light -- one variable, same assets, same renderer. If that frame shows grounded figures and modelled bodies the diagnosis is proven; if it does not, this row is wrong and the shadow-map project is back at the top. |
| 2026-09-04 | GR53 | land | nx_cdp_eval LAPTOP-ONLY 2026-09-04: sovereign CDP witness (buildroot/runtime/nx_cdp_eval.nx 6798 B; built _build/nx_cdp_eval.sov.elf 40253 B sha bfe1756b), usage nx_cdp_eval <port> <ws-path> <expression>; imports nx_syscalls, nx_websocket_client_upgrade, nx_websocket_frame; hand-built sockaddr_in to loopback; the fixed mask key CE_MASK is named as a loopback-only choice. PROVEN: three refusal paths plus a full masked round trip against a WebSocket fixture (FIXTURE-GOT masked=1 len=107, reply printed verbatim, exit 0). NOT promoted, NOT on the NAS, not yet driving a real browser tab -- that is the remaining leg, and it retires the Python ac_cdp.py wire. MEASURED on the way: nx_net.nx does NOT compile standalone on the laptop toolchain (8 errors, Result of i64 and NetError unresolved while NetError is an enum at line 70), so it was dropped from the closure rather than worked around -- a lang-domain defect to adjudicate. |
| 2026-09-05 | GR41 | measure | rigview inventory against the DanceXR viewer, 2026-09-05 (charsim R26 to R36 carry the rungs): rigview serves one NXA (/world/ref9d.nxa, 104 joints, 14164 verts, 4 influences) with ribbon hair, a chest-pinned dress, body colliders, flesh springs and a Natural or Clothed toggle; DanceXR loads PMX XPS FBX by drag and drop, plays VMD and BVH with on-the-fly bone fixes, stages many actors, and runs alive, physics-tool, camera and audio layers -- GR41's sovereign-tier viewer is the page those rungs land on and GR25 garment penetration stays the ruler for its cloth |
| 2026-09-07 | GR39 | measure | Rigview release rigview-20260907-asset-ui is SERVED, VISUAL ACCEPTANCE PENDING. Added visible asset selection, dated release summary and worklog at /world/rigview; ref9e selects the beach asset. Strict selector18/18 and whole script syntax passed. Origin body5773 contains controls and release feedback. Browser discovery empty, so model rendering and interaction not visually accepted; no model appearance or physics improvement claimed. GR39 completion status is unchanged. |
| 2026-09-08 | GR33 | land | Beach hair pause-recovery hotfix h1788833756 is served with a visible release note. A ten-second pause previously left6.933seconds of backlog after60frames; the fixed timing discards excess paused time under the existing four-step budget. Tested normal outputs30/60/120FPS remain identical; pause fixtures1/10/60seconds resume without backlog. Actual native private emission4507 contains the tested function and both scripts parse. Composite nx_craft_emit755840B SHA77e47e930ac2d0b67fb0666ffddd4c3e1c698876806c6e6215a062a747b68964 promoted4585 and restaged4593; exact deployed-copy comparison4595. The one replaced printable timing statement was reviewed; contentdiff4421 remains RED for the intentional replacement. Full served-origin Beach4143 verified. Browser visual acceptance, hair appearance and body collision remain incomplete; GR33 overall completion is unchanged. Reference: https://gafferongames.com/post/fix_your_timestep/ for bounded timestep recovery, not a comparative performance result. |
| 2026-09-08 | GR33 | land | Beach groom-axis hotfix h1788835310 served with visible release worklog (HTTP5942 fullread; stage/live hash5927 identical). HSTR geometry was Z-up but ribbon width/jitter construction assumed Y-up; convert at builder boundaries. Actual ref9e asset SHA6e3078707944141164c50860939057173adf3787a48c6ef936e14ce2322319da has4800guides; actual emitted CPU builder preserves604800vertices and allUVs, finite/nondegenerate triangles and repeatability. Vertical width edges fall from100795quads to0 under the horizontal-width oracle; this is a coordinate correctness result, not visual quality or body-contact proof. Native source5310/build5374/private emit5418 verified; composite5493/promote5828/restage5871/hash5876 preserves regeneration. Complete diff5576 reviewed13lost printable runs (two statements and eleven comments), with5gained; RED remains for the intentional replacement. Browser/GPU/appearance/contact and external comparative performance remain unverified; GR33 remains incomplete. |
| 2026-09-08 | GR33 | measure | Actual Beach ref9e asset re-hashed6363:11728120B SHA6e3078707944141164c50860939057173adf3787a48c6ef936e14ce2322319da. Reproducible local contact-audit.mjs checks raw bind-pose surface intersections:11663of33600guide segments strictly cross body triangles; corrected emitted ribbons34153of100800quads have at least one strictly crossing edge (prior34508). Body has220boundary edges in6loops, so no unqualified solid inside/outside oracle. BVH agrees with brute force on131distributed guide samples; analytic hit/miss/winding/endpoint/coplanar fixtures pass. Counts omit touching, coplanar and exact-boundary cases and do not measure animated shader-deformed or rasterized clipping. This is a contact baseline, no acceptance or new visual release. Next fix must address guide/body fit and rendered children against posed body geometry. Established technique reference: AMD TressFX4 dynamic skinned-mesh SDF collision https://gpuopen.com/news/tressfx-4-simulation-changes/ (2017 primary source; not September2026 SOTA or a measured Nishi comparison). GR33 remains incomplete. |
| 2026-09-08 | GR33 | land | Beach h1788836868 initialized cheek vertex input served: publish1788838826 exit0 746442B; full HTTP1788838842 verified against scoped stage except existing ad insertion. Visible dated release worklog present. Composite adoption8732/restage8798/hash8843 exact. Float64 native scalar checks cover14164 body vertices, not GPU appearance or contact acceptance. User reports scene still looks unchanged: next milestone substantial hair/body-contact before-after, GR33 remains incomplete. Preview capability revoked8682 confirmed reason5. |
| 2026-09-08 | GR33 | measure | Beach release marker must be native in-game GPU geometry at starting camera with screenshot, not HTML icon. Native bush source and T96 under test, not delivered. User requires whole-system evaluation: CPU/GPU/RAM/VRAM/transfers/storage/network/frame pacing/thermal-power/cleanup; unavailable counters are coverage gaps, never passes. CPU-only Node simulation result does not evaluate renderer. Latest gate build refused admission; no queued work. Source verification and paced retry in progress. |
| 2026-09-08 | GR33 | measure | Native Beach release bush gate1788840419 passes96/96: fixture delta14, overwrites0, restamp unchanged, initial-camera ray hits foliage, native GPU palette verified. Native plain/sharedWASM and private page emitted. Exact-artifact VM suite1788840781 TERMINAL timeout120s after parsing/memory/render and initial terrain checks: partial only, NOT passed. No public release or GPU screenshot. Resolve fixture-failure init handling, shared palette scope, VM validation and browser verification before promotion. Full-system performance coverage remains outstanding. |
| 2026-09-08 | GR33 | land | Beach release beach-2026090702 served: publish1788842265 749670B; full HTTP1788842290 exact expected engine/page plus existing edge ad. Native world96/96; final plain/shared WASM19/19. User reports voxel teal bush visible in Waterfox. Automated screenshot and GPU/FPS improvement unobserved; no photoreal or SOTA claim. Worklog below game. Shared foliage recoloring disclosed. Delivery and rollback automation remain open; request-size workaround required17 exact artifact edits. |
| 2026-09-08 | GR11 | requirements | September 2026 Beach production requirements B26-RENDER/STREAM/SAND/CHARACTER/HAIR/HOT/DELIVERY/REVIEW captured in knowledge/compare/sovereign-human-world-contract-20260907.md (47776 bytes after append1788872418). Primary-source corrections separate compute capability from guaranteed speed, texture arrays from sparse residency, writeBuffer from zero-copy, device loss from one-frame rollback, and particle-push shaders from MPM/XPBD. Each requirement needs source and served identities, dependency/owner, comparable hardware/workload, quality and correctness oracles, costs and next experiment. Existing graphics/character/softbody/deploy/toolchain domains own the work; no new random pages. This is requirements capture, not frontier parity or delivered Beach improvements; comprehensive current benchmark research remains open. |
| 2026-09-08 | GR33 | requirements | B26-HAIR/CHARACTER inside-out state: persistent seeded follicle layout, growth direction and length, shaving/stubble/regrowth, region-specific irritation and hair-removal history; shared skin and eye state continuity. Full qualification fixtures and approximation limits in knowledge/compare/sovereign-human-world-contract-20260907.md, subsection persistent follicle, skin and eye state. Pair close/mid/far visual review with reproducibility, save/resume, temporal stability and measured resource costs. Clinical treatment accuracy requires primary evidence; no complete digital twin or Beach implementation is claimed. Reuse GR29 skin and GR33 hair rather than add isolated pages. |
| 2026-09-08 | GR33 | research | Character research references checked 2026-09-08: Meta Codec Avatars https://www.meta.com/emerging-tech/codec-avatars/ provides telepresence datasets and reference implementations; inspect capture requirements, identity generalization and clothing/hair limits before comparison with procedural Beach NPCs. BioWare/Frostbite https://developer.microsoft.com/en-us/games/articles/2024/12/bioware-dragon-age-the-veilguard-strand-hair-technology/ (2024-12-05) reports up to 50000 strands per character on Xbox Series X, compute software rasterization, collision/pinching and extreme-closeup challenges; this is a production reference, not a WebGPU benchmark. EA SEED FAST https://media.contentapi.ea.com/content/dam/ea/seed/presentations/seed-id3-2024-spatio-temporal-sampling-paper.pdf (May2024 DOI10.1145/3651283) optimizes sample distributions for reconstruction filters, evaluated on AO, volumetric ray marching and dithering; qualify with matched filters, error and temporal stability at equal cost. UMBC https://news.cs.umbc.edu/2020/05/umbc-researchers-develop-better-techniques-to-render-characters-with-realistic-skin/ identifies Xie/Olano/Karis/Narkowicz2020 DOI10.1145/3384536, variance-guided adaptive subsurface sampling; underlying paper and current successors need review. Map skin to GR29 and rendering/sampling to existing graphics owners. These four leads are neither a comprehensive September2026 survey nor Nishi parity. Investigate Stanford, USC ICT, Berkeley, Cornell, CMU, NVIDIA, Epic, Disney, Adobe and Meta by specific dated method, reproducible artifact and relevant workload, not institutional prestige. Require source/availability/rights scope, hardware, quality-cost evidence, failure cases and one executable next comparison per adopted lead. |
| 2026-09-08 | GR11 | requirements | Operator target: integrated Beach WebGPU above60FPS with intelligent resource use, assessed on declared device/browser/resolution/NPC-count/scenario and quality tier. Budget for60FPS is16.67ms per frame; report raw intervals and p50/p95/p99/worst, missed-budget frames, CPU simulation/submission and actual GPU intervals when exposed, unavailable counters, memory/residency, network and energy where measurable. Average FPS alone cannot qualify smoothness. Persistent Nishi research must discover dated primary publications, implementations and shipped references; maintain capability coverage, query/source history, access/publication dates, supersession edges, unresolved gaps and evidence freshness. Re-evaluate affected comparisons with pinned reproducible workloads when relevant methods change; do not rerun everything or infer parity from new links. As-of-today ranking requires current evidence and an explicit survey boundary. Native research automation and comprehensive coverage remain unverified; keep deployment and Beach delivery as current implementation priority. |
| 2026-09-08 | GR11 | research | WebGPU4K60/Wasm source audit2026-09-08: https://emscripten.org/docs/porting/multimedia_and_graphics/WebGPU-support.html describes Emdawnwebgpu C bindings on top of browser WebGPU, not direct driver access. https://toji.dev/webgpu-best-practices/buffer-uploads.html recommends writeBuffer for Wasm; mapped uploads can require an extra heap copy, staging rings unmap/copy/remap rather than remain concurrently GPU-accessible. Do not claim zero-copy or guaranteed native speed from webgpu.h or prescribe replacing Nishi compiler. https://ir.pku.edu.cn/handle/20.500.11897/738236 dates FusionRender to2024; https://websplatter.github.io/ and https://arxiv.org/abs/2602.03207 identify a distinct2026 Gaussian-splatting pipeline; original pasted WebSplatter link incorrectly pointed to FusionRender. https://arxiv.org/abs/2604.02344 studies LLM dispatch overhead, not proof that4K Beach rendering is rarely GPU-bound. Benchmark native4K3840x2160 separately from upscaled output; retain internal resolution, quality and raw frame intervals. Candidate experiments: batch dirty uploads, bundles for compatible repeated commands, GPU resident simulation/culling with indirect draws, pipeline preparation and bounded staging reuse. Select by measured bottleneck and total quality/cost, not mandatory-method slogans. Full paper reproduction and current broad coverage remain open; this audit does not assert a delivered renderer change. |
| 2026-09-08 | GR11 | priority | Operator sequencing clarification2026-09-08: Beach big rocks before microdetail. Integrated priority: reliable scoped delivery and measurement; ground/collision stability with no unintended sinkholes or falling through terrain; reproducible procedural beach variants; coherent shoreline/wave/wetness interaction; grounded character movement and broad lighting/material quality; then finer hair/follicle/skin detail. Reuse existing terrain/gameengine/procgen/water owners and dependency graph; rank by user-visible severity, unlocked dependencies, recurrence and measured total cost per accepted result. Terrain qualification must compare rendered surface and collision surface across seeds, terrain seams, slope transitions, spawn placement, walking/running, timestep changes and stream/unload/reload; distinguish intentional depressions from defects. Procedural qualification pins seed/config/version and tests materially different beach layouts with coherent terrain, water level and contact, using Houdini and other verified procedural systems as workflow/capability references rather than name-only scores. Current runtime defect cause and external parity remain unverified. Fine-detail requirements remain retained but must not displace these blocking integrated outcomes. |
| 2026-09-08 | GR11 | benchmark-target | Operator selected False Earth https://github.com/momentchan/false-earth as an external Beach direction, implemented through NishiLang emission. README inspected2026-09-08 describes GPU procedural terrain/grass, shared terrain sampling for blade attachment, stable PCG seeds/grid snapping, distance LOD, character grass push, lighting/postprocessing and adaptive DPR. Map these to existing terrain/procgen/vegetation/renderer owners. Compare matched traversal and camera views for terrain continuity, vegetation anchoring and motion, interaction, shadows, temporal stability and scene coherence. Record actual internal render resolution separately from output resolution because adaptive DPR can hide reduced pixel workload; pin commit, configuration, browser, GPU, seed and capture before timing comparisons. README claims are not reproduced benchmark results. Study mechanisms and build native Nishi equivalents; do not transplant its React/Three/TSL stack or separately licensed astronaut asset. Reference deliberately depicts transforming surreal terrain, so its ground behavior is not an oracle for stable physical Beach ground. Current first defect remains continuous rendered terrain versus voxel player-contact height, reproduced locally with a256Q8 discrepancy; no served correction claimed. |
| 2026-09-08 | GR11 | research | Water foundation read2026-09-08: Mark Finch, GPU Gems chapter1 https://developer.nvidia.com/gpugems/gpugems/part-i-natural-effects/chapter-1-effective-water-simulation-physical-models . GPU geometric waves plus fine normal detail; wavelength/amplitude/direction/speed constraints, analytical derivatives and bounded Gerstner steepness are reusable baseline concepts. Historical shipped Uru technique, not September2026 frontier proof or complete shoreline fluid simulation. Beach acceptance should share parameterized water state between visual displacement and contact/buoyancy queries, verify normal derivatives and nonfolding bounds, preserve temporal continuity during parameter changes, and separately qualify shoaling/breaking/runup/foam and wet sand. Implement via native Nishi emission; compare newer methods by workload, quality and cost rather than adding wave count alone. |
| 2026-09-08 | GR11 | benchmark-target | Operator selected https://www.shadertoy.com/view/Ms2SD1 as an ocean visual reference. Direct fetch failed2026-09-08, so current original source/render was not inspected. Discoverable copies identify Seascape/TDM and carry restrictive attribution/noncommercial/sharealike headers; original identity/version/terms require primary verification before any code reuse. Treat as visual target only pending capture: broad swell, fine ripples, reflective horizon, crest shading and temporal coherence at measured resolution/cost. Nishi implementation remains native; separate visual reference acceptance from interactive shoreline, buoyancy and breaking-wave physics. No copied shader or verified FPS claim. |
| 2026-09-08 | GR11 | reference-pending | Operator reference https://www.shadertoy.com/view/wt3XDj requested2026-09-08. Direct fetch returned402; title, author, contents, appearance, terms and performance remain unverified. Retain exact shader ID for future authorized access; do not invent classification or benchmark scores from an inaccessible URL. Evaluate relevance to current Beach blockers before implementation. |
| 2026-09-13 | GR29 | measure | 2026-09-13 native regional skin asset generated and decoded: resident-region-skin2048-t278.nxa SHA256 1b0ece72f0793618a774ecb2f24dd425393ac89098312b397334abbe7f93dbbf, 41169952 bytes, four regional sets and20 PNGs. Actual native bake3502 exit0 elapsed70650ms VmHWM231244kB; initial3414 timed out at60000ms after three sets and was killed/reaped, no final output, source unchanged. Existing positive inspect3617 and native decoder3616 passed all20 dimensions/RGB byte lengths; all18 source sections byte-equal. Canonical codec and baker safety adopted; native CLI shipping under review. Renderer integration, physical calibration, chart seams and visible photorealism remain unverified. Evidence buildroot/knowledge/gates/region-skin-generated-review-t278.json and texr-renderer-handoff-t278.json. Isolated jobs own temporary allocations until exit; no in-process leak-free claim. |
| 2026-09-13 | GR29 | land | 2026-09-13 canonical native regional baker shipped via guarded workflow3965; exactly5 reviewed diagnostic changes allowed, fresh capacity gate10/10, serving and _offc SHA256 e19ad62435b32c75dc8fd40b3d0ce2df0fd15952842ce4adc6c4844e708c42c9 byte-identical. Actual existing registered inspect call3985 reads2048 candidate and20map descriptors. See buildroot/knowledge/gates/region-baker-canonical-adoption-t278.json. Generated map artifact and decoding are qualified; shared-renderer integration and visible photorealism remain open. No Beach page or production asset replaced. |
| 2026-09-13 | GR29 | land | 2026-09-13 Native regional skin chart-edge normal correction adopted and shipped via existing baker; canonical six-group gate passed, registered inspect passed, live/offc SHA db135be99a0a127731a08d27ce5874431ec0b2a33b6305721140e8a3f71a69db. Controlled 2048 bake preserves 16 non-normal maps and 18 sections; 16642856 nonedge pixels unchanged, 50253 boundary pixels corrected, zero measured new gradient-direction disagreements. Physical calibration, chart topology and Beach renderer integration remain open. Evidence buildroot/knowledge/gates/region-edge-adoption-t278.json |
| 2026-09-13 | GR29 | land | 2026-09-13 Existing native regional inspector now emits artifact SHA256 and explicit normal transfer metadata tied to focused shared encoding owner. Registered call6185 equals qualified4022B fixture; eight normal descriptors, prior fields preserved, producer verification remains a separate hash-joined receipt. Native complete-capture regression and16 boundary checks passed; ship6153 rebuilt stale shared-closure gate before promotion. No Beach shader/page release from this change. Evidence buildroot/knowledge/gates/texr-descriptor-delivery-t279.json |
| 2026-09-13 | GR29 | watch | 2026-09-13 Independent asset-cohort discovery retained: existing rigfloor roster has11 ref variants of one geometry; Catwalk glTF has no declared extensions/morphs. Additional available VRM-derived Seed-san, FBX-derived DarkKnight and static OBJ Bunny are regression candidates with exact hashes; no genuinely unseen holdout verified. Native section checks pass for two NXA assets, while provenance API returns NO-ROW and private_use allowed. Existing asset-catalog backing manifest not found at declared roots and4096 enumeration/index hazard remains; no catalogue mutation or rendering claims. Review buildroot/knowledge/gates/asset-cohort-existing-owner-review-t280.json |
| 2026-09-13 | GR29 | watch | 2026-09-13 Native all-corner UV preparation preserves all84276 source corners and28092 triangles;102-edge shortest source-graph cut leaves7 charts with disk-consistent Euler/boundary counts. Existing clipped-cotangent Jacobi diverges on main chart; new isolated factorized least-squares solver converges6 small charts in31-132iterations, main chart remains unqualified after1024iterations20.99s. No triangles hidden and no usable atlas or Beach texture release claimed. Development fixtures and familiar NPCs are not unseen holdouts. Evidence knowledge/gates/uv-solver-cut-review-t280.json; actual job1789328087; pending bounded followup8229. |
| 2026-09-13 | GR29 | watch | 2026-09-13 UV native qualification exposed compiler dot-store hardcoded integer destination despite resolved f64 field. Isolated existing-parser candidate passes13/13 native exact-value and invalid-field controls including indexed chains; fullreview knowledge/gates/parser-field-final-review-t280.json. Canonical/serving compiler unchanged pending review. Separate retained debt: address-of SSA let returns value as address; MCP64KiB request ceiling. Solver caller conversion misuse corrected using existing explicit numeric intrinsics, not attributed to compiler. No visual acceptance from source gates. |
| 2026-09-13 | GR29 | watch | 2026-09-13 Independent nonhumanoid Bunny qualification: exact source1eb35d1e21ce99e5ce911353b6be278990713448dd9e8f5c9387f9de39b32205 2408417B; native oi_read retains35947vertices/69451triangles but fx1024 creates2011 zero-area faces. Independent source/quantized buffers agree; source already has1113 loosevertices and223 boundaryedges, no winding/nonmanifold/duplicate defects. Existing meshcheck22/22 controls pass. Generated normals not allunitlength; occlusion/rendering unmeasured. Asset NOT accepted, not unseen. Exact native buffer identities, failure provenance and coordinate-contract options retained knowledge/gates/bunny-native-review-t282.json. No asset publication or shared parser change. |
| 2026-09-13 | GR29 | land | 2026-09-13 Shared normal encoding contract and explicit selected-producer qualification published to Alpha Beach. Native fullpage30/30, legacyCLI30/30, wrong reference SHA refused before page emission. Public https://nishifamily.com/alpha/beach exact1417792B SHA8e6fddbede735b7107dd2f622cdff2d79db1111eb106ce9bbadd6393a93048e7; Playwright WebGPU Intel xe-lpg cast pipeline created, clean scope, four legacy visitors, Frame character interaction reviewed. WebGL isolated samecode qualifies rendering. Diagnostic release note under Controls. Skin/hair/eyes/garments remain visibly inadequate; no TEXR integration or photoreal acceptance. Beta/Production unchanged. Evidence knowledge/gates/normal-alpha-release-t297.json. Correct structured release row replaces interpretation of prior free-text t296 append; historical text retained. |
| 2026-09-13 | GR29 | measure | 2026-09-13 UV preparation exposed resolved f64 field stores incorrectly checked/emitted as i64. Reviewed exact function adopted into canonical nx_parse.nx SHA8fb41956e2782826f26c5a445274fc7a2b4293f77f6da11d8d8167eaad2f6915; staged canonical x86 compiler976040B SHA5286c47d09a40271b143c356c66a8502af3e9147b19716de7e8e0261f2b5efd7 passes13/13 real native typed-value and direct/indexed invalid-field checks. Existing gate now reuses only byte-exact fixtures; source history not overwritten. Serving compiler unchanged pending wider toolchain qualification. No usable UV atlas or skin visual gain claimed. Evidence knowledge/gates/parser-canonical-adoption-t298.json. |
| 2026-09-13 | GR29 | measure | 2026-09-13 Native staged compiler field gate16/16: added u8/i32 neighbor preservation, signed negative load and pointer-chain checks; existing float64 and invalid-field controls retained. Evidence knowledge/gates/parser-field-narrow-t299.json job1789329333. Serving compiler unchanged; f32 and concurrent gate isolation unqualified. No Beach visual acceptance implied. |
| 2026-09-13 | GR29 | measure | 2026-09-13 Existing compiler field gate19/19, now f32 direct/indexed/pointer-chain values and neighbor preservation verified through staged canonical compiler. Evidence knowledge/gates/parser-field-narrow-t299.json retains both16/16 and19/19 iterations. Serving toolchain and public Beach unchanged; full conformance and concurrent harness isolation not claimed. |
| 2026-09-13 | GR29 | watch | 2026-09-13 Source-pinned material handoff audit: prepared TANGENT discarded; GPU tangent/normal image absent; imported lit shader uses geometric vN. Changing normal flag alone cannot repair this. Khronos tangent handedness requirements linked in knowledge/gates/material-handoff-audit-t301.json; shared owner integration and both-backend qualification pending. No new visible release. |
| 2026-09-13 | GR29 | measure | 2026-09-13 Structured binary comparison adopted into existing shared library: native16/16 and existing display18/18 pass. Correctly distinguishes DWARF count-byte false loss from genuine removals. Evidence knowledge/gates/contentdiff-shared-adoption-t302.json. Serving CLI/ship/staging migration pending; no allowance or new Beach release. |
| 2026-09-13 | GR29 | watch | 2026-09-13 Root review confirms candidate Bunny precision result0 introduced collapses vs2011 legacy;32 native controls are scoped evidence. Before adoption: reproduce/fix sign-only OBJ suffix acceptance and expose polygon triangulation qualification in result. Evidence knowledge/gates/obj-root-review-t303.json. No imported-model rendering or live repair claimed. |
| 2026-09-13 | GR29 | measure | 2026-09-13 Exact OBJ geometry API adopted in canonical nx_obj_import; actual native35/35 passes, Bunny0 introduced collapsed faces vs2011 legacy, exact source/index hashes retained. Polygon fan state explicit; units/materials/rendering unqualified. Evidence knowledge/gates/obj-canonical-adoption-t304.json. Ingestion-route and Alpha integration remain next; no public visual release. |
| 2026-09-13 | GR29 | measure | 2026-09-13 Native UV candidate conditioning counterexample9814 repaired by explicit Dirichlet elimination; same solver planar controls52/52 including cancellation/resume and existing LSCM24/24. Actual resident adaptive run0160 preserved all84276 source corners but remained rejected:10charts9392collapsed+8flipped atlas triangles, next nondisk split refused. Full15154B evidence knowledge/gates/uv-adaptive8-full-t280.out and knowledge/gates/uv-dirichlet-review-t280.json. No canonical unwrap or visible material adoption; caller cancellation is in-process only. Boundary-seeded partition candidate0337 running, not yet accepted. |
| 2026-09-13 | GR29 | measure | 2026-09-13 Management approval parsing canonical repair passes40/40 actual-handler/helper checks: exact unique yes required; duplicate approval/confirmation refuse. Shared form lookup reused, legacy lookup unchanged. Evidence knowledge/gates/mgmt-unique-approval-t305.json. Serving binary unchanged; structured-refusal propagation and guarded deployment still pending. |
| 2026-09-13 | GR29 | watch | 2026-09-13 Operator requests additional independent models beyond historic NPCs. Cross-model qualification must distinguish source defects from conversion, rig, material and rendered regressions; retain source identity, relevant capabilities and separate held-out assets. Reviewed material-normal-segment-review-t301.json: Catwalk has671 mixed tangent-handedness triangles and0 normal-map parts, so it cannot qualify mapped materials. Existing character worker assigned verified additional mapped asset plus controlled fixtures and no-map compatibility in shared integration; request remains pending, not delivered or published. |
| 2026-09-13 | GR29 | measure | 2026-09-13 Existing native UV preparation candidate now reuses only unchanged source-face/cut-corner sets; independent full5393952B packet parity passed after detecting and correcting double-quantization at atlas stage. Matrix applications10528 to4377, observed solve60.36s to37.79s; wall-time samples are not isolated benchmarks. Evidence knowledge/gates/uv-reuse-review-t280.json12614B. Resident UV still rejected6341collapsed+4flipped triangles; no material deployment. Timeout descendant gap retained, liveness0528 matched0 after valid402process snapshot. Caller residual/refinement policy now explicit; bounded32-refinement run0985 pending. |
| 2026-09-13 | GR29 | measure | 2026-09-13 Shared structured comparison iterator and complete-file CLI adopted in canonical owners via native MCP. Actual canonical build1081 compares material143320/147288B with state0 lost0 source_removed0 source_added1. Reviewed candidate17/17 and legacy18/18; canonical negative reruns pending. Evidence knowledge/gates/contentdiff-canonical-cli-t307.json. Serving binary and downstream staging/adoption guards unchanged; no Beach visual release or behavior equivalence claim. |
| 2026-09-13 | GR29 | measure | 2026-09-13 Canonical native comparison exposes typed result through existing workflow, avoiding console parsing for next adoption integration. Canonical gate1507 passes21/21 including actual material pair, tail loss, unknown/unreadable and stale-result invalidation. Evidence knowledge/gates/contentdiff-typed-workflow-t309.json. Serving and adoption gate not yet migrated; no Beach release. |
| 2026-09-13 | GR29 | measure | 2026-09-13 Canonical adoption content stage now calls typed native comparison directly, removing comparison subprocess and GREEN substring parsing. Native1720 passes6/6 actual material, self, tail loss, unknown, missing and forged text controls. Evidence knowledge/gates/adoption-native-content-t310.json. Behavior probe/capture and downstream staging integration remain; serving unchanged, no visual release. |
| 2026-09-13 | GR29 | measure | 2026-09-13 Native UV precision witnesses1756/1780 isolate source-bound chart0(7373faces) versus chart1(859faces). Exact dyadic independent raw determinants confirm no raw zeros/flips in either. Chart0 work export introduces8zeros; Q16 atlas introduces200morezeros+1flip. Control remains allpositive. Zero source-position/index mismatches. Full packet and per-triangle evidence knowledge/gates/uv-precision-review-t280.json18879B; chart0 solve3.66s, control0.22s, not hardware-normalized benchmark. Existing Q16/baker contract is a representation choice, not physical law. Adaptive32 run0985 timed out despite retained descendant output; final39chart atlas remains rejected. No further blind subdivision or live material adoption. |
| 2026-09-13 | GR29 | measure | 2026-09-13 Canonical staging now compares complete files, separates source loss and rejects UNQUALIFIED even with override; existing raw API preserved. Native2037 passes25/25; shared md_promote always classifies before rename, compatibility2077 passes40/40 (not real promotion). Evidence knowledge/gates/staging-canonical-adoption-t311.json. API/ship/behavior integration pending; serving unchanged. Root reviewed mapped diagnostic images; reversed labels require camera/source parity analysis, no global winding fix approved. |
| 2026-09-13 | GR29 | measure | 2026-09-13 Canonical management API deploy/promote now share qualified override predicate and expose content_state/source_removed/override_can_resolve. Native compatibility2271 passes40/40, not full HTTP promotion test. Evidence knowledge/gates/management-staging-propagation-t312.json. Serving unchanged; typed behavior and ship integration remain before release-chain qualification. No Beach visual release. |
| 2026-09-13 | GR29 | measure | 2026-09-13 Native symmetric-Dirichlet refinement candidate reuses source-indexed UvGramSolve after bounded inventory. Native2419 passes16/16 including derivative finite differences, cancellation and folded-input refusal. Actual chart0 128sweeps adds2.215s, raw/work faces allpositive and packeddefects209 to70; source-normalized minimum area improves2.19e-7 to0.00120. Rejected for whole-atlas use:64packedcollapses+6flips and worstanisotropy2.36 to606.6 despite lower meanenergy. Control859faces stayspositive allstages. Full source/metrics knowledge/gates/uv-stretch-review-t281.json. No canonical changes; no precision increase or further blind subdivision. |
| 644012-10-12 | GR29 | measure | Root t313: full management native build2708 passed without install; typed adoption observation candidate composes canonical content evidence and behavior candidate without parsing approval text, qualification pending. Actual WebGPU mapped/no-map images reviewed: relief visible, orientation unresolved, no NPC/page acceptance. Search503 read-after-accept recovered on same read-only query; default web scope unsuitable for internal-file discovery; root cause unresolved. Evidence knowledge/gates/release-review-followthrough-t313.json. |
| 2026-09-13 | GR29 | measure | 2026-09-13 Existing UV distortion-referee extension candidate now has native stage-specific max-anisotropy/orientation/collapse/unknown measurements and caller-owned acceptance. Actual source-bound gate3123 passes25/25: raw incumbent2.363 max versus t281606.622 is rejected under explicit incumbent-preservation policy; packed collapse still fails independently. Control859faces qualifies allstages under its measured caller policy. Nine native anisotropy measurements match independent finite-triangle Jacobian witness within8.41e-15 relative difference; collapse excluded only from undefined ratio, retained as failurecount. Full source/delta/evidence knowledge/gates/uv-quality-review-t282.json; canonical owner nx_uvunwrap_lib unchanged. No renderer/generation/serving changes; global overlap and whole-atlas acceptance remain unverified. |
| 644012-10-12 | GR29 | land | Canonical shared source adoption t315: runner35901B, probe13761B, integer4104B, adoption10562B, content inspector25941B. Native canonical behavior25/25 integer20/20 sectionlessELF10/10 content regression21/21 composed observation7/7. Typed observation always requires reviewed acceptance; legacy CLI/ship integration and serving deployment unfinished. No new Beach release. Evidence knowledge/gates/shared-adoption-t315.json. |
| 1970-08-23 | GR29 | plan | Weekly investment recommendation, not measured spend: prioritize visible character material/rig/contact and dune occlusion delivery; complete retained integration before expanding features. Shared MCP JSON+SSE repair must pass same attached file/admin calls. Swarm routing requires matched accepted outcomes including root context, delegation, review, rework, cached input, output and latency; current Claude review receipts have rootSupervisionTokens unknown, no positive ROI established. Evaluators require known-good/bad controls and independent visual witnesses: Sept19 anatomy numerical gains failed appearance and product path was retracted. Evidence knowledge/gates/native-mcp-sse-t316.json, shared-adoption-t315.json, claude-public-source-review-t253.json, claude-public-source-revision-review-t253.json; anatomy.plan and aesthetictwin.plan. Source/gate passes are not served visual delivery. |
| 1970-08-23 | GR29 | research | Current primary baseline check: MCP2026-07-28 publication https://blog.modelcontextprotocol.io/posts/2026-07-28/ specifies stateless core, header routing, cacheable discovery and task extensions. Existing JSON/SSE candidate repairs current endpoint compatibility; does not establish2026 conformance. Versioned client/server compatibility tests and migration assessment required before protocol adoption. DORA https://dora.dev/guides/dora-metrics/ defines five delivery metrics; combine with accepted Beach quality and full task effort, never use test count as product-quality proof. Current library/code/serving/visual closure gaps remain tracked; permanent repair requires shared-owner deployment and representative consumer verification. |
| 1970-08-23 | GR29 | research | Humble Unreal realtime simulation FX suite reference candidate: native fetch1789848115 HTTP200 retained840823B knowledge/fetched/rf_e70139e6.raw SHA8901e69b1f7aea387a9cdaec6d5dfb4aa5aa88fe49615fb8a14931dd1a756ffc. Browser observed USD22/100packs; Beach candidates Nature Water, Atmospheric FX, Leaves, Kick Up, Wind Gust. Evaluate three relevant effects for measured appearance/runtime comparison before conversion investment; no purchase, asset qualification, solver-source access, direct Nishi compatibility or positive ROI established. NPC motion/material/contact delivery remains primary. License and format checks pending actual redemption. Assessment knowledge/gates/humble-fx-reference-20260919.json; https://www.humblebundle.com/software/unreal-engine-realtime-simulation-fx-suite-software |
| 1970-08-23 | GR29 | plan | Served Alpha motion release b1789848521, page e673d1ea260dc1856dae419802887fd08276e4023667584e5ec15ff06af3af9c, published through existing site-release/v2 with five hash-bound dependencies. Root Chrome observed full source clip advance, pause holding time/revision, and discrete pose0 to1 collision revisions154 to155. Evidence knowledge/gates/beach-native-motion-release-20260919.json and beach-native-motion-served-review-20260919.json. This is mechanical playback qualification only; manually loaded prepared reference is not an autonomous integrated NPC or photorealism gain. |
| 1970-08-23 | GR29 | plan | Photoreal character acceptance remains FAILED: operator reports ankle/foot deformation, witch head/eye separation, inactive arms and static soldier; root resident Character1 face view confirms protruding eyes, hair intersection and inadequate skin. Audit complete resident and imported cast individually; one passing source-reference clip cannot substitute. Investment order: procedural generator source/bind/unit and attachment coherence, parameter-derived geometry/material/deformation, then representative multiview and integrated scene qualification. Imported fixtures serve oracles; target remains reproducible procedural digital humans, not per-model hand tuning. Automatic detection and full cast motion coverage remain incomplete. |
| 1970-08-23 | GR29 | research | Native discovery verified existing nx_body_proc (76228B) and nx_body_gen (77675B), plus nx_skelgen registered in authoritative serving tool_allowlist.conf169958B. Reuse and qualify these owners before new generator development. Current body recipe emits parameterized ring/feature geometry; nested fixed-radius bone/muscle shells do not establish anatomical digital-twin accuracy. Current resident FBX-derived rig path remains separate pending lineage verification. nx_procgen is a maze generator and nx_procgen_human_gate is a2D silhouette test, neither qualifies human creation. ProcFunc2026v2 revised2026-09-15 is current procedural-composition research, native fetched41712B knowledge/fetched/rf_8fcd074e.raw SHAa97b9d552e4bf41861d9d27e62c90b119b35e5ee5739aef22b8a20e378d89bef; source https://arxiv.org/abs/2604.26943v2 . Paper demonstrates room/material composition, not validated human generation. Next acceptance is repeatable generated human variants through shared rig/material/attachment pipeline and visible Alpha, not status-only metrics. |
| 1970-08-23 | GR29 | plan | Executed existing native procedural pipeline via MCP: nx_skelgen selftest1789849988 passed21/21; rest generation1789850009 wrote568B, 21joints requested1750mm actualvertex1751mm; nx_body_proc1789850015 wrote3563B canon; nx_body_gen1789850023 emitted103824tris/20parts truncated0 to knowledge/gates/procedural-human-20260919.nxmesh. Repeated jobs1789850055/1789850062 samecounts; byte determinism not yet verified. Mesh has no skinning contract established; Alpha inspection/conversion ongoing. Root visually reviewed skullskin-current-20260919_pair.png from existing skull input: severe face/skull geometry failure, not photoreal. Current fresh-skull generation must distinguish stale input from current recipe before repair. Reusable resident rig observation candidate passed21/21 isolatedChrome fixtures; actualserved phase capture pending. |
| 1970-08-23 | GR29 | plan | Historical graphics introduction superseded after served visual review; its old scores were not current acceptance evidence. Retained original: Nishi vs Unreal Engine 5 and Godot and Infinigen and Houdini -- the whole graphics estate on one measured surface: sovereign renderer, procedural bodies and creatures, layered anatomy, motion, gaussian splats, five live wasm worlds, referees and the research library. Measured 2026-08-17: capability breadth 827 permil (nx_graphics_board, 27 grounded axes) against a perceptual headline of about 175 permil clay (headline = MIN of statistical and perceptual, twice Goodhart-proofed); figure-crop CHARJUDGE ladder 0 to 257 vs corpus mean 138 over 169 renders; bodybench honest headline 369-389 detail-bound. The distance between breadth and the perceptual number is the whole roadmap, and the leap rung is the genome-driven splat avatar. |
| 1970-08-23 | GR29 | plan | Published and root-reviewed Alpha b1789851741 SHA49d75368588da4db775f5bf6c9fc3262c54b4a860476342c1585a66d863eed3b: Progress shows actual native procedural body images and static GLB; geometry remains coarse, unrigged and visually rejected. Native repeat mesh comparison proved byte determinism SHA3b58ea74d21012da6146e787243bb4510841729bec1b0d4638f4d9bb3d90d0bb. Real constrained skull fit job1789851408 completed:5525 samples,72live parameters, objective984 to721centi-mm, onepass; not anatomical or photoreal acceptance. Verified skull emission1789851941 pending. Same-frame resident transitional capture retained in knowledge/gates/resident-rig-moving-a-20260919.json; shared hierarchy correction under qualification, no released gait claim. References and failures retained in existing review records; no Beta or Production promotion. |
| 1970-08-23 | GR29 | plan | Operator clarification: the delivery unit is a coherent playable scene, including photoreal, anime and intentionally mixed multiverse directions. Alpha component workbenches feed integrated Beta scene review and accepted Production; component tests do not establish scene acceptance. Assess shared lighting/color/style, motion/contact, environment, audio/interactions and frame-time stability together. Investment denominator is accepted scene improvement; numerator includes all model input/cache/output, delegation, review, retries, rework and elapsed time, with attribution uncertainty retained. The same evidence-driven production pattern applies to physical and digital products, including material/resource efficiency; no current positive token ROI claim. |
| 1970-08-23 | GR29 | plan | Operator correction: top-down intended scene outcomes and bottom-up primitive/model verification must meet in one traceable evidence graph. Proposed native contract: claim, intended use/style, observable, units/frame, tolerance rationale, reference/calibration uncertainty, asset/code/backend version, scenario coverage, evaluator version, counterevidence and qualified scope. Validate evaluators using seeded failures, independent reference measurements, holdout scenarios and false-pass/false-fail rates. Passing component tests cannot substitute for missing system evidence. Known failures remain visible in Alpha and Beta; they do not automatically block experimental publication. Production regression invokes scoped incident containment/rollback then repair through Alpha/Beta. This is architecture direction, not an implemented autonomous validation system. |
| 1970-08-23 | GR29 | gap | Published generated body remains visually failed: coarse white static silhouette, no readable facial/skin details and no rig. Narrow enabled/disabled centerline proxy test passed, but user reports walk-through. Native movement owner uses zero horizontal player footprint; legacy cast lacks prepared collision coverage. General solid-character claim unqualified. Investigate exact asset/default/backend/overlap and carry user counterevidence forward. Researcher search for NASA STD7009 returned unrelated top results despite available primary standard; relevance and internal-vs-web scope need explicit evaluation. |
| 1970-08-23 | GR29 | plan | Nishi studio extension contract: standards must become executable reusable capability contracts, not claims of superiority. Each primitive owns versioned inputs/outputs, units/frame, failure semantics and evidence. Existing release workflow composes preparation, changed-dependency qualification, visible experimental Alpha publication, integrated Beta evaluation, production acceptance and scoped incident recovery. Agent consumes retained evidence and requests bounded tutoring only for unknown or repeated failures. Known visual defects may publish in Alpha/Beta with explicit failed/unreviewed status; production runtime failures trigger containment, repair and regression replay rather than ordinary continuation. No claim this autonomous cycle is implemented yet. |
| 1970-08-23 | GR29 | plan | Delivery efficiency must measure operator actions separately from internal operations, root-agent handoffs, elapsed time, complete provider input/output/cache token usage, retries/rework, accepted scene improvements and escaped defects. Compare matched task scope and quality; unavailable measurements stay unknown. One action is useful only if it produces a reviewable reproducible outcome and retains failure evidence. Existing estate owners must be composed rather than adding duplicate services or pages. |
| 1970-08-23 | GR29 | gap | Workflow ownership verified from current source and native catalogue: nx_workflow is read-only board/summary/show; nx_plan_run is legacy sequence runner with fixed step/argument/output caps and differs built175712B versus serving152683B; nx_wflow is the durable CLI, built/serving117440B, backed by nx_wflow_engine. Actual reusable release definitions are knowledge/wflow/component-release-preparation.flows and .steps at serving root, not buildroot. Current two steps only set release_scope and run site_prepare prepare-release with manifest/hash; no publication step. Extend this existing owner for explicit experimental Alpha publish and uncertain-effect reconciliation after current camera served review; do not add another executor. Catalogue INVOKED NO is bounded MCP-log evidence, not proof of never used. |
| 1970-08-23 | GR29 | qualification | Native complete-person composition retains52parts using intact body/head arenas rather than dropping head features at48. Native oriented mesh job1789857279 emits44994vertices/90028triangles, GLB2161120B; source positions unchanged by Yquad orientation correction, winding disagreements25276 to0. Topology remainsRED:212degenerate triangles22nonmanifold edges; rig absent. Root source review identified INT64_MIN overflow through abs on recipe coordinates, requested direct signed-bound validation before arithmetic. Whole-surface material500/torso is an explicit uncalibrated approximation; source part/material identity is not yet carried into output faces, so eye/hair material fidelity is not established. These are experimental Alpha candidates, not photoreal qualification. |
| 1970-08-23 | GR29 | research-review | Native rank8619 refused stale SB-HUMAN. Exact nx_barfresh grant repaired by union mint8663 preserving45 existing grants and expiry; original call8686 verified46 grants. Current stored feed knowledge/fetched/rf_ff76e849.raw updated2026-09-19 includes GenStream2609.18634 submitted2026-09-16, confirmed against live arXiv and author repository. This is semantic human-media streaming with client generative reconstruction and acknowledged compute/generalization limits, not evidence of superior procedural anatomical generation. Existing July photoreal baseline not relabelled simply to clear gate. Rank remains stale pending scoped baseline review. Reuse/repair/replace remains per-component evidence decision; no retirement authorized or performed. |
| 1970-08-23 | delivery-generalization | operator-correction | Examples including ribs are instances of estate-wide composition, not authorization for separate bespoke projects. Existing workflow owner must resolve typed component contracts, dependencies and units; expose inspectable Alpha artifacts; compose Beta with integration, performance and rights evidence; promote Production with acceptance and rollback. Reuse must be demonstrated on differing component classes; component-specific evaluators remain necessary. Current character delivery continues. Generic autonomous delivery is not yet verified. Identity-length refusal is containment, not completed dynamic identity repair. |
| 1970-08-23 | GR29 | candidate-verification | Frontier candidate t346 now propagates fs_attest_domain failures for single/all CLI paths instead of reporting ATTESTED after domain failure. Write1789860142 verified92636B. Actual native run1789860154 compiled but stopped earlier at missing knowledge/frontier_scan.conf under buildroot qualification cwd; domain-propagation behavior NOT yet exercised. Existing serving-root resource resolution must be reviewed, not silently copied as a workaround. Durable checked worklist publication and dispatch remain pending; no adoption. |
| 1970-08-23 | GR29 | candidate-verification | Existing nx_estate_path.ep_artifact_path now composed by frontier candidate for configuration and registry reads; no copied configuration or cwd mutation. Candidate92991B write0264. Actual CLI run1789860271 reaches requested absent domain, reports failed_domains=1 verdict=INCOMPLETE, exits2 instead of ATTESTED. Native helper regression pending; checked durable worklist write, remaining path consumers, review dispatch and canonical adoption remain incomplete. |
| 1970-08-23 | delivery | verified-enablement | Provenance tool profile gap repaired through existing capability mint0469: exact nx_asset_prov added, all46 prior grants and expiry preserved, profile47 readback verified and actual default-profile help0487 executes. Initial percent-encoded allow CSV mint0450 refused invalidcharset with token_emittedfalse; literal CSV succeeded. No credentials retained here. Root independently reconstructed converter9deltas baseline50748B b4b03d3e4920429bf4ba96dce090c0e5387f1ab6d8f6708d210ddb00eb2bc53d to55801B7ab526d90b962a4921b5cf4b0d6007cbcedf549350e04a6267807885f26bdf7e. Actual body export qualified; general concurrent writer serialization and zero-normal boundary remain under review before canonical adoption. Alpha page build0466 full30/30 output1233643B3bf6a6172f4d4a28ba7331801535f0d417f8d6ff2825ad5e1f2a19fab38cf0d6 is assembled, not yet published or visually accepted. |
| 1970-08-23 | delivery | input-lineage-review | Presence is not consumption: serving-root knowledge/profile_human.dat declares BodyParts3D-derived measured priors, but actual body mesh job1789859918 executed under buildroot and reports prof_rows=0 prof_rows_offered=0 prof_scale=0. Root read full receipt and source relative-file read. Do not assign this unused profile as candidate dependency or erase its separate third-party obligations. Primary references retained by native fetch0607 rf_9c830398.raw SHA0c2913b3c51f2c6b248ac054acfd0b2a72940386199ca95fa9aa848bd69746e0 and0608 rf_aa4e1530.raw SHA4a02a925c115a4f8e0e5018fbed2fcb9388fde1b0c0ce4a61a506bb44e0307ec. Actual recipe and dimorphism lineage still being checked. Browser candidate renders but collision transform refuses; experimental Alpha must expose this failure clearly rather than claim NPC readiness or wait for unrelated completeness. |
| 1970-08-23 | delivery | root-visual-review | Root independently viewed actualcandidate screenshot597813B SHA338f041e99894cfb64481aec755ead6cfb094daf8a3fc00ba692ca98a7cb985e through NodeMCP. Appearance FAILED: coarse toy-like anatomy, shoulder/neck seams, flared pelvis rim, rudimentary face and limbs; collisionproxy visibly refused. Experimental Alpha exposure authorized with explicitfailure labels; not photoreal or NPC acceptance. Neutral-white diagnostic was inconsistently labelled source albedo; assigned to existing UI owner. Exactconditional ownerAlpha policy reviewed against current rights schema and actual recipe; applies only reviewedasset hashes, no code relicensing or blanket productiongrant. Screenshot-upload profilegap repaired mint0773:48grants, prior47+nx_content_put, unchangedexpiry; actualcall0790 dispatched. Workflow preflight must discover required grants before dispatch instead of repeated root minting. |
| 1970-08-23 | delivery | default-transport-review | Attached nx_fs size on reviewedAlpha artifact failed-32004 measured414 with old response-body JSON-object diagnostic; same current-profile HTTPS MCP call0944 returned exact1233643B. Localconfigured bridge5fd9d226c3c9e4eea77b9bee6d9754650e2cbfc762fb0dad80bf233ba6149b5d496028B contains newJSON/SSE diagnostic and lacks observedoldtext; older07625/e734/560 installedbridges contain observedtext. Evidence points to staleattachedprocess, not newsourceparserregression. Attachedtransport remainsunqualified; currentHTTPS MCP pathworks. Conditional assetpolicy actualrow and exact source/GLB/descriptor verdicts verified0877/0878/0885/0886; scopedexperimentalAlpha approval applies, not genericSHIP_OK. |
| 1970-08-23 | GR2 | candidate-gap | Native procedural body candidate t374 generated through nx_body_proc + nx_body_gen: canon build=650 musc=500 noise=250 seed=374, 1700mm, 20 parts, 110 control rings, 1,807,104 triangles, truncated=0; artifact buildroot/knowledge/gates/beach-female-body-t374.nxmesh SHA d426a6bbada2d3fe0f525664caad535d8ea55edb6c4d762bd1ff2b7fc9e83679. Face gate 4/4 GREEN and deterministic face render job1789881061 landed 114173B. Existing nx_gltf_export is write-only; complete native absence scans jobs1789881227/1789881233 found no GLB importer, decoder or GLB-to-NXMSH2 owner. Face Alpha composition remains UNVERIFIED until a reusable adapter preserves identity/material/rig contracts and passes served review. No Beta or Production promotion. |
| 1970-08-23 | GR2 | determinism | Repeated nx_body_gen for t374 with identical canon/parameters produced 1,807,104 triangles and byte-identical 159,025,240-byte mesh; nx_filehash cmp identical=1, SHA d426a6bbada2d3fe0f525664caad535d8ea55edb6c4d762bd1ff2b7fc9e83679. Determinism is qualified for this recipe/seed only; anatomy, rigging, materials and visual acceptance remain open. |
| 1970-08-23 | GR2 | multiview | Native twin-face render for seed374 at yaw45 landed 114638B via job1789881347; front yaw0 landed114173B via job1789881061. Same identity vector and skin values retained. These are deterministic view artifacts, not visual-quality acceptance; matched Alpha/Beach served review still required. |
| 1970-08-23 | GR2 | adapter-contract | Shared prepared-character runtime contract read from buildroot/runtime/nx_character_prepared_runtime.nx: actor input requires pos/aj/aw/nrm/uv/idx streams, optional palette/bind, image/normal materials, geometryMode and shared GPU buffers; static geometry without source image is explicitly supported only when geometryMode=static. A GLB importer must produce these streams, preserve scale/units/material provenance and refuse unsupported skin/animation claims. This is the adapter contract; no implementation or Alpha promotion claimed. |
| 1970-08-23 | EC | resource-health | Native GET /api/health job1789881554 reports overall DEGRADED with all 16 listed services UP, mem_avail_permil=392 and swap_used_permil=1000; reason mem-pressure. This is an estate resource condition affecting search/adapter scheduling, not evidence of a Beach visual regression. No restart or fallback was performed. |
| 1970-08-23 | GR2 | build-contract | Native nx_sov_build_run job1789881827 correctly rejected standalone nx_gltf_export with nxasm rc102 UNDEFINED label: main. The exporter is a library, not an executable; this is a build-entry contract, not a compiler regression. Adapter qualification must use a main-bearing gate/consumer that imports the library. Build admission passed with native policy checks and bounded host measurements. |
| 1970-08-23 | GR2 | json-prerequisite | Native nx_sov_build_run qualification job1789881917 built and ran nx_json_test through nx_cc->nxasm_x86, no GCC, run-exit=0. `nx_json: emitter + tokenizer round-trip PASS`; staged artifact only, serving root deliberately unchanged. This qualifies the existing JSON tokenizer prerequisite for the future GLB accessor resolver, not the GLB adapter itself. |
| 1970-08-23 | GR2 | hires-candidate | Native nx_body_gen job1789882088 produced a higher-resolution procedural candidate: height1700mm, radial128, sub64, relief300, 20 parts, 110 control rings, 7,202,304 triangles, truncated=0, headroom1,186,304, artifact633,802,840B SHA483425e30501be39a12561cc70f5d3703231de32b52a2726134d9f0882695c7f. This improves geometric sampling only; no skin, rig, material, motion or visual acceptance claim. Resource cost must be measured before repeated use. |
| 1970-08-23 | EC | hires-resource-check | After 633,802,840-byte procedural candidate, native health job1789882133 remains DEGRADED: 16 services UP, mem_avail_permil=391, swap_used_permil=999, reason mem-pressure. No additional high-resolution generation or conversion dispatched; candidate retained for controlled qualification, not repeated blindly. |
| 1970-08-23 | AT46 | skin-gate | Native nx_skullskin_gate job1789882182 ran after exact grant repair: 31/32 checks PASS, verdict RED. Tissue-depth geometry passes, 31 landmarks located, no penetration, both eye patches built on sampled orbital rim, skin render written 73,203B; eye-pair adjudication fails (eyes=0, both canthi not placed). Total 80,907ms, peak RSS 61,912KB. This confirms a concrete face/eye integration defect; skin-depth math is not equivalent to photoreal appearance. No Alpha promotion. |
| 1970-08-23 | AT46 | eye-requalification | Direct capability-qualified nx_skullskin_gate rerun with argv=buildroot/knowledge/gates/beach-skin-t375.nxmesh used the correct knowledge root and completed 31/32 checks in 2627ms, verdict RED: tissue geometry, 29 located + 2 fallback landmarks, positive depth/no penetration, both eye patches and face detection pass; photomark still reports eyes=0. Shared nx_tissuefield_lib source now colors the negative-Z visible hemisphere, but nx_sov_build_run reused a stale cache key and did not compile that dependency change. Alpha promotion remains refused. |
| 1970-08-23 | EC | build-cache-imports | nx_sov_build_run --qualify reused the old nx_skullskin_gate cache after nx_tissuefield_lib changed; supported --no-cache forced a rebuild, but the qualification child still lacked its relative knowledge and skull inputs. This is an estate build-runner contract gap: cache keys must include transitive imports and qualification must anchor data paths to the declared knowledge root. No serving binary changed. |
| 1970-08-23 | EC | qualification-path-repair | Repaired nx_skullskin_gate resolves knowledge and cascade inputs from both serving-root and build-root paths. Native nx_sov_build_run --qualify --no-cache with argv=knowledge/gates/beach-skin-t375.nxmesh now runs 31/32 checks in 3411ms, reaches the photomark route, and returns the same honest RED eye-pair result (eyes=0); no missing-precondition skip remains. Staged ELF only; no serving binary or Alpha change. |
| 1970-08-23 | AT46 | eye-material-render | Shared nx_tissuefield_lib globe renderer now carries per-vertex material classes into triangle emission, so iris/pupil triangles are no longer forced to uniform white. Native no-cache qualification compiles this source, but photomark still reports eyes=0; the material repair is therefore unaccepted and remains staged only. |
| 1970-08-23 | AT46 | eye-fallback-telemetry | The photomark chooser now reports its documented skin fallback when the loaded cascade rejects eyes. Native qualification reaches route=1 and emits measured eye spans: right lateral=175, right medial=242, left medial=271, left lateral=345, pixels R=675 L=1547. Pair failure is named PM_PF_ENDO (medial gap 29px below the declared minimum relative to fissure width); this is a geometry/detection mismatch, not a pass. No release. |
| 1970-08-23 | AT46 | eye-pair-diagnosis | After documented cascade-to-skin fallback, the valid native run now reports route=1 and PM_PF_ENDO with measured canthi R 175/242 and L 271/345 (29px medial gap versus 67/74px fissures). This identifies the next repair as eye geometry/render alignment, not a gate threshold relaxation. No Alpha promotion. |
| 1970-08-23 | AT46 | eye-disc-candidate | Added explicit front-facing iris/pupil disc geometry to the shared globe renderer and qualified through native no-cache build. Artifact compiled, but photomark measurements and PNG byte size were unchanged (route=1, PM_PF_ENDO, eyes=0); candidate does not improve the witness and remains unpromoted. This prevents shipping a non-effecting patch. |
| 1970-08-23 | AT46 | eye-coordinate-candidate | Corrected the globe hemisphere and front discs to follow the declared skull anterior +z and tfl_push half-turn. Native qualification compiled and produced a changed 73536-byte render, but measured canthi and PM_PF_ENDO remained unchanged (31/32 RED). Candidate is retained for visual comparison only and not promoted. |
| 1970-08-23 | AT46 | eye-occlusion-candidate | Moved iris/pupil surfaces ahead of the configured lid thickness using the existing tissue-depth patch value. Native build succeeded, but photomark canthi and PM_PF_ENDO remained unchanged; artifact size changed only from 73536 to 73528 bytes. Candidate is not accepted or promoted. |
| 1970-08-23 | AT47 | eye-material-witness | Native no-cache qualification compiled and ran; dark framebuffer regions prove eye material is rendered on both sides (left=442 pixels x=194..232, right=858 x=274..324), while photomark still selects mismatched canthi (R=175..242,L=271..345), eyes=0,pair_fail=5. Diagnosis is ruler/landmark selection, not absent iris/pupil geometry. No Alpha promotion; next owner is photomark candidate selection with preserved negative controls. |
| 1970-08-23 | AT48 | photomark-core-valley-candidate | A shared photomark candidate raised fissure extraction from half-depth to a named core fraction. Native no-cache qualification compiled but remained RED 31/32: failure changed to PM_PF_ASYM with canthi R=196..219,L=330..342 and eye pixels R=256,L=140. Candidate was rejected and reverted; no serving binary or Alpha promotion. |
| 1970-08-23 | AT49 | photomark-baseline-restored | Rejected core-valley candidate was removed and native no-cache qualification rerun from current source. Baseline is restored: 31/32 PASS, dark eye regions present (442/858), photomark PM_PF_ENDO with canthi R=175..242,L=271..345. Staged only; no serving binary or Alpha promotion. |
| 1970-08-23 | AT50 | iris-disc-depth-alignment | Corrected shared iris ring vertices to use the same configured lid-depth surface as the iris center and pupil, removing a cross-layer placement bug. Native no-cache qualification compiled; dark-eye witness changed (left=472,right=862) but photomark remains PM_PF_ENDO, 31/32 RED. Candidate remains staged and unpromoted. |
| 1970-08-23 | AT51 | skin-material-path-audit | Native source audit of nx_tissuefield_lib/nx_trimesh shows qualification resets the renderer, sets no texture or specular material, and renders constant skin albedo under fixed Lambert lighting. The observed flat/halo-like skin cannot be solved by eye geometry alone. A shared configured material contract is required; no procedural-noise toggle was enabled blindly and no release changed. |
| 1970-08-23 | AT52 | material-contract-texture-candidate | Added reusable tfl_material_apply(T) driven by tissue_depth.conf and qualified texture_scale=13/specular=0. Native run compiled, emitted TFL-MATERIAL telemetry, and wrote a 143877-byte render, but photomark changed to PM_PF_ASYM (R=204..242,L=283..348), 31/32 RED. Texture 13 is retained only as a controlled candidate; canonical default restored to 0. No promotion. |
| 1970-08-23 | AT53 | material-contract-default | After the texture candidate comparison, canonical tissue_depth.conf now defaults skin_texture_scale=0 and skin_specular_strength=0 while retaining tfl_material_apply(T) as the reusable callable path. Native no-cache run emitted TFL-MATERIAL texture_scale=0 specular=0 and returned the previous 31/32 RED PM_PF_ENDO result. No serving binary or Alpha promotion. |
| 1970-08-23 | AT55 | alpha-skin-gate-promotion | Guarded nx_organ_ship job1789884913 built nx_skullskin_gate, proved 33/33 GREEN, promoted SHA256 72755ac223b9d912aa31f1a99ba10e90593664a7353fdb06e26aea2168aa534f, installed byte-identical offc twin, and found it already registered. The outer ship timed out at final verification (exit=-1 TIMEOUT-OR-SIGNAL); native nx_filehash independently confirms live nx_skullskin_gate.elf equals buildroot/_build/nx_skullskin_gate.sov.elf (341360 bytes). Behavior probe could not run because the ship comparand .sov.elf.new was absent. Alpha primitive is promoted with verification caveat; no Beta/Production claim. |
| 1970-08-23 | AT56 | alpha-live-behavior-verification | Native nx_behaveprobe self nx_skullskin_gate ran the promoted live executable twice: both exit 0 with substantive output (live_bytes=5320, other_bytes=5319), but differ at offset 2017 by one byte. This is a deterministic-output gap likely attributable to runtime telemetry, so Alpha is live and behavior-clean but not byte-stable under self comparison. No Beta/Production promotion. |
| 1970-08-23 | AT57 | graphics-and-swarm-rank-current | Native nx_compare_rank completed with graphics declared=56, done=11, open=45, finish=3, and swarm done=0, open=11. Both rank artifacts were persisted by the sovereign ranker. Graphics currently declares no sotatarget or treat rows, so frontier and per-rung MCP/API/workflow/agent treatment remain UNDECLARED; swarm critical rungs remain DARK-CONTRACT. This is a measurement of roadmap state, not a quality claim. |
| 1970-08-23 | AT58 | alpha-ship-race-diagnosis | A fresh native no-cache qualification in isolation passed 33/33 GREEN (job1789886387), while the guarded ship run during overlapping self-probes refused at 32/33 RED and self behavior probes alternated 0/1 exits with differing output lengths. The gate writes shared /tmp/nx_skullskin_gate artifacts, so concurrent runs can contaminate each other; this is a real reproducibility defect, not permission to normalize bytes or weaken the gate. No new promotion claimed. |
| 1970-08-23 | AT59 | deterministic-skull-fixture | Recovered the canonical skull fixture already retained as buildroot/knowledge/gates/beach-skin-t375.nxmesh.skull.nxmesh, copied it through nx_filecopy with byte verification to nx_skullskin_gate.skull.nxmesh, and changed the gate to prefer the repository/root fixture before shared /tmp scratch. This removes time-dependent SDF fallback from normal Alpha and ship qualification; an invalid sparse mesh still fails the anchor tooth rather than being accepted. |
| 1970-08-23 | AT60 | alpha-skin-ship-recovered | After the canonical fixture/path repair, guarded nx_organ_ship completed OK: 33/33 GREEN, ADOPT-GATE APPROVED, live and staged SHA256 51676885f7c4c039b129446e407204b9d20cdcc13802f35c4dc64bddfa6b341f, 341720 bytes, byte-identical offc twin, already registered. The behavior evidence is explicitly DIFFERS because this source is ahead; no Beta/Production claim and no served UI verification. |
| 1970-08-23 | AT61 | alpha-live-self-probe | With the promoted artifact and no extraneous argv flags, native nx_behaveprobe self ran both live executions cleanly (live_rc=0, other_rc=0, equal substantive 5293-byte captures). Raw output differs at offset 3545 because the gate emits run-time/host telemetry; this is a probe comparison limitation, not a failed gate. A schema-aware volatile-field comparison remains the next shared probe improvement; no regression was accepted from raw byte equality. |
| 1970-08-23 | AT62 | ranked-roadmap-contract-gap | Fresh sovereign graphics rank completed: declared=56, done=11, open=45, finish=3, bars fresh=1/2 attested; no sotatarget/rungrole/treat rows are declared, so current SOTA comparison and per-rung MCP/API/primitive/workflow/agent treatment remain explicitly unmeasured. Swarm remains 0/11 done with dark contracts. This is a roadmap-contract gap, not a quality claim and not permission to invent thresholds. |
| 1970-08-23 | AT63 | genome-avatar-precondition | Native nx_bodyfit_canon_gate qualification built through nx_cc->nxasm and returned SKIP (not PASS): required canon_merge2.dat is absent. A complete sovereign find scan covered 85,321 files with zero matches. No synthetic replacement was created; GR2 remains unqualified until its canonical input is produced or its owner is corrected to an existing authoritative artifact. |
| 1970-08-23 | AT64 | candidate-canon-validated | Existing candidate buildroot/knowledge/gates/beach-female-canon-t374.dat was evaluated by the canonical nx_canon_solver through the sovereign native build path: parts=20, rings=110, CLEAN, exit=0. It is not promoted or wired because nx_bodyfit_canon_gate currently assumes the missing 16-part canon_merge2.dat and the candidate has no adoption/provenance record. Next owner work is a data-driven canon source/part-count contract, not a hardcoded replacement. |
| 1970-08-23 | AT65 | data-driven-canon-selection | Changed nx_bodyfit and nx_bodyfit_canon_gate to read reviewed knowledge/bodyfit_canon.path, trim transport whitespace, derive the part count from P rows and compare identity/extreme outputs against the selected artifact. Historical canon_merge2.dat remains the fallback; no serving binary changed. Qualification job1789888721 is still running after repeated authoritative polls, so compile/runtime verdict is pending. |
| 1970-08-23 | AT66 | canon-runtime-budget | After the data-driven path repair, nx_bodyfit and nx_bodyfit_canon_gate both compile natively. Full gate execution against the 20-part/110-ring candidate exceeded its configured 300000ms deadline and was killed/reaped; no qualification or promotion claim is made. The richer canon is structurally clean but needs a measured evaluation-budget/LOD strategy before it can enter the photoreal workflow. |
| 1970-08-23 | AT67 | canon-lod-inventory | Complete glob inventory found nine canon candidates. The body-semantic and beach-female candidates are all 20-part/110-ring 3.5-3.6KB canons; the historical canon_male.dat is a smaller 5-part/45-ring artifact but is incompatible with the current face-rule contract. No validated lower-detail LOD exists for the selected family, so no substitute was wired. The full candidate remains structurally clean but runtime-expensive. |
| 1970-08-23 | AT68 | lod-bridge-inventory | Estate search found nx_mesh_decimate (integer vertex clustering) and nx_nxa_tier, but no existing NXMSH2 load→decimate→write bridge connected to nx_bodyfit. The bodyfit runtime uses tfl_mesh_load for NXMSH2; decimation is an NxMesh library path. The reusable next rung is this format bridge with error/provenance checks, not a second reducer. |
| 1970-08-23 | AT69 | nxmsh2-bridge-scope-narrowed | Existing nx_tissuefield_lib already supplies tfl_mesh_load and tfl_mesh_write for NXMSH2, including welding, averaged normals, bounds and mean-edge telemetry. The LOD gap is therefore only a representation-compatible TFL decimator and a bodyfit evaluation hook; no duplicate parser or writer should be built. |
| 1970-08-23 | AT70 | tfl-nxmsh2-decimator-green | Added the representation-compatible tfl_mesh_decimate reducer and a native nx_tfl_decimate_gate. Qualification passed 5/5: input load, reducer return, vertex reduction, triangle reduction and NXMSH2 output write. On beach-skin-t375.nxmesh, 7,739,992B input became 1,329,808B at explicit cell=50 u10. Shared library compiles through nx_cc->nxasm; no serving promotion yet and bodyfit hook remains next. |
| 1970-08-23 | AT71 | tfl-decimator-alpha-promotion | Guarded nx_organ_ship job1789889781 promoted nx_tfl_decimate_gate sha=3bc1448aa9da8d47c785d07cca47f625fb9f66cf1bf67ef0eaa8088900d55d7f; registered and authorized after cap mint job1789889884; invoked via organ_run job1789889899 with 5/5 GREEN. This is a reusable Alpha primitive; no served UI/Beta/Production claim. |
| 1970-08-23 | AT72 | native-face-render-grant-and-health-recovery | Exact nx_twinface grant was missing; management mint job1789890108 issued scoped cap, original native render was retried with that cap and completed job1789890124: PNGWRITE RESULT=LANDED bytes=95255, TWINFACE seed=42 yaw=0 w=512 h=384. During the run /api/health reported DEGRADED from measured memory pressure avail_permil=367 and swap_used_permil=1000; nishi_search returned complete 503 evidence, then recovered on the next native call. No Beta/Production claim and no visual adjudication from the remote PNG in this session. |
| 1970-08-23 | AT73 | alpha-beta-stage-contract-repair | Audited existing Beach owners and found Alpha context mislabeled stage=beta while Beta marked Alpha route=null. Hash-guarded native edits job1789890386 (Alpha) and job1789890408 (Beta) corrected the contracts. Guarded publisher verified both exact artifacts already matched their live owners: job1789890570 Alpha UNCHANGED bytes=1245450 and job1789890580 Beta UNCHANGED bytes=1204542, writes=0. Browser/served interaction review remains unavailable in this session, so this is a source/live-owner contract receipt, not a visual acceptance claim. |
| 1970-08-23 | AT74 | procedural-female-alpha-candidate | Reused existing nx_body_proc -> nx_body_gen -> nx_skullskin path. Generated canon knowledge/gates/beach_female_proc_20260920.dat: sex=800 build=700 musc=500 noise=100 seed=42 stylize=0, 3631B; generated 1750mm native body knowledge/gates/beach_female_proc_20260920.nxmesh: 20 parts, 110 control rings, 461568 tris, no truncation; generated skin with scoped skullskin grant job1789890664. Re-ran nx_skullskin_gate with exact grant minted job1789890707 against the generated skull: 30/33 RED. Positive evidence: field nonpenetration, eyes=2, face found, skin render written. Failing evidence: one unlocated landmark/anchor distance, intercanthal index 272 outside canon 380..420. This is a valid Alpha candidate and a useful liar-killer failure, not a Beta/Production asset. |
| 1970-08-23 | AT75 | procedural-sex-parameter-triangulation | To test whether the AT74 eye-spacing failure was a sex parameter, generated a second canon with sex=300 build=700 musc=500 noise=100 seed=42 stylize=0, then native body+skin. The skullskin gate still measured intercanthal_index=272‰ against 380..420‰ and again passed 30/33 RED; eye spacing is invariant across sex=800 and sex=300. This rules out a parameter-only fix and directs the next repair to the shared face/eye placement contract, while preserving the strict canon band. |
| 2026-09-23 | GR35 | measure | 2026-09-23 scoped Alpha inspection repair delivered at /alpha/beach title b1790154993: page1125385B SHA515e85baba3dd29f3d32f7cd64a8b4c0813a799b0733da06ecd4af97d881daa2, actual shared385136B SHA335df0bdd07f4d8b3f527438b5333a24a82627d827a9823bc1dff3a8bdf7045a. Current canonical native camera owner adopted guarded, memory layout preserved; native112/112, actual WASM53/53, pair40/40, pagefit32/32 and full JS referee20647refs0findings. Existing workflow1790155327 published with30 dependency identities verified,29 prior dependencies unchanged and prior1be4d1e retained. Root full-Chromium Intel xe-lpg served review09:27Z: Catwalk load, Fit full body, Body180degree orbit, resident face front/back, Return to gameplay and dated Progress record PASS; pageerrors0. Imported Face explicitly unqualified; refusing Face leaves orbit inert until Body reselected. Imported collision capability absent. Resident close-up eyes/hair/skin rejected; terrain/player/NPC contact, motion, performance and photoreal quality unaccepted. Beta/Production unchanged. Retained delivery buildroot/knowledge/gates/beach-camera-resume-20260923-delivery.json; served proof beach-camera-served-review-20260923.json SHA1984164d86220ef9232449506d305c74010571a040f0aeb0d1c566d8fb1c7b63;8 reviewed screenshots. Next existing owners: native imported-body integration and inspection refusal state; no new viewer. |
| 2026-09-23 | GR35 | measure | 2026-09-23 Alpha imported-contact repair published at /alpha/beach title b1790160590: page1216243B SHA79cef81700acf33f5c3daeaa3ce6c60d5b9f44f3aff7ecc7a73493066349d648; shared451054B SHAd986bf9b35d1b4e154f3c2b640383584299fb1d7ec30dd7442ad9e82e407be6e. Existing native body/registry/motion owners composed with scene-derived capacity, camera/saves/morph/layout retained; native140/140, actual WASM78/78, pair40/40, pagefit32/32 and native JS referee20728refs0findings passed. Guarded workflow1790160960 retained predecessor515e85ba and29 unchanged dependencies; all63 publication phases rc0. Root served Chromium Intel xe-lpg review: Catwalk proxy blocks actual player from both sides; disable allows passage and reenable blocks with samehandle1048577/revisions1,2,3. Imported Face remains unqualified but Body orbit stays active without moving player. Distinct id/same source binds2097153; actual native source playback updates proxy and stops at authored clip end. Observed player feet3409Q8 equals native terrain3409; render-source interpolation3409.92578125Q8; non-atomic sample, controller support unavailable, no anatomical/NPC or image-derived geometry acceptance. Reproduced defect: same-id replay refuses while residents are culled, succeeds when framed; existing two-owner readiness repair reviewed for private qualification. Skin/face/hair rejected, arbitrary through-dune occlusion/autonomous navigation/Native Browser parity unqualified. Negative unmatched QA timing: world+cast105.971712ms, raw-loop p50=108ms p95=283ms/256 samples including pauses; no60FPS or attributed improvement. Beta/Production unchanged; no rung closure. Retained delivery buildroot/knowledge/gates/beach-body-resume-20260923-delivery.json; served proof beach-body-served-review-20260923.json41764B SHA812496e63325a3d308435fb6d729b7402461f65637b1837c35221b0c7dc93f0e,3 reviewed local screenshots. Next: durable prepared readiness and identity-first replay through existing renderer; no WASM rebuild. |
| 2026-09-23 | GR35 | measure | 2026-09-23 Alpha prepared-character readiness repair published at /alpha/beach title b1790164131: page1218872B SHA52b324caa21e07a1dc793e57050a4201214c108be12a6dc1a59cb6dcade1d695; shared451054B SHAd986bf9b35d1b4e154f3c2b640383584299fb1d7ec30dd7442ad9e82e407be6e unchanged, plain709fc0b3 unchanged. Existing prepared runtime and host renderer converge durable readiness independent of resident culling, snapshotted id+three-artifact replay identity, conflict-aware pending requests and final renderer-lifetime refusal. Matching replay preserves body/allocation/configuration. Shared templates initialize with zero resident rows while row-dependent draws stay gated; zero-row validation cannot relabel old resident submissions. Canonical f3893ffa/24a2dcba guarded adoption and complete511612B readbacks; native pagefit32/32, full JS referee20795refs0reports with negative controls, exact-function isolated Chromium66/66 with synthetic boundary stated; independent reviewcef90c3f accepts scoped change. Existing Alpha workflow1790164641 completed63rc0 phases, all30 dependencies unchanged, prior79cef817 retained. Publication verified, served-review result pending at 2026-09-23 12:07:00 UTC. State: published; waiting for retained actual WebGPU interaction/visual review, next join root proof before scoped acceptance. Retained delivery buildroot/knowledge/gates/beach-prepared-readiness-20260923-delivery.json SHAfd2724eee3ba3586cd3dea488e3013e51a10e3dc7439b6b2bb94bdfb84ad57da; adoption4d6be557 and full publisherd9e3f1d4 retained. No appearance/performance improvement or Beta/Production acceptance claimed; skin/eyes/hair and anatomical support remain rejected/unqualified. |
| 2026-09-23 | GR35 | measure | 2026-09-23 served Alpha readiness review complete: https://nishifamily.com/alpha/beach title b1790164131, page1218872B SHA52b324caa21e07a1dc793e57050a4201214c108be12a6dc1a59cb6dcade1d695; shared d986bf9b unchanged. Actual Chromium143 Intel xe-lpg WebGPU confirms offscreen import and same-id already-loaded replay with identical body handle/revision/anchor/counts and artifact requests7→7. Zero resident-row inference uses the existing valid hair-diagnostic guard, not stale draw telemetry. Same-source distinct scene IDs bind1048577/2097153. Frame/orbit/Return pass; imported Face explicitly refuses and retains Body admission. Conservative whole-mesh proxy blocks enabled, permits disabled traversal and blocks opposite-side motion after reenable; same handle revisions1→2→3. One authored clip ends at1252698752Q with changing bounds while second instance stays revision1; world navigation remains unbound. Dated Progress summary visible, prior history preserved, page-script errors0, test context closed. Complete native-hashed served proof buildroot/knowledge/gates/beach-readiness-served-review-20260923.json38183B SHAe7001089603445e100ee7a9f9bea75b7eab0f11d521c4887308071c9328c5896; delivery buildroot/knowledge/gates/beach-prepared-readiness-20260923-delivery.json. Scoped readiness/body-camera interaction accepted; GR35/GE31 remain open for rejected skin/face/eyes/hair, anatomical/sole support, arbitrary occlusion, autonomous navigation/evaluation and matched performance qualification. No60FPS/photoreal/Beta/Production claim. Both telemetry IDs have ERR_ABORTED plus204/accepted observations; cause/attempt counts unknown. Measured aiming, hidden-workspace30s timeout, asynchronous reads and block-local evidence retention are observer corrections, not delivered runtime repairs. Screenshot binaries remain local. Prior negative unmatched performance evidence retained; no token/time/swarm ROI claim. |
| 2026-09-27 | GR57 | queue | Landed 2026-09-27 from the beach-lounger reference-still analysis. Estate incumbents found by nx_capsearch: nx_gm_cloth_lib (PG29 XPBD), nx_garment_gen (garment shells), nx_kubelka_lib (KM core), nx_skinfit_lib; no sheer-over-skin BSDF anywhere in 2181 tools and 6011 libs (the ranker's corpus_complete=0, so a floor). The skintwin ranker reads ST5 km_layer_over as DARK-CONTRACT: the organ exists, the symbol is undeclared, so the layered core itself is unbuilt. |
| 2026-09-27 | GR62 | measure | nx_ladder graphics 2026-09-27: 56 rungs placed, targets 3, verdict LADDERED before this landing; skintwin and softbody, the two boards GR57-GR60 depend on, read NO-LADDER and are laddered the same session. |
| 2026-09-27 | GR63 | lesson | A SEARCH-ENGINE SUMMARY IS NOT A WITNESS: a search result asserted an NVIDIA hybrid real-time subsurface-scattering talk in the SIGGRAPH 2026 Advances course; the mirrored course page and NVIDIA's events page (1010241 bytes) carry the token subsurface ZERO times, so no bar was raised on it and the claim is recorded as unwitnessed. |
| 2026-09-27 | GR10 | measure | paced nx_frontier_scan attest graphics 2026-09-27: bars=7 fetches=112 venue-tag fan-out rows=49 UNREACHABLE=0 newer_sources=8 stale_bars=1 verdict=STALE-BARS; the stale bar was SB-RTGI (2026-07): the SIGGRAPH Asia 2026 comment tag answered Gaussian Light Transport 2609.11430 published 2026-09-10, and the bar's own listing re-fetched today also reads newest=2026-09; the unpaced run had reported that tag UNREACHABLE (30 of its first 40 venue-tag rows), an abstention nx_barfresh cannot convert into STALE |
| 2026-09-27 | GR10 | land | SB-RTGI re-adjudicated to 2026-09 from the mirrored bytes (knowledge/fetched/rf_a2bd0c41.raw, pinned as ref arxiv_siga_rtgi_2026_09_27, abstract read through nx_arxiv_digest): the September entry expresses GI as a 13D Gaussian mixture fitted on the rendering-equation residual and renders view-independent solutions in milliseconds; July's Split Radiance Cascades and OctaOctree stay named in the bar; nx_barfresh graphics re-read after the edit is the acceptance |
| 2026-09-27 | GR10 | land | ref arxiv_siga_rtgi_2026_09_27 re-pinned 2026-09-27: the fleet refs gate read PROV-MISMATCH because the row cited the scanner's raw capture (rf_a2bd0c41.raw) under the feed's self link while the fetch journal holds that mirror under the export.arxiv.org query; the same query was re-fetched through nx_research_fetch as knowledge/fetched/cmp_graphics_siga_rtgi_202609.xml (2782 B, sha256 4a6d43bf) and the row now cites the journaled url, mirror and pin. LESSON: cite the url the researcher fetched, never the url the feed prints about itself |
| 2026-09-27 | GR10 | queue | fleet refs gate 2026-09-27 (37/44 RED) names six PIN-MISMATCH rows that pin knowledge/library/gfx_*.html: graphics.refs terraindiff26, kerbl3dgs, karis2021, bitterli2020 and procgen.refs terraindiff26, kerbl3dgs. Those library files are rewritten by the papers-index beat (gfx_sig2026.html 288701 B today carries no NX-DERIVED stamp yet differs from every declared pin), so a pin on them is stale by construction. Remedy for the row owners: cite the Ke-Sen listing mirror in knowledge/fetched (cmp_graphics_kesen_sig2026.html, sha256 7e6c166c, immutable) and re-pin; the beat-written library artefact should carry the NX-DERIVED stamp so no future row pins it |
| 2026-09-28 | GR63 | measure | THE CROSS-BOARD PHOTOREAL CRITICAL PATH IS MECHANICAL (2026-09-28, nx_compare_rank program graphics:GR63 on promoted b5450195): 3 boards, 19 nodes, 7 done, 12 open, 0 unresolved, 0 cycles; prerequisites-first path #1 GR42 (beauty-tier first frame) #2 GR55 (skin-tone fidelity of the live render) #3 GR30 (world-quality referee) #4 GR14 (linear-light pipeline) #5 GR23 (one colour transform) #6 gameengine:GE6 (native raster lane) #7 gameengine:GE15 (frame-pacing referee) #8 GR62 (reference-still DE2000 referee) #9 GR61 (camera model as data) #10 gameengine:GE41 (browser-engine parity bars) #11 GR63 (real-versus-render bench, the acceptance ruler) #12 gameengine:GE56. Per-board: XDEP domain=graphics edges=16 done=9 open=7 unresolved=0 after the GE40 path fix and the ST6 repoint (before them: unresolved=3). Read: gameengine GE51 reads DONE on an OPEN prerequisite GE56, a done-on-open inversion the program verb does not yet name as a class (queued) |
| 2026-10-01 | GR42 | land | THE TRACER'S FIRST EVENT TAKES THE CAMERA'S UNIT (2026-09-29, for pano360 PN25): gp_trace_u, gp_march_u and gp_norm3_u in nx_gi_path_lib and wg_sky_u in nx_worldgen carry a per-ray unit u for the primary ray (positions d*t/u in the march and refinement, twater with u, the sky's up and sun terms divided by u), bounce rays stay fx1024 Monte Carlo draws, and every old name is kept as its fx1024 wrapper. PROVEN BYTE-NEUTRAL for every fx1024 caller by a MATCHED-VINTAGE control: the pre-change nx_worldgen and nx_gi_path_lib were rebuilt by reversing the exact edits on scratch copies, proven by hash against the flow-back's agreed base (2560674f, 513c7f4c), compiled in the same lane the same hour, and the control pinhole render (seed 77, 480 x 288) came out eba1436c -- identical to the candidate. nx_gi_path itself is NOT promoted: its served binary (577d0bb3) is 26.6 days old and a rebuild carries other seats' work, so its ship is this lane's owner's, with the stage named |
| 2026-10-01 | GR42 | lesson | A NEUTRALITY TEST AGAINST A BINARY OF ANOTHER VINTAGE CONVICTS A NEUTRAL CHANGE: the pinhole rebuilt with the generalised tracer differed from the served nx_gi_path while every ray-event count matched; the served binary predated other seats' edits. Rebuild the pre-change source in the same lane the same hour and compare against THAT |
| 2026-10-01 | GR62 | measure | REFERENCE 1 INTAKE AND ITS FIRST DEFECT, 2026-09-30: the operator supplied Wikimedia Commons private-ref:6a8ea809; licence read from Commons' own extmetadata: CC BY 2.0, author Tino & Tina - fr (Flickr Patmos rest), Restrictions empty; mirrored private-ref:6a8ea809 sha256 d421bd74 at exactly Commons' declared 186403 B (1174 x 673), metadata mirror private-ref:6a8ea809 5672c82a. Scene-class colour reference only, never the person's twin, not a refs row (refs mirrors are republished). nx_skinsample measure on the whole frame: mask_permil=518, L 62.014 a 4.075 b 12.912, ITA 42.9 deg -- the YCbCr skin rule admits the beige sand and stones, so a whole-frame beach skin mean is a sand mean. OUTSIDE ORACLE landed for scoring the fix: private-ref:6a8ea809 (13 conservative interior boxes: 4 skin, sun and dappled shade; 3 sand, sun and shade; sea, towel, rock, vegetation, hill; hair as a skin-coloured negative), read by Claude vision from a byte-identical copy; a mask's recall is its admitted share of skin-box pixels, its false admission the share of sand, rock, towel and hair pixels. |
| 2026-10-01 | GR59 | land | SUN-TAN AND TAN LINES AS A DERIVED MELANIN FIELD BUILT AND PROMOTED 2026-09-30. gr_melanin_field landed in runtime/nx_skin_ita.nx (11339 -> 19116 B, additive, pure integer, no new imports): per-vertex melanin = constitutive + facultative, the facultative part computed from each vertex's OWN garment-coverage history and the declared exposure, both data rows in knowledge/skin_tan.conf. The absorber chain is the new runtime/nx_skin_tan_lib.nx (21525 B): melanin -> epidermal absorption -> a finite Kubelka epidermis over a semi-infinite dermis through the estate's km_layer_rt, km_layer_over and km_r_from_ks -> CIE XYZ on the CIE's own 1931 2-degree CMF, D65 and A tables at the 5 nm abridgement CIE 15 accepts over 380-780 nm -> cs_lab_from_xyz -> ITA, and a relight that re-integrates the same reflectance under A and adapts it to D65 through cs_cat_build (Bradford). The constitutive melanin is SOLVED from the genome's ITA on this organ's own locus, never typed. Dose-response and persistence from Ravnbak 2008 [ravnbak_tankinetics_2008] (MMD per Fitzpatrick type; the pigmenting dose fraction lowered 2x and 3x by 6 and 12 exposures; Table II steady state; peak a week after the last exposure; a linear fade over the published days to total fading); absorber and scatterer spectra from Jacques 1998 [jacques_skinoptics_1998]; the band from Amano 2020 [amano_naturaltan_2020]; colorimetry on [cie_cmf1931_2deg] [cie_illum_d65] [cie_illum_a]. nx_skin_ita_gate 9 -> 25 teeth (all 9 incumbents still GREEN), built sovereign 127720 B sha256 5dfb04e0b8d0c00a68fa9d50fd8b57f1d693537fe6abe9d3e78d7ede23912253, staged run 25/25 GREEN, contentdiff lost exactly 1 run (the gate's own rewritten verdict note, named), promoted under allow_capability_loss with prev banked (pre-change live 35d5fdee 28666 B), _offc refreshed, LIVE run 25/25 GREEN (nx_job_run pid 32241). The first render-side adoption of the skintwin ST1 referee cs_de2000 |
| 2026-10-01 | GR42 | measure | TRIANGULATED BAR as of 2026-10-01 for reflections and global illumination: the best in class is three variance-reduction families, not one method. Resampling: ReSTIR reuses candidate light samples spatially and temporally, unbiased at equal error 6-60x faster than the prior best and biased 35-65x faster at some energy loss, 3.4 million emissive triangles under 50 ms at 8 rays per pixel. Learned guiding: practical path guiding learns the radiance field in an SD-tree, unbiased, no hyperparameter tuning, MSE 0.018 against 0.104 for Vorba et al. at equal time, and Intel ships the family as a production library with per-frame learned fields on surfaces and in volumes and product sampling with BSDF lobes. Fitted transport: the September 2026 entry minimises the rendering-equation residual over a 13D Gaussian mixture and renders view-independent solutions in milliseconds. The September cs.GR listing (170 entries) carries only that paper and RenderFormer-V2 as light-transport titles, none ReSTIR or path guiding, October has no entries yet, and arXiv holds only three ReSTIR entries, newest ToF ReSTIR in May 2026 ; witnesses [bitterli2020] [ppg2017_t1001] [openpgl071_t1001] [arxiv_siga_rtgi_2026_09_27] [csgr202609_gfx_t1001] [csgr202610_gfx_t1001] [restir_search_t1001] ; where they disagree: ReSTIR buys real time by sharing samples across pixels and frames and accepts bias or correlation, path guiding stays unbiased and pays a training phase and memory, the Gaussian fit replaces sampling with an optimisation whose view independence makes glossy reflections its weak case, and the Intel library calls itself pre v1.0 with an API in flux while the 2017 paper calls the same family production compatible ; the estate's running lane builds a brute-force unidirectional integer path tracer (nx_gi_path: primary ray, shadow ray, N-bounce diffuse with a hashed sampler, the sky as the light, a heightfield march with no triangle BVH; first frame 4 spp and 2 bounces at 480x288 in 21.96 s) and it is BELOW the triangulated best because it carries none of the three reducers -- no reservoir resampling, no learned guiding, no cached or fitted transport -- so its noise at equal time is the noise all three witnesses were built to remove |
| 2026-10-01 | GR59 | measure | THE ESTATE'S NUMBERS BESIDE THE SOURCES' (live nx_skin_ita_gate 5dfb04e0, 25/25 GREEN, values by gv_kv, one declared genome and one declared scenario, not a population): genome hue 790 asks ITA 49.863 deg and the solved constitutive melanin is 55404 ppm, 5.54 percent, INSIDE the 1.3-6.3 percent Jacques gives for light-skinned adults [jacques_skinoptics_1998], a range the solver was never shown. A three-week beach holiday at the density of Ravnbak's 12-exposure protocol (12 sessions of 3.2 SED, about 0.6 MMD for type II, a UPF 30 garment whose edge wanders 296-304 mm) read at its peak, day 26: the line falls at 301 mm, always-covered skin keeps 0 ppm facultative, always-exposed skin gains 21774 ppm (Table II 9.0 over the type-II baseline 22.9 [ravnbak_tankinetics_2008]). Inside Lab 72.75 3.01 19.18 (ITA 49.86), outside 66.83 4.83 21.22 (ITA 38.42) against the measured outer-forearm mean ITA 38.8 [amano_naturaltan_2020]; DE76 6.51 against the measured natural tan 6.03 SD 2.64 (band 3.39-8.67); DE2000 4.99 by cs_de2000; dITA 11.45 deg inside the band 4.60-13.59 that the same measurements imply at this subject's protected colour; relit under illuminant A and adapted to D65 by Bradford, dITA 11.27, a shift of 0.18 deg against a bound of 4.49; the line's optical-density log-log exponent 2.01 against melanin's 3.33 and a neutral paint's 0.00 (nearest class decides, the paint is told apart); the D65 white summed from the mirrored tables Xn 95.042 Zn 108.880 against 95.047 108.883 [cie_cmf1931_2deg] [cie_illum_d65]; persistence: 21774 ppm at day 26, exactly half (10887) at day 110, 0 at day 194, 175 days after the last session as the paper measured |
| 2026-10-01 | GR62 | land | THE REFERENCE-STILL REFEREE LANDED, contract symbol present: gr_refstill_referee with gr_rs_acc_new and gr_rs_acc_add appended to buildroot/runtime/nx_colorsci_lib.nx (+8357 B additive, 38414 to 46771 B, sha256 eec69f03; compile-proven inside nx_refstill_lib and gated there first, only then moved, so no importer of colorsci ever saw a broken state). buildroot/runtime/nx_refstill_lib.nx carries the two-layer two-chromophore skin model (the dermis IS sf_forward over km_r_of_mix, the epidermis a double-pass melanin filter), every coefficient computed at run time from private-ref:fe89c6bb (sha256 31c01400) whose rows cite the mirrored Jacques 1998 Skin Optics and the Prahl haemoglobin table; the scale-free residual sf_sse(k t, m)/sf_sse(m, 0); the bound (sqrt(floor)+sqrt(quant))^2 with the lattice floor measured over every cell centre and edge midpoint and the quantisation term half the larger adjacent 8-bit code step per pixel; the declared regions private-ref:b7de2687 (sha256 a8d08af3, CHOSEN boxes, stated so); the class reducer. nx_refstill PROMOTED 142080 B sha256 4ee8bcbc, offc installed, registered (measure, class, oracle). nx_refstill_gate PROMOTED 110897 B sha256 cb8ffdac, offc installed, registered, rostered and mapped in graphics.gates; live run 22/22 GREEN (nx_job_run pid 9609): T3 sand-only neg-control BITE-PROVEN (5 sand colours the AT23 YCbCr rule admits, the model admits 0, closest residual 1413779 e-9 against its bound 176400 e-9), T4 positive control 16 of 16 chain-made colours recovered, T5 Sharma pairs 17 and 20 at 27.149 and 19.454 exact, T6 ITA quartiles exact, T8 sf_fit recovers the dermis blood fraction (22583 vs 22571 micro). Class file private-ref:d9101c46 (sha256 a2a387ad): 4 ADMITTED Commons stills (private-ref:6a8ea809 CC BY 2.0, private-ref:06caaa3c CC BY 2.0, private-ref:ff3b177c CC BY-SA 3.0, private-ref:80de5b96 CC BY 2.5; each mirror equals Commons' declared size and is pinned by nx_filehash) and 2 NOT-ADMITTED (licence No restrictions); not refs rows, nothing published. nx_refintake not used: its unit is a declared subject (a person slug), the opposite of this class's never-an-identity contract. |
| 2026-10-01 | GR62 | measure | THE DEFECT AND ITS FIX SIDE BY SIDE, nx_refstill class on the promoted binary 4ee8bcbc (nx_job_run pid 26615, /tmp/nx_refstill_class_live_1.out), every pixel to one region, partition SUMS on all four. private-ref:6a8ea809 790102 px: model skin 369 permil, L 49.465 a 8.086 b 10.735, ITA median -15.5 deg (q1 -72.8, q3 52.4) against the AT23 whole-frame YCbCr mean 518 permil, L 62.014 a 4.075 b 12.912, ITA 42.9; the model refuses 131293 YCbCr pixels and the bright sand becomes its own region (137 permil, L 78.570 a 0.462 b 15.158). private-ref:06caaa3c 3812256 px: model 146 permil L 49.390 a 9.448 b 9.685 ITA -23.0 vs YCbCr 142 permil L 55.505 a 8.505 b 10.220 ITA 28.3. private-ref:ff3b177c 9980928 px: model 400 permil L 59.304 a 3.658 b 9.212 ITA 47.7 vs YCbCr 408 permil L 59.614 a 3.540 b 9.336 ITA 45.8. private-ref:80de5b96 5018112 px: model 64 permil L 44.605 a 12.966 b 35.318 ITA -4.0 vs YCbCr 157 permil L 62.954 a 6.025 b 39.556 ITA 18.1. Bands private-ref:1d118014 (sha256 b882576e, 1348 B, NX-DERIVED, carries a CAVEAT row): skin centre L 50.691 a 8.540 b 16.238 radius dE00 10.801; sky 72.936 -1.823 -15.640 r 9.848; sea 50.095 -2.707 -14.652 r 7.178; sand 71.526 0.982 17.119 r 11.905; fabric 48.505 13.937 25.160 r 26.910. |
| 2026-10-01 | GR62 | measure | SCORED BY THE OUTSIDE ORACLE private-ref:6a8ea809 (13 boxes, scored by the oracle file's own rule), nx_refstill oracle on bytes identical to the promoted 4ee8bcbc (nx_job_run pid 8035, /tmp/nx_refstill_gr62_oracle1.out): RECALL over 22500 skin-box px, model 831 permil vs YCbCr 938; FALSE ADMISSION over 26400 sand, rock, towel and hair px, model 151 permil vs YCbCr 395. Per box, model vs YCbCr: skin sun 574 vs 1000, 906 vs 922, 1000 vs 986; skin shade 886 vs 892; sand sun 306 vs 995 and 140 vs 988; sand shade 484 vs 689; rock 175 vs 189; towel 0 vs 0; hair 7 vs 0; outside the negative set hill 739 vs 845, vegetation 0 vs 120, sea 0 vs 0. The model cuts false admission 2.6x for 107 permil of recall; what it still admits is shaded sand and the sunlit brown hill. |
| 2026-10-01 | GR62 | lesson | A COLOUR MODEL REMOVES THE DEFECT IT WAS BUILT FOR AND EXPOSES THE NEXT: the two-chromophore mask refuses yellow sand, but a colour-only mask cannot tell skin from a mineral of the same chromaticity -- private-ref:ff3b177c's model skin is 400 permil of the frame at a* 3.66 with an L* spread of 4.4 (grey-beige sand, by its statistics), and reference 1's sunlit brown hill is 739 permil admitted. The CAUSE is UNMEASURED; the leading hypothesis is the model's redness at its low-blood edge (the domain floor is Jacques' typical average 0.2 percent), and the deciding experiment is the model's own L*a*b* along its domain against MEASURED skin (queued) -- not a floor moved until one image scores better, which would be a threshold tuned to flatter one oracle. ALSO: the conf's checkable criterion caught a too-coarse lattice before any photograph was read (41 per axis: floor 200535 e-9 above a mid-grey pixel's 80059 e-9; 81 per axis: 60815 e-9 below), and the gate now holds it as tooth T1b. |
| 2026-10-01 | GR62 | queue | REMAINING DONE-RULE LEG, exact next steps: (1) the beach page -- a new nx_refstill verb that reads private-ref:1d118014 into gr_refstill_referee's band rows (GR_RS_BAND_HAVE, L, a, b of each centre), labels a captured beach frame with rfs_measure_rgb and prints each region's dE00 against its band, then the page emitter's row beside the GR30 world grade, through the beach page's own release lane. (2) BEFORE the skin band is shown as skin: calibrate the skin model against MEASURED skin (mirror a published skin reflectance or colorimetry set, walk the model's domain with cs_de2000 against it, fix the KM conversion or the channel wavelengths it indicts), re-score with the outside oracle, and land a second oracle file for another class reference before any domain change is believed. (3) recalibrate regions.conf against the class: private-ref:80de5b96 reads 470 permil fabric with mean a* -15.1 b* 47.8 (hue about 108 deg), a yellow-green surface the sand box (40 to 100 deg) misses, and 289 permil unmeasured (clipped). (4) mirror van Gemert and Star 1987 for the KM conversion row of skin_model.conf. |
| 2026-10-01 | GR59 | measure | TRIANGULATED BAR as of 2026-10-01 for facultative pigmentation (UV dose to melanogenesis and tan persistence, ITA of sun-exposed versus protected skin): HOW A TAN IS SCORED -- as a difference from a constitutive baseline site: degree-of-tan is the ITA of buttock minus the ITA of the exposed site, measured every 6 to 7 weeks for a year with personal UV dosimeters [schmalwieser_tan_2025], or outer minus inner forearm, whose natural-tan difference averaged Delta E 6.03 (SD 2.64) [amano_naturaltan_2020]. DOSE TO TAN -- under controlled sub-MMD UVB the absolute pigment increase is linear in dose and independent of skin type, repeated exposure lowers the dose to minimal pigmentation by factors of 2 and 3, the protocol takes the peak at one week after the last exposure, and fading takes 4 to 5 months (132 to 175 days to total fading by linear regression on 1 week, 2 and 4 months) [ravnbak_tankinetics_2008]; in outdoor farmers the change in tan regresses linearly on daily mean erythemal dose with the zero crossing at 1.4 SED per day (above it the tan grows, below it fades) and an obvious tan persists through late winter [schmalwieser_tan_2025]; the newest on-subject measurement (August 2026, L'Oreal) adds depth: in vivo multiphoton microscopy measured melanin density and its epidermal depth distribution, and melanin density and skin colour rose from winter to summer under moderate real-life exposure but not under daily SPF and PPD protection [epmc_tanning_202609]. THE COLOUR PATH -- a natural tan lowers lightness without raising yellowness, following Chardon's curved (L*, b*) skin-colour volume with peak yellowness near L* 50, while a straight L*-b* shift reads as unnatural [amano_naturaltan_2020]. THE INSTRUMENT -- ITA is weak outside controlled colorimetry: fixed-threshold ITA from uncontrolled images scored 22.9 percent against a 51.3 percent majority baseline [alipour_ita_limits_2026], and ITA was built for the suntanning pathways of Caucasian skin, so apparent-colour work reports L* and hue angle [beyondskintone_iccv2023]. LISTINGS -- the arXiv skin-tanning search returns 3 hits, none on this axis (one is leather tanning) [arxiv_skintanning_202609], the arXiv cs.GR October 2026 listing reads no updates [csgr202610_gfx_t1001], and the newest on-subject biomedical month is August 2026 [epmc_tanning_202609] ; witnesses [schmalwieser_tan_2025] [ravnbak_tankinetics_2008] [amano_naturaltan_2020] [epmc_tanning_202609] [alipour_ita_limits_2026] [beyondskintone_iccv2023] [arxiv_skintanning_202609] [csgr202610_gfx_t1001] -- independent groups: the Copenhagen Bispebjerg photobiology group with Vienna (Schmalwieser and Ravnbak share Philipsen and count as ONE), Liverpool and Leeds colour science (Amano, Xiao, Wuerger, Meyer), L'Oreal Advanced Research, Technological University Dublin, Sony AI with the University of Tokyo -- five, all in peer-reviewed venues ; WHERE THEY DISAGREE: the controlled study's abstract says fading was unrelated to ethnic origin, yet its own subgroup medians fade faster in Scandinavians (101 and 104 days) than in Indian and Pakistani volunteers (152 and 132 days); the field study finds the 1.4 SED per day crossing only in farmers, with no dose correlation at any site in spouses and children, and finds the supposedly unexposed inner arm also tans and fades; the instruments differ (a reflectance pigmentation value, tristimulus ITA, multiphoton melanin density) and the imaging papers reject ITA from photographs; the defining erythema standard (ISO/CIE 17166) was mirrored but did not extract, so the SED unit is witnessed through the papers that use it, not through the standard ; VERDICT on the running lane (corrected after reading its land row 1790824660, which preceded this row): GR59 landed 2026-09-30 as gr_melanin_field over nx_skin_tan_lib and is AT the triangulated best on controlled tanning kinetics and baseline -- Ravnbak's per-type minimal melanogenic dose in SED, its 2x and 3x lowering by repeated exposure, a peak one week after the last exposure and a linear fade over the published 132 to 175 days, constitutive melanin solved from the genome's ITA, and a Kubelka epidermis over dermis integrated on the CIE 1931, D65 and A tables with a Bradford relight -- and BELOW it on five named axes: the field maintenance threshold (the 1.4 SED per day crossing measured in outdoor farmers, where the lane's kinetics come from controlled sub-MMD UVB), the melanin depth axis, a single fade per type where the source's own subgroups fade faster in Scandinavians, L* and hue reported beside ITA, and a check that the colour moves along the natural curved (L*, b*) path |
| 2026-10-01 | GR62 | measure | TRIANGULATED BAR as of 2026-10-01 for the render-versus-photograph colour referee (skin segmentation in photographs and the colour-difference metric): SURFACE-COLOUR METRIC -- CIEDE2000 (CIE 2001, Technical Committee 1-47, for industrial colour differences) is still the standard formula for small differences between region-mean colours under declared reference conditions, but a correct implementation must pass the Sharma supplementary pairs because the CIE worked examples miss common implementation errors, and the formula is mathematically discontinuous in mean hue [sharma_ciede2000_2005]. DISPLAY METRIC -- for display-referred HDR and wide-gamut signals the standard is ITU-R BT.2124 DeltaE ITP from ICtCp, scaled so 1 is a just noticeable difference, deliberately over-predicting by presuming the most sensitive adaptation, and only ordinal for scene-referred signals without an assumed nominal peak luminance [itu_bt2124_2019]. PHOTOGRAPH METRIC -- the image-quality school finds CIEDE2000 no better than plain CIELAB on photographic pairs, pixelwise CIELAB fragile to slight misregistration, and a learned metric better on photographic colour distortions (STRESS 14.1 against 18.2 on the TID2013 colour subset) though worse than CIEDE2000 on uniform patches [cdflow_cvpr2023]; the newest listing entry (submitted 2026-09-30) is again a data-driven regressor that reports CIEDE2000 inconsistent with judgments, beside a September revision of a luminance-scaled Euclidean line element for surface colours, and October has no entry yet [arxiv_colordiff_202609] [csgr202610_gfx_t1001]. SKIN IN PHOTOGRAPHS -- apparent skin colour is read on extracted facial skin regions and reported as L* and hue angle rather than ITA alone [beyondskintone_iccv2023], because ITA from uncontrolled images is lighting-sensitive and unreliable [alipour_ita_limits_2026]; the closest published precedent to this rung, a photo-to-virtual-human pipeline, scores CIELAB Delta E and ITA under cheek and full-face masks over about 19848 renders and finds consistently higher error for darker skin [truetotone26]; the newest arXiv skin-segmentation entry is SkinMap 2510.05296 (October 2025, full-body skin regions weighted for remote photoplethysmography) and that listing holds no 2026 entry [arxiv_skinseg_202609] ; witnesses [sharma_ciede2000_2005] [itu_bt2124_2019] [cdflow_cvpr2023] [beyondskintone_iccv2023] [alipour_ita_limits_2026] [truetotone26] [arxiv_colordiff_202609] [csgr202610_gfx_t1001] -- six independent groups plus the current listings (Sharma at Rochester; ITU-R Study Group 6; Ma's group at City University of Hong Kong; Sony AI with the University of Tokyo; Technological University Dublin; PUCRS), one standard and four peer-reviewed papers (True to Tone is a preprint) ; WHERE THEY DISAGREE: standards keep handcrafted formulae (CIEDE2000 for surface colours, DeltaE ITP for HDR displays) while the photographic-image school says neither transfers to photographs and 2026 preprints keep proposing replacements no standards body has adopted; the skin-measurement papers reject ITA as the single reported number; and none of the witnesses read today builds a skin mask from a chromophore-fit residual, an absence not proven by exhaustive search (their skin regions come from landmark-based facial extraction or cheek and full-face masks) ; VERDICT on the running lane (updated after its land rows 1790825400 and 1790826001): GR62 landed gr_refstill_referee and the nx_refstill frame verb, which print each region's CIEDE2000 against its band centre beside the class radius through ST1's ruler, with the skin mask taken from the two-chromophore fit residual; it is AT the bar in comparing region statistics rather than pixels with CIEDE2000 under a declared illuminant, and BELOW it in four named places: prove cs_de2000 on the Sharma supplementary pairs (neither board cited them before today), add DeltaE ITP for the display-referred HDR frame, report L* and hue beside ITA, and widen the mask's validation beyond one reference photograph scored against seat-annotated oracle boxes (its land row reports recall and false-admission pairs of 831 vs 938 and 151 vs 395 permil) to a landmark or parsing mask over many photographs, because no witness read today builds a mask from a chromophore residual |
| 2026-10-01 | GR59 | measure | THE NEW TEETH BITE THEIR OWN SUBJECT, PROVEN BY AN AIMED MUTANT (2026-09-30): nx_gate_bite over nx_skin_ita.nx (aim gr_melanin_field) killed its first compiling comparison mutant at byte 6805, which sits in the pre-existing locus code, so it proves the gate non-vacuous and NOT the new block (the aim limit already filed as debt 1790543778); the aimed proof was taken by hand instead: gr_melanin_field's dose line mutated to ignore the coverage history (the garment blocks nothing), built sovereign, run staged: RED 18/25, killed by T12, T12b, T13, T13b, T14, T15 and T16 while the zero-dose, open-fabric and persistence controls stayed GREEN; the source restored byte-identical (sha256 38164447, 19116 B) within the same minute, the scratch rows reaped, and the bite harness's twin custody reported restored=2 mutant_live=0 with live still 5dfb04e0. A rebuild against the CURRENT closure (nx_colorsci_lib 38414 -> 46771 and nx_kubelka_lib 18047 -> 29088 moved under two sibling lanes during this leg) built 28970df2 at the same 127720 B and ran 25/25 GREEN with output byte-identical to the promoted binary's (sha256 92db0439, 4048 B): the sibling changes are behaviour-neutral for GR59 |
| 2026-10-01 | GR62 | measure | THE DECIDING EXPERIMENT, FIRST HALF -- THE MODEL'S OWN COLOURS, MEASURED: nx_refstill model on the promoted binary f03a67db (nx_job_run pid 17063, /tmp/nx_refstill_model_walk_1.out) walks the skin model as absolute reflectance (white diffuser 100000) at eighths of f.mel and at the floor, middle and ceiling of f.blood, 27 rows. At the blood FLOOR (0.2 percent) a* is 3.600 at no melanin, 3.105 at f.mel 5.4 percent (L 77.305 b 11.918 ITA 66.4), 3.575 at 10.75 percent (L 68.274 b 19.840), 4.590 at 16.1 percent, rising to 10.670 only at 43 percent; at the middle of the blood range (2.6 percent) a* is already 19.3 to 21.8 at every melanin; at no melanin b* stays near 0 (1.046, -0.150, 0.942) at every blood level. So the hypothesis in the lesson row is now a MEASUREMENT: the model's low-blood edge for light-to-moderate melanin is a pale beige at a* 3 to 5, exactly where private-ref:ff3b177c's model-admitted sand sits (a* 3.658 b* 9.212), and the model reaches no yellowness without melanin. The second half (the same walk against MEASURED skin colorimetry through cs_de2000) is the queued step; the floor is NOT moved on this evidence alone. |
| 2026-10-01 | GR62 | land | THE NUMBER THE PAGE OWES NOW EXISTS AS A VERB: nx_refstill frame <jpeg> <bands> reads private-ref:1d118014 into gr_refstill_referee's band rows (tc_atoi_signed, one row per region the bands name), labels the frame and prints each region's CIEDE2000 against its band centre beside the class's own radius. Promoted 174656 B sha256 32ac3a1b (contentdiff GREEN, lost_from_live=0, ONE-WAY-GAIN), offc installed; chain this session 4ee8bcbc (measure, class, oracle) then f03a67db (+model) then 32ac3a1b (+frame), the oracle scores reproduced exactly on f03a67db (recall 831 vs 938, false admission 151 vs 395; pid 17145). First frame refereed, reference 1 ITSELF (in-class, so within-class is expected and proves only the plumbing): skin dE00 3.875 vs radius 10.801, sky 4.831 vs 9.848, sea 7.178 vs 7.178 (private-ref:6a8ea809's own sea defines the radius, so the reducer and the referee agree through the one ruler), sand 5.297 vs 11.905, fabric 15.473 vs 26.910; other and unmeasured carry no band (nx_job_run pid 22393, private-ref:6a8ea809). |
| 2026-10-01 | GR62 | queue | SUPERSEDES step (1) of the 1790825404 queue row (the verb now exists): what remains of the done-rule is the PAGE -- capture a beach frame from the live beach page as a JPEG the sovereign decoder reads (baseline or progressive; a PNG needs a decode path the verb does not have), run nx_refstill frame on it against private-ref:1d118014, and have the beach page's emitter print each region's dE00 and within-class beside the GR30 world grade, through the beach page's own release lane, never by this lane editing the page. Steps (2) to (4) of 1790825404 stand: the skin band is NOT shown as skin until the model walk is compared with measured skin colorimetry and the outside oracle re-scores; a second oracle file before any domain change; regions.conf recalibration; the van Gemert and Star mirror. Also owed: the class verb should check each mirror against the sha256 its refs row pins before measuring it. |
| 2026-10-01 | GR59 | land | NO SECOND SPECTRA READER: skintwin ST6 landed km_spectra_load in nx_kubelka_lib DURING this leg (the estate's one pinned-spectra reader: the bytes proven against the refs pin by sha256, a truncated, reordered or unpinned table refused by name), so the CSV parser written beside it this afternoon (tn_isdig, tn_dec_q30, tn_csv_load) was retired the same hour and nx_skin_tan_lib now reads the CIE CMF, D65 and A tables through km_spectra_load and km_spec_interp under the pins of [cie_cmf1931_2deg] [cie_illum_d65] [cie_illum_a] (21525 -> 20399 B). The gate gained T10c, a biting control: illuminant A offered under the D65 pin refuses KM_SPEC_PIN (-3) by name. Rebuilt sovereign 154232 B sha256 4fd921dcd406757ff4ae0da1ea588693db6b8b5d99600fb671652ba8bd9e3ac5, staged 26/26 GREEN with EVERY published value byte-identical to the first build (the parse changed, the numbers did not), contentdiff lost exactly the four retired symbol names (fq_to_int no longer referenced, tn_isdig, tn_dec_q30, tn_csv_load) and gained T10c, promoted under allow_capability_loss with every lost run named, _offc refreshed, LIVE run 26/26 GREEN (nx_job_run pid 13097). NAMED COST: two promotes in one leg rotated the original pre-change binary (35d5fdee, 28666 B, 9 teeth) out of the prev slot, banked by hash only; prev now holds 5dfb04e0, the first GR59 promote, whose 25 teeth include all nine of the original's |
| 2026-10-01 | GR59 | measure | NO CURRENT-MONTH BAR EXISTS FOR THIS AXIS, AND THE READING SAYS SO (2026-09-30): the arXiv abstract search for skin tanning, newest first [arxiv_skintanning_202609], holds three entries and all three are OFF-SUBJECT by subject (a skin-lesion classifier of January 2026, a non-Hermitian skin-effect paper, leather-tanning defect segmentation); the export API answered HTTP 429 eight times the same hour, so the melanin-AND-rendering query is UNATTESTED rather than empty. Nothing the field published renders sun-tanning as a dose-driven absorber field. The bars this rung is held to are therefore measurements, not rival methods: Ravnbak 2008's photobiology [ravnbak_tankinetics_2008] for dose, threshold and persistence, Amano 2020's 60-subject natural-tan colorimetry [amano_naturaltan_2020] for the band, and the vendor bar this board already carries for painted skin (MetaHuman 5.8 SB-MHSKIN-2609, palm and sole tone applied automatically, no exposure history). Against that vendor bar the estate's line is DERIVED, relit by re-integration, and persistent by a measured fade; parity on absolute skin colour is NOT claimed while the dermis is bloodless |
| 2026-10-01 | GR59 | lesson | TWO INDEPENDENT MEASUREMENTS AGREED THROUGH ONE MODEL WITH NO KNOB BETWEEN THEM, BECAUSE THE TAN WAS MADE RELATIVE: the facultative melanin is the constitutive melanin scaled by Ravnbak's increase over pre-exposure pigmentation, and the constitutive melanin is SOLVED against the genome's ITA, so the two-flux convention (K = 2 mua, S = musp, against the diffusion-matched 3/4) cancels out of the line instead of being tuned into it. What fell out: the solved constitutive melanin, 5.54 percent, inside Jacques's published 1.3-6.3 percent for light skin, and the tanned side's ITA 38.42 against Amano's independently measured outer-forearm mean 38.8. A SECOND LESSON, PAID FOR TWICE THE SAME DAY: a sibling lane mirrored the SAME Jacques page and CIE tables under its own keys while this one did, and a skintwin.refs append created a duplicate key that only a find caught (removed within minutes, the sibling's keys cited instead); and the reader this lane wrote landed beside the one that lane was writing. Before appending a ref or writing a parser, search the board's refs by PIN and the libs by CAPABILITY, not by the name you were about to use |
| 2026-10-01 | GR59 | queue | NEXT STEPS, EACH WITH ITS RECIPE: (1) HAEMOGLOBIN IN THE DERMIS, the named fidelity gap: the inside skin reads a* 3.0 where measured skin reads near 10, and the line raises b* by 2.04 where Amano's natural tans raise none [amano_naturaltan_2020]; add mua.blood = ln10 x epsilon x 150 g/L / 64500 g/mol (the OMLC conversion) at Jacques's f.blood 0.2 percent [jacques_skinoptics_1998] to tn_reflect's dermis, read through km_spectra_load from a mirrored, pinned Prahl TABLE (the omlc_hb row mirrors only the summary page) with an oxygen saturation taken from a pinned source, then re-measure a*, b*, DE2000 and the band teeth -- the constitutive melanin is re-solved against the genome automatically. (2) THE LIVE FRAME: the field is gated, the beach frame does not read it yet; feed gr_melanin_field from the garment lane's per-vertex coverage and bake tn_reflect through the beauty tier, so GR55's per-class ITA error is measured on a tanned body. (3) POPULATION: one genome and one scenario are measured; sweep the six ITA classes (Ravnbak's absolute increase is skin-type independent, so the relative facultative fraction must fall as constitutive melanin rises) and publish dITA per class beside GR55's medians. (4) re-attest the arXiv melanin-AND-rendering listing once the export API stops answering 429. (5) the watch cell _ABSENT_:gr_melanin_field should read LANDED after the orchestrator's next regen of /compare/graphics; check the cell, not this row |
| 2026-10-01 | GR59 | measure | THE CHAIN LIB IS BITE-PROVEN AND THE SCRATCH IS DISPOSED (2026-09-30): nx_gate_bite nx_skin_ita_gate over nx_skin_tan_lib.nx killed its first compiling comparison mutant at byte 5800 (inside the new chain lib), valid_mutants=1 killed=1 not_reached=0, verdict GREEN, the lib restored byte-identical (20399 B, sha256 3b35fb18) and both twins restored (custody twins_checked=2 restored=2 mutant_live=0, live still 4fd921dc). Final sources: runtime/nx_skin_ita.nx 19116 B sha256 38164447 (the gr_melanin_field contract), runtime/nx_skin_tan_lib.nx 20399 B sha256 3b35fb18, runtime/nx_skin_ita_gate.nx 21461 B sha256 658b8d20, knowledge/skin_tan.conf 3529 B. The four scratch binaries (two staged candidates, the current-closure rebuild and the coverage-blind mutant) were retired reversibly into knowledge/retired with nx_retire_path, and every scratch registry row (nx_skin_ita_gate_gr59, gr59m, gr59c, gr59p _staged) was reaped; the first GR59 promote 5dfb04e0 survives both as prev and as knowledge/retired/nx_skin_ita_gate_gr59.sov.elf.new-1790826503 |
| 2026-10-01 | GR10 | measure | OCTOBER ATTESTATION 2026-10-01: nx_barfresh graphics verdict FRESH, 7 of 7 bars ATTESTED (it read STALE with 7 UNATTESTED before this seat looked), every listing read by the seat; the morning's organ attests could not reach arXiv and export.arxiv.org answered this hour only after 429 retries. Literal readings: SB-RTGI 2026-09 (its own venue-tag query, totalResults 1, Gaussian Light Transport 2609.11430 of 2026-09-10), SB-HUMAN 2026-09 (GenStream 2609.18634 arXiv-posted 2026-09-16, originally ACM MM 2025; on-subject September entries such as Heartian 2609.28539 of 2026-09-22), SB-ADVANCES-2607 2026-07 (course page byte-identical to its 2026-09-27 pin), SB-VENUES-2609 2026-09 (the SIGGRAPH Asia 2026 venue-tag listing, 59 entries, 29 in September, newest 2609.37167 of 2026-09-29). Two refs RE-POINTED with the reason in each row: SB-VENUES-2609 from kesen_sig2026, a static July index that can never carry a September month, to arxiv_co_siga2026, whose entries set it; metahuman from https://www.unrealengine.com/en-US/metahuman, which answered 403, to the MetaHuman 5.8 release notes the estate can read. Reading-month convention, as on this morning's vendor bars on aesthetictwin, skintwin and anatomy: infinigen-princeton (newest dated entry 2026-08, the 2.0.0a2 release), metahuman (5.8 of 2026-06-17) and siggraph-rtr (the July 2026 course) carry newest=2026-09 meaning nothing newer than the September reading, the true newest dated entry named in each barscan (queue row 1790828670). RANK by nx_compare_rank (artefact graphics.rank 24361 B sha256 f84a967e870a085f793abf4993c286da82a3dc9e642b4c1fbeca6576e50ed8c9, target 0.1 milestone M1, 63 rungs, done 13, open 50, finish 3, unmapped 7, PM filed 1 of the top 3): the critical path is one rung, #1 GR2 the genome splat avatar (gs_avatar_genome, priority 1000, DARK-CONTRACT); staged next GR27 ground material 3800, GR14 linear-light pipeline 3000, GR42 the beauty tier's first frame 2025, GR15 sovereign strand system 2000, GR61 camera model as data 1000; the reference-still human rungs rank low on value alone (GR60 support surface 125, GR63 real-versus-render bench 100, GR57 sheer fabric over skin 75, GR58 wet state 66), and every open rung is TREATMENT-UNDECLARED with sponsor=none. |
| 2026-10-01 | GR10 | queue | READING-MONTH CONVENTION FOR VENDOR AND REFERENCE BARS NEEDS A DECISION (an estate-wide ruler question for the nx_barfresh owner, raised by the seat that applied it): the barscan grammar says newest is the newest entry's month and nx_barfresh attests only when it equals the bar's month, yet the vendor and reference bars (here infinigen-princeton, metahuman, siggraph-rtr; on aesthetictwin VISIA, 3dMD, Artec and others; on skintwin and anatomy the MetaHuman, UE5 and DARPA rows) were dated by the month they were read, so an honest literal barscan (2026-08, 2026-06, 2026-07 here) reads UNATTESTED, and the only literal fix is re-dating each bar to its newest dated entry, as this seat did for aesthetictwin MB8 and SB-FACEANAT-2609. Both readings now stand on the boards. Decide one: re-date every vendor and reference bar to its newest dated entry (strict, the 2026-09-06 aspect-sentiment precedent), or give barscan an explicit baseline-unchanged kind that nx_barfresh accepts and prints as its own state, so a reading-month bar is never mistaken for a newest-entry bar. Until decided, each convention barscan names the true newest dated entry in its note. |
| 2026-09-30 | GR59 | measure | ADVERSARIAL VERIFIER, method class 3 (independent hand re-derivation, not the gate's arithmetic) over a fresh run of the promoted nx_skin_ita_gate 26/26 GREEN on the NAS (job /tmp/verify_gr59_ita.out). ITA=atan((L-50)/b) by hand from the printed Lab: covered L 72.748 b 19.180 -> 49.864 deg vs printed 49.863; tanned L 66.828 b 21.218 -> 38.418 vs 38.418; dITA 11.446 vs 11.445 (inside the 4.599..13.585 band); the 1 mdeg gap is inside the milli quantisation of the printed Lab (about 1.2 mdeg). CIE76 from the same Lab 6.519 vs printed 651 centi (the gate truncates, 603 plus or minus 264 band 339..867 holds). Fade from skin_tan.conf: last exposure day 19 + Ravnbak peak 7 = 26 vs 26; 19 + Ravnbak 175 = 194 vs 194; midpoint (26+194)/2 = 110 vs 110 at 21774/2 = 10887 ppm vs 10887 (the half-at-midpoint result holds only because the decay between peak and zero is linear, a model choice, Ravnbak publishes only the total fading time). Bare melanin: Jacques 6.6e11 nm^-3.33 at 400 and 700 nm = 1427.9 and 221.5 cm-1, ln ratio/ln(700/400) = 3.330 vs printed 3.329 (the printed value is the conf exponent echoed through Q30, not a fit). NOT REPRODUCED: the line OD exponent 2.013 cannot be derived from Jacques law at two wavelengths, which gives 3.33 for any Beer-Lambert double pass; 2.013 is a least-squares fit over the diffusion-model reflectance (tn_od_exponent over tn_reflect), plausible for a diffusion regime between sqrt(mua) and linear but unverified by this class, owed a re-derivation through an independent reflectance model. Verdict PARTIAL: ITA, dITA, DE76 and the fade schedule CONFIRMED by hand, line OD exponent UNREPRODUCED, nothing changed. |
| 2026-10-01 | GR62 | land | THE CIEDE2000 RULER IS NOW PROVEN PAIR BY PAIR ON THE FIRST AUTHOR'S OWN DATA FILE: ciede2000testdata.txt mirrored from Sharma's Rochester page through nx_research_fetch (knowledge/fetched/cmp_graphics_sharma_ciede2000_testdata.txt, 1830 B, sha256 44aebb39, refs row sharma_ciede2000_testdata, class dataset), 34 rows with the signs printed. nx_colorsci_gate T2b adds one gv_check_near tooth per pair (actual, expected, delta and tol emitted in 1e-6 units; tolerance 60e-6 = half an ulp of the four-decimal table, 50e-6, DERIVED by cg_half_ulp_micro, plus the incumbent Q30 chain headroom 10e-6, both written into every tooth name from the constants), tags the paper's own arctangent and mean-hue cases (pairs 7-16, read 2026-10-01 from the mirrored PDF through nx_pdf_text), and COMPUTES which pairs straddle the 0/360 cut from the lib's own G and cs_hue_deg under the lib's knife-edge epsilon (7 pairs: 11 12 15 17 18 19 20; pairs 10 and 14 tagged ON-THE-180-DEGREE-KNIFE-EDGE). T2c: the KAT table whose signs were INFERRED from the PDF hue column equals the data file on 238 of 238 L a b dE fields. Planted fixture: the naive mean hue without the 360 wrap is refused against the published h_bar_prime on all 7 boundary pairs while the lib's own is accepted. Staged build da5969fe (127128 B; nx_job_run pid 1676, /tmp/gr62_colorsci_gate_staged_1.out) passed 64/64 GREEN, worst pair 23 at 49e-6 (inside the half-ulp alone); nx_contentdiff live dc7b9777 (80042 B) vs candidate GREEN lost_from_live=0 ONE-WAY-GAIN; PROMOTED nx_colorsci_gate sha256 da5969fe09846ed2f35b2c9855a7a68dfc8e325e05a49eeaea3051b5298c288b with prev banked; offc installed. cs_de2000 itself is UNCHANGED (no defect found), so no skintwin ST1 pointer row is owed. |
| 2026-10-01 | GR62 | measure | BITE-PROVEN BY A REAL MUTATION OF cs_de2000 HUE HANDLING, ISOLATED FROM EVERY OTHER LANE: the lib and the gate were copied to private module names (nx_colorsci_lib_gr62bite, nx_colorsci_gate_gr62bite, importing only each other) and the wrap branch of the mean hue was replaced by the paper's named mistake, the naive (h1'+h2')/2; built 110435 B and run (nx_job_run pid 3758, /tmp/gr62_colorsci_gate_bite_mutant_1.out): 54/64 RED, and the ten failures are exactly the seven boundary-pair teeth (deltas from 23751e-6 on pair 20 to 4784644e-6 on pair 19; the mutant gives pair 11 7.179153 against the published 7.2195), the two T2 aggregates (dE00 within half an ulp, h_bar_prime) and the naive-mean fixture (FALSE-POSITIVE on the good input); all 27 non-boundary pair teeth stayed PASS, so the new teeth discriminate this bug class and nothing else. The shared lib was never mutated, so no restore was needed; the pristine staged run above is the GREEN half. Scratch retired reversibly into knowledge/retired (both sources and the staged elf); registry rows reaped (nx_toolreg_reap reaped=2); the _build fossil of the mutant is refused by nx_retire_path as a protected .elf name and remains an unregistered, sourceless build artifact. |
| 2026-10-01 | GR62 | lesson | A ZERO HUE IS NOT ALWAYS ZERO IN Q30: for the (2.5, 0) sample the lib's cs_hue_deg returns 359.999999 on three of the five pairs that carry it (16, 18, 20) and exactly 0 on the other two (17, 19), because CORDIC atan2(0, +a') leaves a one-micro-degree residual that depends on a' = 2.5(1+G). The mean-hue branch therefore takes the wrap on 18 and 20 where exact arithmetic would not, and skips it on 16 where exact arithmetic would take it, and all three still land on the published h_bar_prime and dE00 because the hue circle makes the two branches agree. dE00 is unaffected (34 of 34 within half an ulp); a reported hue of a pure +a* colour can print 359.999 instead of 0. A boundary class derived from a fixed-point hue is a property of the representation, so the tooth tags say which branch the LIB took, never which branch exact arithmetic would take. |
| 2026-10-01 | GR62 | measure | THE FIRST REAL EXPERIMENTS ON THIS BOARD, AS DATA (ecosystem EC86 renders them on every page; both PRIVATE: the subject is a real person, so the page carries their numbers and never an image). E1 (ab, mask YCbCr control against the two-chromophore skin model, the oracle boxes of private-ref:6a8ea809 as the unit of replication), re-measured tonight by nx_refstill oracle on promoted 32ac3a1b (direct call, verdict=ORACLE-SCORED, the same pixel totals as the land row: recall 938 vs 831 and false admission 395 vs 151 permil, pixel-weighted): per skin box YCbCr 1000, 922, 986, 892 against the model 574, 906, 1000, 886; per negative box (sand sun left strip, sand sun mid beach, sand shade, towel, rock, hair) YCbCr 995, 988, 689, 0, 189, 0 against the model 306, 140, 484, 0, 175, 7. Four skin boxes and six negative boxes are far under the incumbent sample floor, so the page reads INSUFFICIENT with that reason rather than calling the model better on recall or worse on false admission: the box means favour the model on false admission (the sand boxes) and YCbCr on recall (the sunlit thigh), and only more oracle boxes or more references decide it. E2 (factorial, mask x reference over the four admitted references private-ref:6a8ea809, private-ref:06caaa3c, private-ref:ff3b177c, private-ref:80de5b96), re-measured tonight by nx_refstill measure (nx_job_run pids 5224, 5795, 5994, 6105; /tmp/exl_measure_*.out): skin share permil and the admitted skin's mean L*, a*, b* under each mask; one frame per cell, so the mask main effect rests on four frames a side and reads INSUFFICIENT, while the effect sizes are on the page. The YCbCr side is the whole-frame rule (at23_ycbcr_whole_frame), the model side the region=skin row. |
| 2026-10-01 | GR42 | measure | THE ReSTIR FAMILY BEYOND THE 2020 PAPER, MIRRORED AND READ 2026-10-01 from NVIDIA Research's open copies: ReSTIR GI, HPG 2021, five NVIDIA authors: resampling of multi-bounce indirect paths, MSE 9.3x to 166x better than path tracing at one sample per pixel, 8.9 ms against 8.0 ms equal-time at 1080p on an RTX 3090 with MSE improved 15 to 1 [restirgi2021_bf1001]; GRIS and ReSTIR PT, TOG 41(4) 2022, Utah and NVIDIA: variance bounds and convergence conditions for reuse of correlated samples with unknown PDFs, Fig. 2 shows unbiased ReSTIR reuse converging to a wrong result, ReSTIR PT shifts whole paths between pixels [gris2022_bf1001]; Area ReSTIR, DOI 10.1145/3658210, Utah and NVIDIA: reservoirs over each pixel's 4D film and lens ray space for antialiasing and defocus, Fig. 1 times 4.2 ms path tracing, 9.5 ms ReSTIR PT and 14.0 ms Area ReSTIR, its body text not recovered by the extractor [arearestir2024_bf1001]; with Bitterli et al. 2020 [bitterli2020] the family is now read as four methods. COUNTED BY GROUP it is still ONE lineage: every paper has NVIDIA authors and Kettunen, Wyman, Lin or Bitterli recur across them (NVIDIA with Dartmouth in 2020 and with Utah in 2022 and 2024), so these fetches add method witnesses, not independent groups, and the GR42 bar's group independence remains unmet |
| 2026-10-01 | GR42 | queue | a ReSTIR witness from a group with no NVIDIA author is still owed (a candidate lead named from memory and unverified: AMD's world-space spatiotemporal reservoir reuse for ray-traced global illumination, 2021); the newest ReSTIR entry on the board's listing is ToF ReSTIR (May 2026, DOI 10.1145/3811299), whose authors were not read [restir_search_t1001] |
| 2026-10-01 | GR62 | queue | DeltaE ITP FOR DISPLAY-REFERRED HDR FRAMES IS NOT BUILT, AND WHY: the standard is in hand (ITU-R BT.2124-0 (01/2019), mirrored 2026-10-01 at knowledge/fetched/cmp_graphics_itu_bt2124_2019.pdf, pinned, refs row on GR62) but no second, INDEPENDENT witness that applies it is mirrored, and the triangulation rule refuses a metric that stands on one document. Exact next steps: (1) mirror through nx_research_fetch and read through nx_pdf_text one peer-reviewed paper, from a group other than the standard's proponents, that evaluates DeltaE ITP on HDR or wide-gamut imagery, and pin a refs row naming the claim it backs; (2) build cs_de_itp in nx_colorsci_lib from the equations read off the mirrored BT.2124 and BT.2100 bytes (absolute linear BT.2020 light, the PQ inverse EOTF, ICtCp, the BT.2124 scaling), its constants as named UNIT rows, never typed from memory; (3) gate it in nx_colorsci_gate with gv_check_near teeth against any worked values the standard prints plus an independent re-derivation of at least two pairs (one at SDR white, one at a highlight), tolerances derived from the printed digits; (4) only then give nx_refstill an HDR frame verb. |
| 2026-10-01 | GR23 | queue | CLAIMED by seat colour-pipeline (cross-board program step 3 toward gameengine GE83): mirror the AgX reference as an OUTSIDE oracle, then build the transfer curve and the LUT generator in NishiLang as a lib plus a gate whose teeth read oracle values from the mirrored bytes; in flight, nothing shipped yet |
| 2026-10-01 | GR14 | queue | CLAIMED by seat colour-pipeline (cross-board program step 4 toward gameengine GE83): organ-level proof only -- sRGB encode and decode known answers incl. the linear-segment knee, a compositing case where blending in linear light and blending encoded values differ by a computed amount, and a neg-control tooth that catches a double display-encode; the page-level A/B screenshot referee in the done-rule needs a browser capture this headless seat does not have and is NOT claimed here |
| 2026-10-01 | GR59 | measure | BACKFILL-HUMAN (c) tanning kinetics: THREE independent groups now counted, the two new ones read in full text: Copenhagen (Ravnbak 2008 and Schmalwieser 2025, one cluster through Philipsen, existing rows); Fudan University, Shanghai (Ma 2024 Phenomics, refs gr59_ma_fudan_2024_uvtime, mirror knowledge/fetched/cmp_graphics_bfh_ma2024_hanchinese_tanning.xml sha256 bf16a7b5: six UV doses on 31 backs, erythema index peaks on D1 and melanin index on D7 regardless of dose over 14 days, confirmed in a 664-person cohort); NCI-NIH with FDA CDRH (Coelho 2009, refs gr59_coelho_nihfda_2009_llp, mirror knowledge/fetched/cmp_graphics_bfh_coelho2009_longlasting_pigm.xml sha256 24e45089: pigmentation robust at day 16 and visible at 9 months or more in 32 percent after 2 MED and 61 percent after 4 MED). Independent count 3. SED: was 0 witnesses, now 1 primary: the CIE's own International Lighting Vocabulary term 17-26-069 (1 SED is an erythemal radiant exposure of 100 J per square metre, referring to ISO/CIE 17166; refs gr59_cie_eilv_17_26_069_sed, sha256 5e809d7c), with the CIE and ISO catalogue pages for ISO/CIE 17166:2019 (refs gr59_cie_pub_17166_2019, sha256 a7f03424; ISO page knowledge/fetched/cmp_graphics_bfh_iso_17166_page.html). Still open: the 12-page standard text with the s_er(lambda) table is sold (CIE Webshop), not mirrored; a first CIE URL variant answered 403. Search listings: cmp_graphics_bfh_epmc_tankin_search.json and cmp_graphics_bfh_epmc_tankin_search2.json |
| 2026-10-01 | GR42 | measure | THE FIRST ReSTIR WITNESS FROM OUTSIDE THE NVIDIA LINEAGE, MIRRORED AND READ 2026-10-01 (the 1790830011 queue row's lead, now verified): Boisse, World-Space Spatiotemporal Reservoir Reuse for Ray-Traced Global Illumination, SIGGRAPH Asia 2021 Technical Communications, PDF from AMD's GPUOpen site, a single author on none of the four NVIDIA-lineage papers: reservoirs of path vertices cached in a GPU hash grid and reused across space and time, so light can be sampled at any vertex of the eye path rather than only where screen-space reuse reaches [worldspace_restir_amd2021_bf1001]. Counted by group the ReSTIR bar now stands on two lineages, NVIDIA with Dartmouth and Utah [bitterli2020] [restirgi2021_bf1001] [gris2022_bf1001] [arearestir2024_bf1001] and the GPUOpen author, both peer reviewed; a third independent group, and a 2026 entry of the family read in full, are still owed |
| 2026-10-01 | GR59 | land | BACKFILL-HUMAN refs rows landed this lane in graphics.refs, each mirror pinned by nx_filehash, cited here so the refs referee's inline-citation tooth sees them: [gr59_ma_fudan_2024_uvtime] [gr59_coelho_nihfda_2009_llp] [gr59_cie_eilv_17_26_069_sed] [gr59_cie_pub_17166_2019] |
| 2026-10-01 | GR62 | land | L*, C*ab AND HUE ANGLE NOW REPORTED BESIDE ITA PER REGION, LIVE: the promoted nx_refstill (sha256 95c9d457b96e9c03e1071c904cf8f89d841f7c6710b87e4aeae89748a4b8a71b, 175448 B; serving root, _offc twin and buildroot/_build fossil byte-identical by nx_filehash cmp; supersedes 32ac3a1b) appends Cab_e3 and hab_e3 at the END of every measured region row beside the incumbent L_e3, a_e3, b_e3 and ita_* keys (rfc_lch_kv, composed from fq_hypot and cs_hue_deg, the same arithmetic cs_de2000 uses, so no second colour formula exists; every earlier key keeps its name and order). Witness run on the admitted private-ref:6a8ea809 reference (nx_job_run pid 18426, /tmp/gr62_lch_live_verify_1.out, verdict=MEASURED, partition=SUMS): skin L_e3=49465 a_e3=8086 b_e3=10735 Cab_e3=13440 hab_e3=53012; sky Cab_e3=24934 hab_e3=271039; sand Cab_e3=15165 hab_e3=88254. Hand re-derivation from the printed a and b agrees to the last printed digit (skin sqrt(8.086^2+10.735^2)=13.440, atan2(10.735,8.086)=53.01 deg; sky atan2(-24.930,0.452)=271.04 deg). |
| 2026-10-01 | GR62 | queue | OPEN, journaled at the lane deadline (NAS clock past 1790832600, so no new build was started): (1) nx_refstill_gate (promoted 110897 B) carries NO tooth on the new Cab_e3 and hab_e3 keys (nx_txtscan find hab in its source: total=0), so the additive fields are proven by the live witness run and a hand re-derivation only, not by a gate; owed: gv_check_near teeth on a synthetic region with known a and b (one in each hue quadrant, one at a=b=0 asserting h=0 per the CIE convention, one pure +a* row asserting the Q30 359.999999 residual named in the 1790829802 lesson), plus a contract tooth that every pre-existing key precedes Cab_e3 in the row; (2) DeltaE ITP stays queued exactly as row 1790830110 states (BT.2124 mirrored, the independent applying witness not); (3) nx_refstill_gate is REGISTERED-UNAUTHORISED per nx_catalog (no cap names it). |
| 2026-10-01 | GR30 | queue | CLAIMED by seat world-referee (second launch, T0 1790856829). State read from nx_catalog: nx_world_quality_grader built and promoted 32285 B and registered, its only live caller nx_worldgrade promoted 91681 B, its gate nx_world_quality_grader_gate built 43338 B and never promoted, contract symbol wqg_build_wired not declared in the organ. Plan: record real-world grades through nx_worldgrade, declare wqg_build_wired as the guarded entry that reports can-see, wire nx_worldgrade through it, extend the gate with a real-engine-world tooth, a planted stair-step control, determinism and emitted values, then promote and register the gate. The grader is a heightmap statistics ruler and not an image referee |
| 2026-10-01 | GR62 | land | THE FIRST REAL EXPERIMENTS ARE PUBLISHED ON THIS BOARD (ecosystem EC86): scoped regen job 1790858482 verdict=GREEN, FORK-FRESH matrix e99dd74f, index.html 1365034 B (prev 1169619), section anchor #experiments with #exp-GR62-E1 and #exp-GR62-E2, both privacy private, no image rendered for either; nx_page_verify job 1790858564 VERDICT=GREEN (page 200, 5 of 5 declared assets PNG DECODE-OK, all of them estate world renders, a11y-issues=0). WHAT THE PAGE HONESTLY SAYS: E1 per box (unweighted, unit of replication = an oracle box) recall YCbCr 950.0 against skin model 841.5 permil, n=4 per arm, and false admission YCbCr 476.8 against 185.3 permil, n=6 per arm; every corrected verdict reads INSUFFICIENT because the incumbent sample floor is 30 units per arm and one photograph's oracle file carries 4 skin and 6 negative boxes. E2 is a 2 by 4 factorial with one frame per cell, so its 16-contrast family reads INSUFFICIENT throughout and the factor-effects table prints the mask main effect as a difference of marginal means, descriptive. The pixel-weighted totals of the 1790829840 row (938 vs 831, 395 vs 151) are a different aggregation of the same boxes and are not what the table prints. |
| 2026-10-03 | GR62 | land | THE PRIVATE REFERENCE STORE IS PARTITIONED BY LICENCE AREA, AS DATA (operator 2026-10-01: "we want to make sure that we actually keep media sources in their proper license areas not just reference things"). private-ref:5c6374be (3135 B): ordered rule rows (contract, NC, ND, SA, BY, CC0, PDM, Public; restricting tokens first, NO MATCH = reserved) and seven area rows each carrying its obligations as key=value (publish, attribution, sharealike, commercial, derivative): pd-cc0, cc-by, cc-by-sa, cc-nd, cc-nc, reserved, contracted. Each area is a directory under private-ref:a74101e4; the two admitted images were copied into their areas byte-verified (private-ref:6a8ea809 186403 B, private-ref:052c57ab 385722 B) and the flat copies retired reversibly to knowledge/retired/ (rename only). private-ref:e051925b (3539 B) gained an 18th field, area, and its header now admits every licence class each in its own area. New organs, all PROMOTED and REGISTERED: nx_refarea_lib (rfa_classify, rfa_check: ALLOW or a NAMED refusal REFUSE-PRIVATE, -RESERVED, -NONCOMMERCIAL, -NO-DERIVATIVES, -NO-ATTRIBUTION, -SHAREALIKE, -NO-GRANT, -UNKNOWN-AREA, -UNKNOWN-USE, -NO-ROW; privacy decided before the licence, every unreadable input fails closed), CLI nx_refarea (classify, check, audit; promoted e31de083, 33037 B), gate nx_refarea_gate (promoted cd19b346, 206996 B, _offc installed, row in knowledge/compare/graphics.gates) |
| 2026-10-03 | GR62 | measure | THREE METHOD CLASSES on the licence-area and privacy guarantees: (1) nx_refarea_gate 40/40 GREEN on the staged and again on the promoted binary, every refusal tooth printing the outcome NAME beside the expected one and paired with an allowed positive control (pd publish without attribution, cc-by publish and commercial with attribution, nc publish non-commercially, sa adaptation under the same licence), plus neg-controls (a token must be WHOLE: BYTE is not BY; an unreadable conf classifies everything reserved and refuses every public use); the publisher predicate rp_private_in (nx_swcompare_lib, in front of rp_publish) refuses a store path, a ../ spelling and a manifest-named also_at copy while an unnamed mirror and a lookalike SUBSTRING of a named path stay publishable; (2) bite-proven: nx_gate_bite nx_refarea_gate nx_refarea_lib.nx compiled a comparison mutant and the gate went RED (killed=1, restore byte-identical 14006 B); (3) independent live witnesses: nx_refarea audit reads rows=2 clean=2 (area equals the classification of each row's own licence words, privacy private, filed in the area directory, stored bytes equal declared) and nx_refarea check private-ref:6a8ea809 publish attrib=yes reads REFUSE-PRIVATE. The publisher refusal is LIVE on both generators: the promoted nx_swcompare_matrix (cb4a476f, 1441520 B) and nx_swcompare_sota (e4b5ae25, 812896 B) each carry the private-ref:e10e037d refusal (nx_txtscan count 2) and are byte-identical to their buildroot/_offc forks (nx_filehash cmp identical=1); nx_absent over buildroot/knowledge/compare *.refs finds no row naming private-ref:e10e037d, private-ref:6a8ea809 or private-ref:052c57ab (ABSENT-PROVEN, corpus_complete=1) |
| 2026-10-03 | GR62 | measure | EXPERIMENT GR62-E3 LANDED AS DATA, PRIVATE, NUMBERS ONLY: a 2 by 2 factorial (mask YCbCr vs skin model, by ref private-ref:6a8ea809 vs private-ref:052c57ab, both cited by private-store id, no path and no image on the page), nx_refstill measure on the area-filed copies. Skin share permil: private-ref:6a8ea809 518 vs 369, private-ref:052c57ab 905 vs 906; mean skin L*/a*/b* (thousandths): private-ref:6a8ea809 YCbCr 62014/4075/12912, model 49465/8086/10735; private-ref:052c57ab YCbCr 70581/9915/11250, model 67828/9745/11415. Re-derivation by hand from the instrument's own pixel counts: 291780/790102 = 369.3, 3915579/4320000 = 906.4, 3911529/4320000 = 905.4 permil; the private-ref:6a8ea809 model row reproduces E2 and the 1790856400 land row to the last digit. What it says: on a studio frame with no sand the two masks agree to 1 permil, so E1's large false-admission gap is a property of sand, not of skin; one frame per cell, so every contrast reads INSUFFICIENT and the result is descriptive. The model labels 68 permil of the studio ground as sand (L* 79.8, b* 4.7): the region vocabulary has no studio-ground class |
| 2026-10-03 | GR62 | lesson | A RETIRED COPY LEAVES THE PRIVACY FENCE: nx_retire_path renames into knowledge/retired/, a path carrying no private-ref:e10e037d word and named by no manifest field, so the two retired flat copies would be publishable by rp_publish if a refs row ever named them. No row does (ABSENT-PROVEN above), but the fence is by path, not by content: a retired private image must keep its name inside the store, or the manifest must list retired copies |
| 2026-10-03 | GR62 | queue | OWED, exact: (1) rp_private_in should also refuse private-ref:abf0ed69 (the retire rename keeps the word refstill inside the leaf, so extending the word rule to match refstill_ as well as private-ref:e10e037d closes it with one tooth); (2) the other three beach references of E2 (private-ref:06caaa3c, private-ref:ff3b177c, private-ref:80de5b96, stored in knowledge/fetched/ and named in private-ref:d9101c46) are real-person photographs outside the private manifest: register each with its area and also_at so the publisher fence covers them; (3) rp_publish_gated could compose rfa_check so a public refs mirror is refused by licence area too, once refs rows carry a licence word (today the class field is a source-type vocabulary); (4) a studio-ground region class for nx_refstill so a light backdrop is not counted as sand |
| 2026-10-03 | GR62 | land | PUBLISHED: scoped regen job 1791041959 verdict=GREEN, FORK-FRESH matrix cb4a476f, graphics/index.html 1390566 B (prev 1378701) carrying #exp-GR62-E3, privacy private, no image for it; nx_page_verify job 1791042000 VERDICT=GREEN (page 200, img=5 all estate world renders PNG DECODE-OK, a11y-issues=0). Every .jpg string on the page is a filename inside journal prose, none is an img or a link (nx_txtscan find .jpg: 12 hits, all prose) |
| 2026-10-03 | GR62 | land | THE PRIVATE-STORE TEXT LEAK IS CLOSED IN THE EMITTER, ONE RULE FOR BOTH GENERATORS (independent verifier 2026-10-03: the store was never served as an image, but plan journal text rendered verbatim, 34 store-path occurrences on this page and the manifest path on aesthetictwin). New leaf lib buildroot/runtime/nx_privref_lib.nx (prvr_redact, prvr_redact_in, prvr_token_for): a span around any case-insensitive match of the store word, a manifest row id, a former id, or the STEM of a stored path, source url, description page, meta mirror or also_at copy is replaced by private-ref:<8 hex of the row id> (a store-word span no row claims is hashed from its own text; a query url contributes its whole basename so the word index is never a needle; - and none are never needles); no match returns the same pointer. Wired at the three text funnels: wnote and wj in nx_swcompare_lib (every note, plan, log and api.json string) and exl_esc in nx_experiment_lib (every experiment cell, redacted before escaping). rp_private_in now reads also_at as a comma LIST (rp_priv_field_find). Store items carry OPAQUE ids rs-<first 8 hex of the sha256 pin> (private-ref:052c57ab, private-ref:6a8ea809, private-ref:06caaa3c, private-ref:ff3b177c, private-ref:80de5b96), the manifest is the only id-to-file map and gained field 18 former (the old stems), E1/E2/E3 exparm rows cite only the opaque ids. The three E2 beach stills are registered (licences read from the mirrored Commons extmetadata: CC BY 2.0, CC BY-SA 3.0, CC BY 2.5, Restrictions empty), copied into their areas byte-verified against their pins, and every retired flat copy is an also_at item. licence_areas.conf: rule NC+ND -> new area cc-nc-nd (commercial=no, derivative=no) ordered ahead of NC, nx_refarea_lib rfa_has_all (a + token fires only when every part is a whole token); before this CC BY-NC-ND landed in cc-nc with derivative=yes and the ND obligation was lost. |
| 2026-10-03 | GR62 | land | THE PRIVACY FILTER NOW SITS AT THE PUBLISH CHOKEPOINT, NOT ONLY AT TWO FUNNELS (re-measured after the 1791044400 land): wnote and wj were filtered but the ladder writer sg_text was not, so /compare/aesthetictwin still printed the manifest path inside the AT57 acceptance text (the one store-path occurrence left on either page; a per-char leaderboard field writer bypassed both funnels too). Fix in the path: nx_compare_regen rg_emit_receipted runs every captured byte of every generator (index.html and api.json, every domain) through nx_privref_lib prvr_redact_n before the receipted write and prints private_spans=N on each published line. nx_privref_lib gained the bytes form (prvr_redact_core, prvr_redact_n, prvr_free: exactly sized output, no NUL needed), the retired-copy word private-ref:83b549b7 with or without a manifest (the retire rename keeps the store word in the leaf), and backslash and both parentheses as span delimiters (a JSON escape is never eaten into a token; the prose bracket survives). Promoted with every LOST line read: nx_compare_regen 726f4b57 389736 B (LOST 2 = the old sys_mmap ENOMEM wording, superseded by the shared syscalls arena rewrite), nx_swcompare_matrix 2afa1f7b 1448024 B and nx_swcompare_sota 9a7d7acb 823360 B (LOST 2 each = STS-LOCK wording superseded by the plane-lock rewrite), nx_refarea_gate 3cd60ae1 245808 B; all offc-installed, regen FORK-FRESH on the new matrix |
| 2026-10-03 | GR62 | measure | THREE METHOD CLASSES on the chokepoint filter: (1) nx_refarea_gate 71/71 GREEN via /api/gate_run on the promoted binary (85cff2c8), new teeth each paired with a control: the retired-copy word redacts with no manifest, the manifest's bare file name (derived from its path, no new literal) redacts and keeps its numbers, the bytes form returns an exactly sized buffer (26 for 26) with both JSON escapes outside the token and nothing read past n, and public text returns the same pointer with zero spans; (2) mutation: nx_gate_bite nx_refarea_gate on the filter lib killed=1 of 1 valid mutant, restore byte-identical, both twins restored; (3) independent byte scans of the served files after republish (nx_txtscan count): the store word with its slash, the retired-copy word, the manifest file name, and the stem of every one of the five stored items (searched by name, named here by count only) read 0 on graphics index.html and api.json, aesthetictwin index.html and api.json, and feed.xml; the published lines carry private_spans (aesthetictwin index 1 = the ladder acceptance text the funnels missed). Residual, named: the bare word refstill remains only as code names (graphics: 46 nx_refstill and 15 gr_refstill_referee, the matrix contract symbol and its cell anchor, plus 4 prose words; aesthetictwin api: 2 nx_refstill); a code name carries no store path and is deliberately not redacted. Generators matrix 2afa1f7b and sota 9a7d7acb carry the filter one revision behind the lib (no manifest-file-name needle); the regen net applies it to every byte they emit. Regen promoted 7024b2cc (contentdiff GREEN, lost 0) |
| 2026-10-03 | GR62 | lesson | THE FILTER MATCHES WHOLE NEEDLES, SO A JOURNAL THAT QUOTES A SHORTENED STEM LEAKS PAST IT: this seat's own measure row listed the searched stems in shortened form, two of them were not whole needles, and the republished page carried them until the row was rewritten to name the items by count (corrected in place minutes after landing; the served bytes were re-scanned after). Imprecision chosen and named: containment of a whole stem, id, former id or store word; a fragment of a stem is not a needle (splitting stems into words would redact common English). Rule for every seat: name a private item by its opaque id or by count, never by any part of its file name |
| 2026-10-03 | GR62 | queue | INTAKE CLASSES SOURCE-LANDED, NOT YET PUBLISHED (build admission refused every build for 55 minutes, load1 19 to 25 against the 16.43 ceiling; nothing promoted, served pages unchanged). Measured before the fix by served-byte count on aesthetictwin: the pictured person's surname 68 in index.html and 6 in api.json, the intake store path 28 and 4, the vault root 6 and 0, vault cids 10 prose occurrences, a third-party site name 12; graphics read 0 on name, intake path and vault root. nx_privref_lib gains four needle classes, all containment and over-redacting: (3) the intake store word and its retired leaf plus the vault root, as consts so an unreadable intake still redacts; (4) a vault cid span unless it is a public gallery url; (5) the stem of every oracle file in the intake directory LISTING (names only, no intake file opened); (6) rows of a private names conf kept inside the intake store (name rows: containment in any case and spelling; stemnum rows: a span that is the prefix followed by digits). Tokens are a class-salted hash of the folded needle, one token per name across spellings. Fixture forms (prvr_redact_in, prvr_redact_n_in) read no intake so existing teeth stay deterministic; new prvr_redact_in2 and prvr_redact_n_in2 take the conf and directory; live prvr_redact and prvr_redact_n (the regen chokepoint) read both. nx_refarea_gate gains 20 teeth (section E: each class redacted, public gallery url and non-hex dash and non-oracle stem and contract symbol and citation key left alone, one token per name, unreadable intake still redacts the words, fixture form reads no names, plus live teeth that every declared name row and every live oracle stem redacts). RECIPE when the box clears: queued rows BQ-nx_refarea_gate and BQ-nx_compare_regen fire on the first nx_orchestrate pass with headroom; then run the gate (expect GREEN), nx_contentdiff live vs staged regen and read every LOST line, /api/promote with expect_sha256, nx_offc_install, scoped regen aesthetictwin then graphics, nx_page_verify, and recount served bytes. Residual named: the bare prefix of contract and citation symbols stays (it is code, not a stem); a third-party video host named in one public acceptance row is not declared |
| 2026-10-04 | GR62 | land | INTAKE CLASSES PUBLISHED (the 1791063220 queue row closed). Queued rows had not fired (builds predated source by ~1.6 h), so built by hand once admission granted (load1 3.4, blocked 0). nx_refarea_gate --qualify read 87/88 RED from CWD buildroot: section E 20 of 20 PASS incl. live names 6 of 6 and live oracle stems 2 of 2; the one RED was live-conf-files-cc-by-nc-nd-in-cc-nc-nd plus 0 live manifest rows, both because the live conf and manifest are serving-root relative and --qualify anchors CWD to buildroot (the conf does carry the NC+ND row). The same candidate staged and run from the serving root: nx_behaveprobe other_rc=0 (GREEN). Content ruler, every LOST line read: gate lost 0 (gained 189), regen lost 0 (gained 29), matrix lost 0 (gained 120), sota lost 0 (gained 122), all GREEN ONE-WAY-GAIN. Promoted with expect_sha256: gate f12f25eb, regen 664f55d4, matrix 32acd2d6, sota f91c638a; nx_offc_install promoted on all four, buildroot/_offc REFRESHED for regen, matrix and sota; regen FORK-FRESH on the matrix fork. Scoped republish aesthetictwin (index private_spans=20, api 0) and graphics (0 and 0), each GREEN and each rewriting feed.xml; nx_page_verify GREEN on both. Served-byte recount (nx_txtscan count, items named here by count only) over aesthetictwin index.html and api.json, graphics index.html and api.json, and feed.xml: every private.names name row in three case forms 0 in all five files; the third-party site name 0; stemnum spans (prefix plus each digit 0-9) 0, the bare prefix surviving only as function names and citation keys; the intake store path 0; the vault root 0; both oracle stems 0; optout.names carries no rows yet (0 needles). Vault cids: 1 in each aesthetictwin file, both the declared public gallery url; feed 1 prose mention with no hex. private-ref tokens on the pages: aesthetictwin index 371, graphics index 260. Residual, named: aesthetictwin index.html carries two non-oracle intake subject stems at 2 occurrences each (exempt by design, the allow-non-oracle-intake-file-stem control) and 16 plan-plane ids sharing a prefix with intake plan file names; one gallery image path under the NAS ai image tree (not the vault). Gate binary on its serving-root run is the evidence; a NAS /api/gate_run receipt with the live manifest rows was not taken this leg |
Rendered evidence





Measured head-to-head — a receipt, not a claim
headless-edge-on-west-laptop|Chromium DevTools protocol via nishi-ops/ac_cdp.py|viewport SMALL: canvas 735x368 at devicePixelRatio 1|NOT the operator's display and NOT an RTX 5080 -- this receipt bounds nothing about that hardware on 2026-09-04 (unix 1788538300), 256 runs per arm, reference arm budget-60fps = 1.00x. Every arm had to print the same checksum or its row is VOID and never ranked; an arm whose toolchain is not declared on the host is UNMEASURABLE, an absence rather than a loss. Written by hand-recorded from the served page's own telemetry; a MEASURING ORGAN is owed (see the note on budget-87 below) and until it exists this receipt is a floor, not a beat, re-derived here by nx_bench_receipt_lib: reader verdict UNMEASURABLE, writer wrote VALID (WRITER AND READER DISAGREE -- the rows were edited after the receipt was written).| Arm | Toolchain | Version | Median µs | Min µs | vs reference | Checksum | Status | Runs | Source sha256 | Binary |
|---|---|---|---|---|---|---|---|---|---|---|
| served-beach-webgl2 | WebGL2 + npc-atlas 2048, q=2 so scl=0.707 | served build b1788537121 | - | - | - | - | UNKNOWN MEASURED 2026-09-04: p50=54000 p95=71000 p99=83000 worst=89000 worst_ever=250000 us. Against budget-60fps this is 3.24x OVER AT THE MEDIAN and 5.0x at p99, and worst_ever is a quarter-second hitch. AND IT IS ALREADY PAYING IN PIXELS: q=2 means the adaptive ladder has cut linear scale to 0.707, so this time is being taken at 70 percent resolution and would rise further at native. | 256 | 667335ae5430 | - 0 B |
| budget-87-the-number-actually-in-force | the page's own budget_ms as shipped | served build b1788537121 | - | - | - | - | UNKNOWN THE GOVERNOR IS HITTING ITS TARGET. budget_ms=87 is 11.5 fps, a number nobody would choose as a goal, and it is what the adaptive quality ladder optimises toward -- which is why resolution falls whenever work is added. This arm exists to make that constant VISIBLE beside the published bars rather than buried in an emitted string literal where nx_magic cannot see it. It is not a benchmark result; it is the bar the estate has been unknowingly grading itself against. | 256 | 667335ae5430 | - 0 B |
bench arms=2 valid=0 void=0 unmeasurable=0 build_fail=0 run_fail=0 unknown=2 (partition sums) verdict=UNMEASURABLE
reproduce:
receipt: knowledge/compare/graphics.bench · a rerun on the same host that changes the ranking is a finding, not noise; a rerun on a different host is a different receipt and says so in its host line.
Person · product · place — not yet measured for this domain
knowledge/compare/graphics.ppp (rows surface|nishi or c1..c4|label|url|connect naming OUR live surface and each rival's front door), run nx_ppp_probe domain graphics, and this section fills itself on the next beat: the same ruler on both sides — privacy and CX (third-party hosts, tracker classes, cookies, security headers), design and longevity (design hygiene, computed WCAG contrast, render-blocking resources, unsized media, script weight, theme and motion queries), findability (landmarks, skip link, on-site search, breadcrumb, headings, internal links).Mined from the field — review and competitive intelligence proposed to this board
Rows written by nx_intelmine_propose from nx_reviewmine_lib signals (Steam reviews under the exceeds / meets / mixed / does-not-meet rubric, the installed-title census) and routed here by capability_map.conf. DEFECT names a rival failing that a rung here answers; DEMANDED a capability buyers ask for and do not get; WE-DO-BETTER an exceed the reviews corroborate; THEY-DO-WELL a bar this board must meet; SHIPS a rival capability to match. A proposal is a lead, never a rung — it closes only when a rung with a gate lands, and nothing in this band changes a measured cell.
| Signal | Kind | Term | Title | Proposed rung | Evidence |
|---|---|---|---|---|---|
| THEY-DO-WELL | feature | visuals | Portal 2 620 | Visual fidelity: 620 -- referee-graded rendering against the field | praised=2 complained=0 mixed=0 demanded=0 |
proposals 1|malformed rows 0 (counted, never rendered: a row that is not prop| with 10 fields)|read-capped 0
The field — discovered, not chosen
Rows written by nx_field_discover from graphics.seeds: the industry's own lists (Wikipedia wikitext, GitHub topics, awesome lists) read mechanically, every candidate counted across seeds. The matrix columns above are a SEAT'S pick; this band is the population they were picked from, and the stats line measures one against the other. A rival here is a lead, never a verdict — it earns a column when its capabilities are read and pinned.
| Class | Count | Note |
|---|---|---|
| seeds | 2 | |
| scanned | 2 | |
| fetched | 2 | |
| reused | 0 | |
| failed | 0 | |
| named | 2 | |
| candidates | 2 | |
| mentions | 2 | |
| capped | 0 |
| Rank | Rival | Seeds | Mentions | First seed | Kind | Link |
|---|---|---|---|---|---|---|
| 1 | Unreal Engine 5 | 1 | 1 | col0-unreal | named | https://dev.epicgames.com/documentation/unreal-engine/creating-human-skin-in-unreal-engine |
| 2 | Infinigen | 1 | 1 | col2-infinigen | named | https://infinigen.cs.princeton.edu/docs/latest/index.html |
field candidates 2|shown 2 of 2|matrix columns in the field 2 of 4|discovered rivals with no column 0|malformed rows 0 (counted, never rendered)|read-capped 0
Exclusive and shared capabilities
This board grades 105 capabilities: 65 exclusive to it and 40 shared with other boards. Shared rows on this page: 40 = home here 9 + home elsewhere 10 + ambiguous 7 + undecided 14 + refused 0 (partition 40); undecided = unmapped 0 + unproven 14 + unevidenced 0; homonyms 0; since the previous census: new 0, resolved 0. Census FRESH age_s=186.
| Capability | Organ | Boards grading it | Home | Rule | Evidence | Standing on this board |
|---|---|---|---|---|---|---|
| cjc_hw | nx_charjudge.nx | mangagen, charsim+, graphics, koikatsu*, gen | koikatsu | rung | plan | home elsewhere adopt: @xrow koikatsu|cjc_hw|home by rung (fleet duplicate census) |
| m_joints | nx_bvhmotion.nx | mangagen, charsim, graphics, koikatsu, gameengine, motion | — | none | unproven | undecided |
| ntx_apply | nx_nxa_texc_lib.nx | charsim, graphics, dcc | — | none | unproven | undecided |
| ntb_bake | nx_nxa_texbake_lib.nx | charsim+, graphics+, dcc | — | ambiguous | gates | ambiguous |
| ntb_bake_pbr | nx_nxa_texbake_lib.nx | charsim*+, graphics*+, phototwin, dcc | — | ambiguous | plan | ambiguous |
| sr_sss_preint | nx_sdfrender.nx | charsim*, graphics* | — | ambiguous | plan | ambiguous |
| fa_facs_au | nx_faceanat.nx | charsim, graphics | — | none | unproven | undecided |
| fa_expr_morph | nx_faceanat.nx | charsim*, graphics* | — | ambiguous | plan | ambiguous |
| wp_shade | nx_worldpipe.nx | charsim, graphics | — | none | unproven | undecided |
| wc_mob_mesh | nx_wasm_craft.nx | charsim, graphics, gameengine | — | none | unproven | undecided |
| cjc_panel | nx_charjudge_census.nx | charsim+, graphics*, aesthetictwin | graphics | rung | plan | home here |
| gs_avatar_genome | nx_gsplat.nx | charsim, graphics* | graphics | rung | plan | home here |
| g4_avatar_rig | nx_gsplat4d_lib.nx | charsim, graphics*+ | graphics | rung | plan | home here |
| si_albedo_of_gene | nx_skin_ita.nx | graphics+, aesthetictwin | graphics | gates | gates | home here |
| bg_cmr | nx_body_gen.nx | graphics+, procgen | graphics | gates | gates | home here |
| bg_parse_layers | nx_body_gen.nx | graphics+, procgen, aesthetictwin | graphics | gates | gates | home here |
| bp_load_dimorph | nx_body_proc.nx | graphics, procgen, aesthetictwin+ | aesthetictwin | gates | gates | home elsewhere adopt: @xrow aesthetictwin|bp_load_dimorph|home by gates (fleet duplicate census) |
| emitDigitP | nx_body_proc.nx | graphics, aesthetictwin+ | aesthetictwin | gates | gates | home elsewhere adopt: @xrow aesthetictwin|emitDigitP|home by gates (fleet duplicate census) |
| bp_species_canon | nx_body_proc.nx | graphics*, aesthetictwin*+ | aesthetictwin | rung | plan+gates | home elsewhere adopt: @xrow aesthetictwin|bp_species_canon|home by rung (fleet duplicate census) |
| creaturegen_build_plan | nx_creaturegen.nx | graphics, procgen, aesthetictwin | — | none | unproven | undecided |
| cg_genome_parts | nx_creaturegen.nx | graphics*, procgen* | — | ambiguous | plan | ambiguous |
| as_load_sites | nx_anat_sov.nx | graphics, anatomy | — | none | unproven | undecided |
| rg_fk_y | nx_anat_rig.nx | graphics, procgen, anatomy | — | none | unproven | undecided |
| sd_step | nx_softdyn.nx | graphics, procgen, softbody, dcc, gameengine | — | none | unproven | undecided |
| rd_analyze | nx_ringdown_lib.nx | graphics, synthroom | — | none | unproven | undecided |
| dbs_displace_free | nx_dyna_bind_skin.nx | graphics*, procgen, aesthetictwin | graphics | rung | plan | home here |
| gg_stats | nx_nxa_groom_gate.nx | graphics+, aesthetictwin | graphics | gates | gates | home here |
| gs_render_aniso | nx_gsplat.nx | graphics, procgen | — | none | unproven | undecided |
| gs_color_grad_step | nx_gsplat.nx | graphics, procgen* | procgen | rung | plan | home elsewhere adopt: @xrow procgen|gs_color_grad_step|home by rung (fleet duplicate census) |
| gb_bind | nx_gsplatbind.nx | graphics, procgen* | procgen | rung | plan | home elsewhere adopt: @xrow procgen|gb_bind|home by rung (fleet duplicate census) |
| ig_room_solve | nx_indoorgen_lib.nx | graphics*, procgen, gameengine | graphics | rung | plan | home here |
| bb_detail_score | nx_bodybench.nx | graphics, procgen, anatomy, aesthetictwin | — | none | unproven | undecided |
| gpe_build | nx_game_page_emit.nx | graphics, gameengine | — | none | unproven | undecided |
| gpe_emit_webgpu | nx_game_page_emit.nx | graphics, gameengine | — | none | unproven | undecided |
| wgsl_emit_module | nx_wgsl.nx | graphics, gpu+ | gpu | gates | gates | home elsewhere adopt: @xrow gpu|wgsl_emit_module|home by gates (fleet duplicate census) |
| wp_erode | nx_worldpipe.nx | graphics, procgen* | procgen | rung | plan | home elsewhere adopt: @xrow procgen|wp_erode|home by rung (fleet duplicate census) |
| wc_veg_mesh | nx_wasm_craft.nx | graphics*, gameengine* | — | ambiguous | plan | ambiguous |
| gpu_raster_rung | _gpu_dxg_submit_gate.nx | graphics, gameengine*, gpu* | — | ambiguous | plan | ambiguous |
| gpe_wasm_npc_bind | nx_game_page_emit.nx | graphics, gameengine* | gameengine | rung | plan | home elsewhere adopt: @xrow gameengine|gpe_wasm_npc_bind|home by rung (fleet duplicate census) |
| wc_hair_tick | nx_wasm_craft.nx | graphics, gameengine* | gameengine | rung | plan | home elsewhere adopt: @xrow gameengine|wc_hair_tick|home by rung (fleet duplicate census) |
Gaps from the record — what the estate does that no board carries
The record census (nx_goalmap record) reads the invoked-tool population and every plan queue row and files each organ or directive that NO matrix, plan or gates row names. A row here is a callout the boards missed: adjudicate it onto a board or declare it infrastructure. Census state STALE (age 1396633 s), sources read 4 of 7 declared — a BLIND census is a FLOOR: unread sources can only add rows.
| kind | name | board | source | evidence |
|---|---|---|---|---|
| directive | nx_wgsl_pixel_gate | graphics | plan-queue | rung=GR11 |
| directive | nx_png_write | graphics | plan-queue | rung=GR42 |
rows shown 2|this board's directives 2|estate-wide un-boarded organs 452 (listed in full on /compare/ecosystem)|census rows 639|malformed 0 (counted, never rendered)
References
- [mei2007] Mei, Decaudin, Hu. Fast Hydraulic Erosion Simulation and Visualization on GPU. Pacific Graphics 2007. publisher · accessed 2026-08-18 · published-paperGrounds: EH1's hydraulic lane: the pipe-model shallow-water erosion this rung implements integer-exact, with mass conservation promoted from float hope to gate tooth.
- [musgrave1989] Musgrave, Kolb, Mace. The Synthesis and Rendering of Eroded Fractal Terrains. SIGGRAPH 1989. publisher · accessed 2026-08-18 · published-paperGrounds: EH1's thermal-talus lane and the founding statement that noise terrain without erosion reads wrong -- the classical bar the game tier meets.
- [terraindiff26] InfiniteDiffusion / Terrain Diffusion: training-free spatially-infinite diffusion with seed-consistency and constant-time random access; a learned real-time infinite terrain generator with a procedural-noise interface. SIGGRAPH 2026 (top-10 attention). publisher · read in our library
knowledge/fetched/cmp_graphics_terraindiff26.html· pinh8c695fd94b91a2891f69e194c98c2e16ed00baec069f39939c01af8dfe53f789· accessed 2026-09-28 · published-paperGrounds: THE 2026 CEILING for the beauty tier: the frontier moved from hand-tuned noise+sim to LEARNED generators that keep procedural properties. Our integer angle rides the nofloat ML stack; any beauty-tier terrain claim grades against this class, not against classical noise. - [metahuman-dna] Epic Games. MetaHuman DNA, Rig Logic and pipeline reference: DNA encodes the head and body geometry and the skeleton in its neutral pose at multiple levels of detail with skin weights stored per LOD, and Rig Logic drives a predetermined set of expressions as joint animations augmented by a per-vertex displacement present at LOD0 only. publisher · read in our library
knowledge/fetched/cmp_graphics_mhdna.html· pinh2c10945784e4d8f879b27e69f29a8d30a03bcc720b5f4a724a70cc69a02b8321· accessed 2026-09-04 · vendor-docGrounds: CORRECTED AND PARTLY STRUCK 2026-09-04, and the correction is the finding. This row previously attributed to this URL: head 24K-60K tris at LOD0, about 700-800 joints, 710 FACS-driven expressions, 8K source textures streamed 4K, pore-level deformation, and strand grooms at 3-6ms per character on 3070-class GPUs -- while carrying mirror and pin both unset, so no tooth on this board could ever compare the claim against a stored byte. The page was fetched and pinned for the first time on 2026-09-04 and read end to end as 95151 bytes of extracted text: it contains no triangle count, no joint count, no expression count, no texture size, no occurrence of the strings pore or FACS or groom, and no millisecond timing or GPU model anywhere. A POSITIVE CONTROL on the same bytes passes -- DNA appears 58 times, joint 90, LOD 45 -- so the absence is a property of the document and not of a failed fetch or a JS shell. The frame-cost claim of 3-6ms per character on 3070-class GPUs is STRUCK OUTRIGHT as unsourced, and it is the one that mattered because a per-character millisecond budget is what a renderer plans against. The structural numbers are SUPERSEDED by mhlodspec, the Epic page that does publish them and now carries them pinned, and three of the four were wrong there too: 24000 head VERTICES at LOD0 rather than 24K-60K triangles, exactly 713 joints rather than a 700-800 band, 669 blendshapes rather than 710 FACS expressions, and max texture 8192 confirmed. This row is kept, mirrored and pinned for the qualitative Rig Logic architecture only, and NO NUMBER may be cited from it. - [metahuman-ref0] Epic Games. MetaHuman landing-page hero still (social-share render of a MetaHuman head, 1200x675 PNG), fetched through the sovereign fetcher 2026-08-19 and banked as the far-bar GOOD reference for the referee panel; truecolor twin cmp_graphics_metahuman_ref0_rgb.png via nx_charjudge_census topng. publisher · read in our library
knowledge/fetched/cmp_graphics_metahuman_ref0.png· pinh598d5eaedd10f0a65d9d914c884bc9d641175c224d6b913235cc4c5cd4520c55· accessed 2026-08-19 · vendor-docGrounds: REFEREE: multi-lens judge panel that can see beauty - [metahuman-ref1] Epic Games. MetaHuman landing-page hero still (cinematically graded MetaHuman face, 960x540 JPEG from the CMS asset pipeline), fetched 2026-08-19 and banked as the second far-bar GOOD reference; truecolor twin cmp_graphics_metahuman_ref1_rgb.png. The second-corpus retest on these two stills refuted the Kovac daylight skin floor (graded skin reads 0-6 permil) while the percept lens separated at fp=0 fn=0. publisher · read in our library
knowledge/fetched/cmp_graphics_metahuman_ref1.img· pinh36f796370a2e8795b0085296903754b1eaff82698a0b0ed9251b349a8a95d32e· accessed 2026-08-19 · vendor-docGrounds: REFEREE: multi-lens judge panel that can see beauty - [webgpu-status26] W3C GPU for the Web Community Group. WebGPU Implementation Status (living page); Safari 26 / Firefox 147 ship notes. Accessed August 2026. publisher · accessed 2026-08-18 · vendor-docGrounds: The DELIVERY tier ladder and its coverage claim: ~82 percent global WebGPU support as of 2026-08 (Chrome/Edge 113+, Safari 26 default-on, Firefox 147 Windows/ARM64-macOS with Linux through 2026) -- why WebGL2 remains the second tier and sovereign wasm the floor.
- [webgpu-roadmap26] Chrome GPU team. What's next for WebGPU (subgroups shipping behind flags, bindless in proposal via texture_and_sampler_let, no ray-tracing extension in spec). Accessed August 2026. publisher · accessed 2026-08-18 · vendor-docGrounds: The WebGPU-tier contract scope: RT stays compute-shader on the web in 2026; subgroups and bindless are the watched upgrades, not assumptions.
- [wasm3-state26] WebAssembly 3.0 in browsers: WasmGC, Memory64, relaxed SIMD, threads (COOP/COEP), tail calls -- stable in Chrome 119+/Firefox 120+/Safari 18.2+. Accessed August 2026. publisher · accessed 2026-08-18 · vendor-docGrounds: The sovereign-wasm floor's headroom: our MVP-profile emission can adopt SIMD/threads/Memory64 wherever the vm-gate proves equivalence -- the 85-95 percent-of-native band the delivery rows lean on.
- [infinigen23] Raistrick, Lipson, Ma et al. Infinite Photorealistic Worlds Using Procedural Generation (Infinigen). CVPR 2023. publisher · accessed 2026-08-17 · source-readGrounds: The Infinigen column codes and the single-surface-vs-layers differentiator: read from princeton-vl source (BSD-3), banked on the graphics board -- NURBS part lofting, genome part interpolation, per-species material node groups.
- [infinigen-indoors24] Raistrick, Mei, Kayan et al. Infinigen Indoors: Photorealistic Indoor Scenes Using Procedural Generation. CVPR 2024. publisher · accessed 2026-08-17 · published-paperGrounds: The ig_room_solve rung: constraint-based room solving with furniture placement is their published method; our contract is the same capability through our engine, refusing unsatisfiable sets.
- [infinigen-docs] Infinigen 2.0 documentation (Princeton Vision and Learning Lab). publisher · accessed 2026-08-17 · vendor-docGrounds: The 2.0 alpha capability read: controllable APIs, indoor materials and simplified rooms, procfunc composable generators, ground-truth annotation pipeline, USD/OBJ/FBX and MuJoCo export.
- [infinigen-flying26] Infinigen team. 2026-08 flying-indoor-preview dataset (17,500 video rows, 331 GB, BSD-3, under construction). publisher · accessed 2026-08-17 · datasetGrounds: The ws_groundtruth_emit contract: their annotation pipeline is why this dataset exists; ours makes the estate a dataset factory on the same axis.
- [kerbl3dgs] Kerbl, Kopanas, Leimkuehler, Drettakis. 3D Gaussian Splatting for Real-Time Radiance Field Rendering. ACM TOG 42(4), SIGGRAPH 2023. publisher · read in our library
knowledge/fetched/cmp_graphics_kerbl3dgs_inria.html· pinh846c9743d36c7f7335f2e999e13bfded5f5cb831b53ad3eb3b2471544207fbca· accessed 2026-08-17 · published-paperGrounds: The SPLAT rows: gs_render_aniso implements the representation sovereignly in integer; gs_color_grad_step is the first tooth of the trained-fit loop the paper defines. - [splattingavatar24] Shao, Wang, Zhu et al. SplattingAvatar: Realistic Real-Time Human Avatars with Mesh-Embedded Gaussian Splatting. CVPR 2024. publisher · published-paperGrounds: The gb_bind hybrid: splats bound barycentrically to mesh triangles, mesh deforms and splats follow -- the operator-ordered outside-splat inside-mesh architecture.
- [penner2011] Penner, Borshukov. Pre-Integrated Skin Shading. GPU Pro 2, 2011 (SIGGRAPH course lineage). publisher · accessed 2026-08-17 · published-courseGrounds: The sr_sss_preint contract and the shipped wrap-SSS floors: diffusion widens the terminator per channel, red furthest -- the T66 equality tooth pins exactly the consequences no tuning can fake.
- [walter2007] Walter, Marschner, Li, Torrance. Microfacet Models for Refraction through Rough Surfaces. EGSR 2007. publisher · published-paperGrounds: The gp_ggx row: GGX/Trowbridge-Reitz D with Smith G and Schlick F is this paper's model, implemented integer-exact in the sovereign renderer.
- [karis2021] Karis, Stubbe, Wihlidal. Nanite: A Deep Dive. SIGGRAPH 2021, Advances in Real-Time Rendering in Games. publisher · read in our library
knowledge/fetched/cmp_graphics_karis2021.html· pinhe567b6ff0a2b157ada8222dcde9486ed4563f3b19a1af282e74e15cf4d9344c4· accessed 2026-09-28 · published-courseGrounds: The geometry-scale bar the retired hand page named; the s2021 Advances page carrying this talk is now mirrored in the library (the GR7 crawl, landed 2026-08-17). - [bitterli2020] Bitterli, Wyman, Pharr, Shirley, Lefohn, Jarosz. Spatiotemporal Reservoir Resampling for Real-Time Ray Tracing with Dynamic Direct Lighting (ReSTIR). ACM TOG 39(4), SIGGRAPH 2020. publisher · read in our library
knowledge/fetched/cmp_graphics_bitterli2020_nvidia.html· pinh39d1efe0b42a0ab60b071666eee49988dd02af27646ff1dc4a271793a9160608· accessed 2026-08-17 · published-paperGrounds: The gp_bounce_gi contract names ReSTIR-class sampling as the bar beyond our cached-form-factor radiosity. - [chardon91] Chardon, Cretois, Hourseau. Skin colour typology and suntanning pathways. International Journal of Cosmetic Science 13, 1991. publisher · published-standardGrounds: The si_albedo_of_gene row: ITA (Individual Typology Angle) classes dark to very-light are this paper's typology; nx_skin_ita computes them in CIE Lab integer arithmetic.
- [bodyparts3d] Mitsuhashi, Fujieda, Tamura et al. BodyParts3D: 3D structure database for anatomical concepts. Nucleic Acids Research 37, 2009. CC BY-SA 2.1 JP. publisher · datasetGrounds: The oracle behind the BODY and ANATOMY rows: 1.26k meshes used as MEASURING STICK only, never shipped as product -- the product is the procedural generator measured against it.
- [kesen-mirror] Ke-Sen Huang's computer graphics paper index (SIGGRAPH, SIGGRAPH Asia, EG, I3D, EGSR, SCA, SGP, HPG, PG, 2000-2026), mirrored into the estate library. publisher · read in our library
knowledge/library/gfx_papers_index.txt· pinh1650826094a4b15ed0ac0877414162d2309cdcac4d41115ad292053ff0314aa5· accessed 2026-08-15 · mirrored-indexGrounds: The RESEARCH rows and every gfx_papers_index mirror path above: the fixed-point crawl proved link-completeness (reachable=236 mirrored=235 papers=9762, partition reconciled); one link was already dead at capture, which is why bytes beat bookmarks. - [advances-archive] Advances in Real-Time Rendering in Games, SIGGRAPH course series 2006-2026 (official archive). publisher · read in our library
knowledge/library/gfx_adv_s2026_index.html· accessed 2026-08-17 · published-courseGrounds: LANDED 2026-08-17: the archive is mirrored beside the Ke-Sen index under the gfx_adv_ prefix -- reachable=26 mirrored=25 dead=1 failed=0 truncated=0, partition RECONCILES (the one dead link had already rotted at capture, which is the argument for bytes over bookmarks). The mirror link here opens the newest year; the remaining tail is the extractor citing adv_ pages into the paper index. - [metahuman] Epic Games. MetaHuman Creator documentation. publisher · accessed 2026-08-17 · vendor-docGrounds: The Unreal Engine 5 column codes on character rows and the photoreal far bar; the north star (the inside-and-out twin) is graded against MetaHuman-class stills by the cjc_panel contract.
- [infinihuman25] Xue, Xie, Kostyrko, Pons-Moll. InfiniHuman: Infinite 3D Human Creation with Precise Control. arXiv:2510.11650, October 2025. InfiniHumanData: 111K identities annotated with text, multi-view images, clothing and body-shape parameters; InfiniHumanGen: diffusion-based controllable avatar generation. publisher · accessed 2026-08-18 · published-paperGrounds: The operator's realistic-human oracle ask (2026-08-18, an Infinigen-2.0 search): Infinigen ships NO human generator (source-read: its genome classes are creatures), so the human-realism bar on this axis is InfiniHuman-class controllable synthesis beside the MetaHuman stills -- the gs_avatar_genome leap and the cjc_panel referee grade against these external oracles, never our own eyeball.
- [kkguides] Community Koikatsu character-creation guide pack (KK Guides): character_creation_guide.pdf 4.5MB, Wubsy material-editor guides 1-4, shadow-texture guide, 60+ studio and shading items. pCloud public share, folder read 2026-08-18 via the publink API. publisher · accessed 2026-08-18 · published-courseGrounds: Operator-supplied oracle for the KK retarget lane (/compare/koikatsu, nx_kkfacts, nx_retarget_showcase) and the MATERIAL and skin rows: the pack names the material-editor and shadow-texture practices a KK-class card expects; our sovereign emit is measured against what the guides teach -- reference only, no third-party asset ever ships as product.
- [relooper-zakai11] Zakai. Emscripten: An LLVM-to-JavaScript Compiler. OOPSLA 2011 (SPLASH companion). The paper that names, specifies and proves the termination of the Relooper, whose three block kinds are Simple, Loop and Multiple. publisher · read in our library
knowledge/fetched/cmp_graphics_relooper-zakai11.pdf· pinhcfe4d5d4207254a34fc9f4937a622d2e7cbdc57aac4dbaa0e6ad0972006253ca· accessed 2026-08-23 · published-paperGrounds: DELIVERY: shaders compiled from NishiLang (nx_wgsl backend) -- the rung 2b structured control-flow recovery. Extracted with nx_pdf_text and quoted only from bytes read: a Loop block's Inner block is the one where "when execution reaches the end of that block, flow will return to the beginning" and which "will contain a conditional break defining where it is exited", which is exactly the S_LOOP plus S_BREAK form shl_region emits. The paper's label variable is introduced there because it "greatly simplifies the problem that the Relooper needs to solve" for multi-entry regions; our measured corpus has none, so we carry none and refuse by name instead. - [llvm-cfgstackify] LLVM Project. WebAssemblyCFGStackify.cpp, the WebAssembly backend's CFG stackification pass, which inserts BLOCK, LOOP, TRY and TRY_TABLE markers because scope boundaries serve as the labels for WebAssembly's control transfers. publisher · read in our library
knowledge/fetched/cmp_graphics_llvm-cfgstackify.cpp· pinh9bae82c88eb9ec5aba93cf7e9323da5a54b17ccb62cf3f8e7a3051e60345d413· accessed 2026-08-23 · source-readGrounds: DELIVERY: shaders compiled from NishiLang (nx_wgsl backend) -- the precondition that licenses a shape reducer instead of a full relooper. Read from the mirrored source: marker insertion "is sufficient to convert arbitrary CFGs into a form that works on WebAssembly, provided that all loops are single-entry". Every loop header in the measured shader corpus has exactly two predecessors, one preheader and one latch, so that precondition holds here; shl_region refuses BY NAME the moment a header appears with n_preds != 2. - [syn4d] Jiang, Lan, Luo, Deng, Lai, Sucar, Rupprecht, Laina, Larlus, Zheng, Vedaldi. Syn4D: A Multiview Synthetic 4D Dataset. ECCV 2026 Highlight. Visual Geometry Group University of Oxford, Nanyang Technological University, and Naver Labs Europe. publisher · read in our library
knowledge/fetched/cmp_graphics_syn4d.html· pinh11884d72fd639550e8f6828c96a8161aee786f82bda16f1a9b565afac6f7db6f· accessed 2026-08-25 · datasetGrounds: WORLD: ground-truth dataset emission -- RGB, depth, normals, semantics. WITNESSED in the pinned 42665-byte mirror and never through a summarising fetcher: their annotation pipeline renders dynamic scenes with a procedural multi-camera setup that jointly outputs RGB, depth, instance segmentation, optical flow, dense 3D tracks, and parametric human pose, and every scene carries per-pixel depth, multi-view camera trajectories, dense long-range 3D point tracks, and parametric SMPL-X human pose and shape. THIS IS THE 2026 PUBLISHED BAR THE ws_groundtruth_emit CONTRACT IS MEASURED AGAINST, and it is the same shape as our own emit verb because our renderer already holds depth and normals per pixel. Their stated differentiator is the ability to unproject any pixel into 3D at any frame and re-project it into any other camera, a cross-view and cross-time consistency property our single-camera snap does not yet claim -- named here as the gap rather than assumed away. Sits beside infinigen-flying26 as the second dataset-factory oracle on this board, and unlike Infinigen it is HUMAN-centred, which is what ties it to the charsim body rows. - [physrig] Zhang, Xu, Feng, Jampani, Ahuja. PhysRig -- Differentiable Physics-Based Skinning and Rigging Framework for Realistic Articulated Object Modeling. arXiv:2506.20936v2 cs.CV, 27 Jun 2025. University of Illinois Urbana Champaign and Stability AI. publisher · read in our library
knowledge/fetched/cmp_charsim_physrig.pdf· pinh744fc02485dfc52444abc672e445fff96ba284a36684f79ac830a14abbcd3f5d· accessed 2026-08-25 · published-paperGrounds: MOTION: DAMPING identified from video -- the parameter the field does not fit. Quoted ONLY from the pinned bytes via pdftotext and never through a summarising fetcher, because WebFetch on a PDF returns plausible sentences that are not in the document, measured twice on this estate. It embeds a rigid skeleton in a tetrahedral volume simulated as a soft body under continuum mechanics and introduces MATERIAL PROTOTYPES to shrink the learning space -- the same shape as our XPBD tet cage and the st_profile genome draw, which is why it is the right rival to name here. THE REASON IT BACKS THIS ROW IS WHAT IT DOES NOT DO: it optimises Young's modulus and Poisson ratio and NEVER damping, and damping is the parameter that governs jiggle decay and the one nx_dynaoracle identifies from video. - [softsmpl] Santesteban, Garces, Otaduy, Casas. SoftSMPL -- Data-driven Modeling of Nonlinear Soft-tissue Dynamics for Parametric Humans. Computer Graphics Forum, Proc. Eurographics, 2020. publisher · read in our library
knowledge/fetched/cmp_charsim_softsmpl.html· pinhce83469210b59198e67c539cffc9d2492d588110cbe0e1d5c18ae440e8ab5cef· accessed 2026-08-25 · published-paperGrounds: MOTION: one damped-oscillator identifier shared by the solver AND by video. The closest published prior art to this lane: soft-tissue dynamics as a function of body shape and motion for PARAMETRIC humans, which is the measure-then-generate loop this board is building. Its motion descriptor disentangles the pose representation by removing subject-specific features -- the same separation our gait DRIVE and tissue PLANT are deliberately kept apart by, and the reason a plant tuned into its own drive band reads as wrong. IT LEARNS FROM 4D SCANS, NOT FROM MONOCULAR VIDEO, so it does not close the video-identification gap this row claims and is cited as the bar rather than as the answer. - [springbones] Artell. Spring Bones -- Blender addon adding a spring and bouncy dynamic effect to bones. GitHub artellblender/springbones, master branch commit 77ee0fb, 807 stars. publisher · read in our library
knowledge/fetched/cmp_charsim_springbones.html· pinh52ad217c3ddebc0079ac99a22096bf754764cefce6cae71caf9fe339d71a85ee· accessed 2026-08-25 · source-readGrounds: MOTION: hair secondary motion -- spring-bone chain on the ONE solver, measured. The game-convention lineage sitting behind the chest_k_q10 and hair_k_q10 bands: a parent bone that moves drives a child bone with bouncy motion, applied to position and rotation, which is the spring-bone chain our hair rung implements on the tissue solver. CITED AS CONVENTION AND NEVER AS MEASUREMENT -- its own README calls it an alpha proof of concept and a toy to play with, exposing Bouncy and Speed as artist sliders with no biomechanical source. That is precisely why our K and C are DERIVED from cited bands rather than dialled, and why the measured human wins wherever a game-convention band and a measured band disagree. - [anny] Baradel, Bregier. Anny -- a free-to-use 3D human parametric model for all ages. NAVER LABS Europe research blog, 2025. Released under the Apache 2.0 license. publisher · read in our library
knowledge/fetched/cmp_charsim_anny.html· pinhf34c997530b459bd78b9969702266bd4afc6038ca21502e12df98e2c88d799f9· accessed 2026-08-25 · published-paperGrounds: MOTION: METRIC SCALE for video measurement -- the declared limit of the video oracle. An Apache-2.0 parametric human covering babies to the elderly with interpretable age, height, weight and muscle controls, witnessed in the pinned 524248-byte mirror. THE PERMISSIVE LICENCE IS THE POINT: unlike SMPL it can lawfully serve as our metric oracle at all. It supplies the metric body frame an uncalibrated clip cannot -- our oracle reports only dimensionless quantities and prints amp_mm as UNMEASURABLE precisely because no such frame exists yet. ORACLE ONLY AND NEVER PRODUCT, the same rule this board already applies to BodyParts3D and the same reason the referee panel's VLM lens abstains rather than acquits. - [multihmr2] Baradel, Fiche, Weinzaepfel. Multi-HMR 2 -- a unified framework for human-centered scene understanding. NAVER LABS Europe research blog, 2026. publisher · read in our library
knowledge/fetched/cmp_charsim_multihmr2.html· pinh65ad258353a91b55bd7fe43df0d7a7114fa5ad5395d35b045b741d6014126d6d· accessed 2026-08-25 · published-paperGrounds: MOTION: METRIC SCALE for video measurement -- the declared limit of the video oracle. Monocular multi-person whole-body mesh recovery in the Anny representation, so it RECOVERS a metric body from a single camera where our oracle can only measure ratios. It is the instrument that would convert our dimensionless fn, zeta, phase lag and amplitude ratios into millimetres and let the cited tissue_amp_walk_mm 32..52 and tissue_amp_run_mm 110..194 bands finally be MEASURED against our render instead of only quoted at it. A learned model, therefore ORACLE ONLY AND NEVER PRODUCT. - [nagano15] Nagano, Fyffe, Alexander, Barbic, Li, Ghosh, Debevec. Skin Microstructure Deformation with Displacement Map Convolution. ACM Transactions on Graphics 34(4), SIGGRAPH 2015. USC ICT Vision and Graphics Lab project page (the Imperial RGI page is also mirrored as cmp_graphics_nagano15-ic.html). publisher · read in our library
knowledge/fetched/cmp_graphics_nagano15.html· pinhfc4e1003c0813603a6a147c9c33a540f24c9ea3c6d17658c9b6291f4cfaabba3· accessed 2026-08-30 · published-paperGrounds: MATERIAL: skin MICRO-SURFACE on the live path -- pores and areola structure, generated not painted. WITNESSED in the pinned mirror: We use a 10-micron resolution scanning technique to measure several in vivo skin samples as they are stretched and compressed in different directions, quantifying how stretching smooths the skin and compression makes it rougher. Microstructure is a DISPLACEMENT that responds to strain, which is why the asset half bakes a height channel in microns and names ntr_tension_attenuate as the next contract rather than asserting a response no stretch field exists to drive. - [ue-skin] Epic Games. Creating Human Skin in Unreal Engine. Unreal Engine 5 documentation, accessed 2026-08-30. publisher · read in our library
knowledge/fetched/cmp_graphics_ue-skin.html· pinhb512e921e73f7d7ef9aa597fc3b87ed24e79aca0fa3a7e109604949ee276a6c4· accessed 2026-08-30 · vendor-docGrounds: MATERIAL: skin MICRO-SURFACE on the live path -- pores and areola structure, generated not painted. WITNESSED: The Normal Map that is used for the face should only contain details about wrinkles. Pore details should be done with a tiling texture as you can get better results this way; and: because the pore detail needs to be very fine, it is very difficult to get the needed quality by just adding the Pores to the base Normal Map. The periodic micro-relief tile in TEXR is this practice made physical: its texel is two per pore orifice, its repeat is the LCM of the band lattices, and the page repeats it per chart at circumference/repeat by length/repeat from the joint table the section carries. - [mh-materials] Epic Games. MetaHuman Materials and Textures. MetaHuman documentation, accessed 2026-08-30. publisher · read in our library
knowledge/fetched/cmp_graphics_mh-materials.html· pinh1c960a93d44e6fe71a466cf2f0deab7bc625fdf001466c76a0244ef9e6919035· accessed 2026-08-30 · vendor-docGrounds: MATERIAL: skin MICRO-SURFACE on the live path -- pores and areola structure, generated not painted. WITNESSED in the LOD table of the pinned mirror: Micro Skin Detail normals are disabled at LOD1 (so a separate micro skin detail normal layer exists at LOD0), the hero assembly bakes normal and specular maps up to 8K, and Base detail normal map baked to a single map. A unique normal plus a separate micro layer is the split TEXR carries: per-region unique height+normal, plus the tiling micro tile. - [txyz-vface] Texturing.xyz. VFace multi-channel Displacement (documentation page), accessed 2026-08-30. publisher · read in our library
knowledge/fetched/cmp_graphics_txyz-vface.html· pinh615ae23880bd348f0825ee72fd9489f9b41f3e6706a593b3dd8fd7da2ad1b72d· accessed 2026-08-30 · vendor-docGrounds: MATERIAL: skin MICRO-SURFACE on the live path -- pores and areola structure, generated not painted. WITNESSED: Our multi-channel displacement map, also known as channel packed map ... fully-complementary, meaning that all these frequencies are independent and unique across the channels; a ready to use high-quality 16k multi-channel displacement map; the 32bit float exr map is stored into the red-channel. The scan-vendor bar is DISPLACEMENT per frequency band at 16k -- the class the asset half answers procedurally with a 16-bit height channel per region and a separate micro band, since licensed scans are not the genome route. - [graham13] Graham, Tunwattanapong, Busch, Yu, Jones, Debevec, Ghosh. Measurement-Based Synthesis of Facial Microgeometry. Computer Graphics Forum 32(2), Eurographics 2013. Eurographics digital library record. publisher · read in our library
knowledge/fetched/cmp_graphics_graham13.html· pinhb6602e4b50938983a5113a75c542bfb27a2ed126690f0a29d2a3d98bb127f863· accessed 2026-08-30 · published-paperGrounds: MATERIAL: skin MICRO-SURFACE on the live path -- pores and areola structure, generated not painted. WITNESSED: a technique for generating microstructure-level facial geometry by augmenting a mesostructure-level facial scan with detail synthesized from a set of exemplar skin patches scanned at much higher resolution. The measured-statistics route: our nx_relief_lib bands are the statistics (amplitude, wavelength, pore pitch and orifice) rather than exemplar scans, which is the only form that survives a 1001-genome cast. - [disney23] Weiss, Moulin, Chandran, Zoss, Gotardo, Bradley. Graph-Based Synthesis for Skin Micro Wrinkles. Computer Graphics Forum 42(5), Symposium on Geometry Processing 2023. Disney Research Studios and Industrial Light and Magic. publisher · read in our library
knowledge/fetched/cmp_graphics_disney-microwrinkle23.pdf· pinh6aed7f4dab974a2ed2a885afcdb5202efffacfa4fb990285ff4877016de02775· accessed 2026-08-30 · published-paperGrounds: MATERIAL: skin MICRO-SURFACE on the live path -- pores and areola structure, generated not painted. WITNESSED in the pinned PDF text stream, never through a summarising fetcher: The simulation first samples pores on the skin and treats them as nodes in a graph; primary wrinkles are between 20 to 100 um deep, depending on age, the secondary wrinkles between 5 to 40 um deep; On the scalp, only 1-2 such wrinkles are observed in an area of 1mm width, on the forearm around 10; we target a resolution of 16k2, i.e. a mean pixel size of 0.02 mm. Our shipped bands (primary 60um, secondary 20um) sit inside those published depth bands, and the micro tile is pore-first as their graph is. - [rgca23] Saito, Schwartz, Simon, Li, Nam. Relightable Gaussian Codec Avatars. arXiv:2312.03704, CVPR 2024. publisher · read in our library
knowledge/fetched/cmp_graphics_rgca23.html· pinha69296554dee891faaa2a9d19b441deaeab23caadae7761e90d54f195005e448· accessed 2026-08-30 · published-paperGrounds: MATERIAL: skin MICRO-SURFACE on the live path -- pores and areola structure, generated not painted. FRONTIER, noted not adopted. WITNESSED: Our geometry model based on 3D Gaussians can capture 3D-consistent sub-millimeter details such as hair strands and pores on dynamic face sequences. Learned appearance reaches pore scale from capture of one subject; our cast is procedural from genome, so this is the ceiling to grade against, not the route. - [hravatar25] Zhang et al. HRAvatar: High-Quality and Relightable Gaussian Head Avatar. arXiv:2503.08224, 2025. publisher · read in our library
knowledge/fetched/cmp_graphics_hravatar25.html· pinhfb6750f7a909fcabf3a4ce141c75b9611f568671fa17c031040b6f103ab41747· accessed 2026-08-30 · published-paperGrounds: MATERIAL: skin MICRO-SURFACE on the live path -- pores and areola structure, generated not painted. FRONTIER, noted not adopted. WITNESSED: decomposes head appearance into several physical properties and incorporates physically-based shading to account for environmental lighting -- the 2025 splat-avatar line carries physical decomposition, which is the property our baked height plus derived normal also has by construction. - [tega25] Li, Gotardo, Bolkart, Garbin, Lattas, Beeler et al. TeGA: Texture Space Gaussian Avatars for High-Resolution Dynamic Head Modeling. arXiv:2505.05672, 2025. publisher · read in our library
knowledge/fetched/cmp_graphics_tega25.html· pinha7e684b4161583a42687a86c2728802124620ad76cca4c58b5e91427b91b8454· accessed 2026-08-30 · published-paperGrounds: MATERIAL: skin MICRO-SURFACE on the live path -- pores and areola structure, generated not painted. FRONTIER, noted not adopted. WITNESSED: capturing facial motion and other transient high-frequency features such as skin wrinkling -- texture-space splats that carry wrinkling as a transient, i.e. the learned counterpart of the tension response our lane names as ntr_tension_attenuate. - [nac-review21] Guiding Nipple-Areola Complex Reconstruction: Literature Review and Proposal of a New Decision-Making Algorithm. PMC8144123, 2021 (citing Sanuki, Fukuma, Uchida, Aesthetic Plastic Surgery 2009, 600 breasts). publisher · read in our library
knowledge/fetched/cmp_graphics_nac-review21.html· pinhd9921bab4383faec792727e438a924450084e386b6cb594086a635488f83b6cb· accessed 2026-09-05 · published-paperGrounds: MATERIAL: skin MICRO-SURFACE on the live path -- pores and areola structure, generated not painted. WITNESSED: A review of 600 reconstructed NAC reported an average diameter of 1.3 cm, a height of 0.9 cm and a 1:3 nipple-to-areola ratio; and: some authors recommend a 4-7 mm nipple diameter, a height of more than 1 cm and an areolar diameter of 4.2-4.5 cm. The ledger knowledge/areola_relief.conf takes nipple 13 mm and 9 mm from this and DERIVES the areola as 1:3 x 13 mm = 39 mm; the PubMed page for Sanuki itself came back as a bot-wall and is not cited as a mirror. - [nac-tattoo24] Fading and Color Reproducibility of Nipple-Areola Tattoos in Asian Patients. PMC11288766, 2024. publisher · read in our library
knowledge/fetched/cmp_graphics_nac-tattoo24.html· pinh336c769f4ae84e777d381fb8f62a6c43ccfb60635595c82daed8aea80a5d2d11· accessed 2026-09-05 · published-paperGrounds: MATERIAL: skin MICRO-SURFACE on the live path -- pores and areola structure, generated not painted. WITNESSED: Skin tone is defined as the mean value (R 209, G 172, B 146) of normal breast skin near the contralateral NAC preoperatively; Luminance = 0.298912R + 0.586611G + 0.114478B; the rate of tattooed/contralateral NAC luminance was mean 1.09 +- 0.13. The paper reports the NAC-to-skin relation only as figures, so the ledger's areola lightness step (L* 16) is DERIVED from this measured skin tone against the estate referee nx_areola_pigment's nulliparous-young band and is labelled DERIVED, not quoted. - [montgomery-sd] ScienceDirect Topics. Gland of Montgomery -- an overview (quoting Doucet et al. 2010b and Schaal et al. 2006), accessed 2026-08-30. publisher · read in our library
knowledge/fetched/cmp_graphics_montgomery-sd.html· pinh6a65c1d7fda2e8568f83422c0c56ecc77959aecb09f3ba33a8c85432ddbcd590· accessed 2026-08-30 · mirrored-indexGrounds: MATERIAL: skin MICRO-SURFACE on the live path -- pores and areola structure, generated not painted. WITNESSED: 97 percent of (Caucasian) women bore more than 1 MG per areola, and 83 percent, from 1 to 20 units per areola. The ledger's mont_count 10 is the midpoint of that band; Schaal 2006 itself (Developmental Psychobiology) is paywalled and not mirrored, so the count is cited through this overview and attributed to it. - [doucet09] Doucet, Soussignan, Sagot, Schaal. The Secretion of Areolar (Montgomery's) Glands from Lactating Women Elicits Selective, Unconditional Responses in Neonates. PLoS ONE 4(10) e7579, 2009. publisher · read in our library
knowledge/fetched/cmp_graphics_doucet09.html· pinhe30b909bd5d4f632ae1dbbf0992eb9a2c256151bda3bd1c23b31aa721f97b1e1· accessed 2026-08-30 · published-paperGrounds: MATERIAL: skin MICRO-SURFACE on the live path -- pores and areola structure, generated not painted. The primary the gland-count assessment is attributed to (the same Dijon group); WITNESSED in its pinned text it speaks of the mother's AG number and their nonrandom distribution on the areolae, while the count band itself is quoted through montgomery-sd because this paper's own text does not restate it. Cited for attribution and for the areolar-gland anatomy, never for a number it does not carry. - [hairswrast26] Lipp, Jarabo, Wimmer, Bode. Deferred Software Rasterization for Efficient Real-time Hair Rendering. Meta and TU Wien, arXiv:2607.04230v1, 5 July 2026. MEASURED on an NVIDIA RTX 5080 at 1920x1080 on a 127k-strand groom: the deferred software rasterizer takes 3.1 ms against 13.6 ms for the mesh-shader baseline at MSAA 1 without LOD, a 4.4x speedup; adding LOD drops the far-view frame from 4.5 ms to 0.4 ms, an 11x reduction, as strand count falls from 127k to 21k; strand generation plus software rasterization is about 96 percent of frame cost. Figure 1 renders multiple grooms totalling 483k strands at 4K resolution in 9.4 ms against 37.6 ms for the mesh shader at 1x MSAA. publisher · read in our library
knowledge/fetched/cmp_graphics_hairswrast26.pdf· pinh063e6cc22d48577094092722572ecbc083f22c149a62dbca58e3151be2b69d17· accessed 2026-09-04 · published-paperGrounds: HAIR: strand budget and frame cost against the published 2026 real-time bar - [hairlod25] Huang, Zhou, Lin, Zhu, Yan, Wu. Real-time Level-of-detail Strand-based Rendering. Eurographics Symposium on Rendering 2025, Computer Graphics Forum volume 44 number 4. LIGHTSPEED, UC Santa Barbara and Nvidia. MEASURED on an NVIDIA RTX 3090 with an AMD Ryzen ThreadRipper 3970X and 256 GB of memory: as hair screen occupancy falls from 50 percent to 0.3 percent the LoD rasterizer moves from a 0.99x slight overhead at 10.3 ms against 10.2 ms, through about 2x at the 10 percent middle view, to a 13x improvement at 0.4 ms against 5.3 ms at the far view, with a constant 0.1 ms per frame to reorganise hair data for the next LoD level. publisher · read in our library
knowledge/fetched/cmp_graphics_hairlod25.pdf· pinh4f22d84adbed087fad6c632b1821ab93f374d518ab084011c5374a2946b31fd1· accessed 2026-09-04 · published-paperGrounds: HAIR: strand budget and frame cost against the published 2026 real-time bar - [mhlodspec] Epic Games. Platform Support and LOD Specifications for MetaHumans. The published per-LOD table, read from the pinned mirror at byte offsets 31280 34880 and 45980: Head Vertices 24000 at LOD0 then 12000 6000 2500 1300 560 270 130; Blendshapes 669 at LOD0 only; Joints 713 529 397 283 84 70 41 26; Skin Influences 12 12 12 8 8 8 4 4; Body Vertices 30500 7600 3350 1507; Hair Style Strands 50000 at LOD0 and 25000 at LOD1 with no strand tier below that; Hair Style Card Vertices 30000 15000 10000 3000 1500; Facial Hair Strands 10000 and 5000; Facial Hair Card Vertices 15000 7000 3000 1000 500; Physics on LOD0 to LOD2 only. Platform table: PC at Epic and Cinematic quality supports strands and cards with max texture 8192, Mac is cards only, iOS and Android are cards only with best LOD 3 and max texture 2048. publisher · read in our library
knowledge/fetched/cmp_graphics_mhlodspec.html· pinh6258c72259656e03f2889da55d4db01c7dadf98d80c946e627f442c952b717ab· accessed 2026-09-04 · vendor-docGrounds: HAIR: strand budget and frame cost against the published 2026 real-time bar - [truetotone26] Schneider, Menezes, Mecenas, Knob, Araujo, Musse. True to Tone -- Quantifying Skin Tone Fidelity and Bias in Photographic-to-Virtual Human Pipelines. PUCRS and collaborators, arXiv:2604.02055v1, 2 April 2026. MEASURED over 19848 rendered MetaHuman images in Unreal Engine against Chicago Face Database references, scored in CIELAB by Delta E CIE76 and by Individual Typology Angle error. Extraction-method medians: Cheek 12.69, MMM 9.67, T-Cheek 4.90, T-MMM 3.77. Lighting medians: CFD light Delta E 2.2 to 4.5, Paramount 5.8 to 14.6, Frontal 7.6 to 18.7. ITA error is flat for classes I and II and strongly non-linear below: class V rises from a median 4.14 under CFD light to 31.32 under Paramount and 39.51 under Frontal, class VI from 11.37 to 46.61 to 53.85. The confusion matrix over all 19848 images shows darker phenotypes rarely preserved and compressed toward lighter classes. ITA class bounds after Chardon: I above 55, II 41 to 55, III 28 to 41, IV 10 to 28, V -30 to 10, VI below -30. publisher · read in our library
knowledge/fetched/cmp_graphics_truetotone26.pdf· pinhb1a8dbbca700505b4f602237da7e9b608f6ec204e95562a7199960bb3b115c38· accessed 2026-09-04 · published-paperGrounds: SKIN: skin-tone fidelity as CIELAB error and ITA-class preservation - [polbssrdf25] Ha, Hwang, Monzon, Cho, Kim, Baek, Munoz, Gutierrez, Kim. Polarimetric BSSRDF Acquisition of Dynamic Faces. KAIST, Universidad de Zaragoza and POSTECH, arXiv:2501.01980. A polarimetric multispectral capture rig of two polarization machine vision cameras synchronised at 20 fps, each behind a different multispectral filter, with 40 LEDs in eight modules of five in a near-coaxial setup and stereo modules for dynamic geometry, fitting spatially varying single scattering and refractive index. Validated against a hyperspectral camera and against ten real-world objects of known refractive index at a mean reconstruction error of 0.028; the polarimetric loss term with spatially varying single scattering yields the minimum RMSE on the specular-dominant polarization observation, and the smallest RMSE on specularity and full rendering across 200 poses of 11 participants. publisher · read in our library
knowledge/fetched/cmp_graphics_polbssrdf25.pdf· pinhcb3d1fe81b00b4b9585ec6b4863bc49d2b7a1a056d129162fae889757eb4aad8· accessed 2026-09-04 · published-paperGrounds: SKIN: measured ground-truth BSSRDF as the calibration oracle - [arxiv_rtgi_2026_09] arXiv API listing fetched and read 2026-09-06: search real-time path tracing OR real-time global illumination OR ReSTIR OR radiance caching, sorted by submission date descending, 30 of 32 results; no September 2026 entry; newest Split Radiance Cascades (2607.20384, 2026-07-22: real-time 3D diffuse GI from world-space sparse radiance probes with ray splitting), then HoloPathTracer (SIGGRAPH 2026, TOG 45.4), OctaOctree Neural Radiosity for real-time glossy GI (2606.08469), Megakernel vs Wavefront GPU path tracing (2605.27323: wavefront 16 percent faster on RT cores), ToF ReSTIR (2605.11536) publisher · read in our library
knowledge/fetched/cmp_gameengine_arxiv_rtgi_2026_09.xml· pinh058f51b4ac854a68a54f45ec979d11d0b2b5631f4c8318d202464dbe7d401d5e· accessed 2026-09-06 · mirrored-indexGrounds: Path-traced GI - [arxiv_avatar_2026_09] arXiv API listing fetched and read 2026-09-06: search relightable avatar OR photorealistic avatar OR photorealistic human OR (subsurface scattering AND real-time), sorted by submission date descending, 30 of 78 results; no September 2026 entry; newest STEER steerable dyadic head avatars (2607.23840, 2026-07-26: a causal flow-matching motion prior driving a Universal Gaussian Head-Avatar Prior), How to Build Digital Humans, the Eurographics 2026 STAR (2607.04341: priors, personalised avatar and animation as the three stages), Generative Relightable Avatars (2606.22718: UV-space material optimisation, feed-forward texture refinement, video diffusion), D-Rex diffusion relighting (2604.27871), generalizable human Gaussian splatting (2604.25466, CVPR 2026) publisher · read in our library
knowledge/fetched/cmp_gameengine_arxiv_avatar_2026_09.xml· pinh1a73ba1a7a0dfee552cb666b0a07a9ec7ed2c50f964787722f613947ff3af281· accessed 2026-09-06 · mirrored-indexGrounds: Pre-integrated SSS - [arxiv_genstream_review_20260919] GenStream: Semantic Streaming Framework for Generative Reconstruction of Human-centric Media; abstract reviewed2026-09-19, submitted2026-09-16; metadata streaming with client-side generative reconstruction, partial skating evaluation, client compute/generalization limitations; not anatomical ground truth publisher · read in our library
knowledge/fetched/rf_f913e989.raw· pinha91525e1d400f058bfc157206feb045b2af0110fd289363a1354a2c039e29d65· accessed 2026-09-19 · primary-abstractGrounds: Human rendering and semantic streaming - [kesen_sig2026] Ke-Sen Huang's SIGGRAPH 2026 papers index (Los Angeles, 19-23 July 2026): 288 dt-B entries counted by the estate on the mirrored bytes; sessions read for humans, hair, cloth, sky, splats and VR publisher · read in our library
knowledge/fetched/cmp_graphics_kesen_sig2026.html· pinh7e6c166c5ae0fca3a235a74073fb3d85e055ca46fe59d7b20e3abd2519bb206b· accessed 2026-09-27 · mirrored-indexGrounds: SB-VENUES-2609 - [kesen_eg2026] Ke-Sen Huang's Eurographics 2026 papers index: 108 entries counted on the mirrored bytes; Neuralocks real-time neural hair (Disney Research), SkinCells sparse skinning, CANRIG neural face rigging, GTAvatar relightable editable Gaussian avatars (ETH), a texture-free woven-fabric model (Manchester), neural garment fold inter-reflection (POSTECH), a neural irradiance volume publisher · read in our library
knowledge/fetched/cmp_graphics_kesen_eg2026.html· pinh92ca15926547daecb0f36c2fa3af9948295dc6a7fbebe807a4f74aaadf6e029b· accessed 2026-09-27 · mirrored-indexGrounds: SB-VENUES-2609 - [kesen_siga2026] Ke-Sen Huang's SIGGRAPH Asia 2026 papers index (Kuala Lumpur, 1-4 December 2026): 7 conditionally accepted entries on the mirrored bytes, a partial list that grows until the proceedings publisher · read in our library
knowledge/fetched/cmp_graphics_kesen_siga2026.html· pinhb745c68e9eda411a35406d31894a14300abb2ef2c563745d38868dbe88c50119· accessed 2026-09-27 · mirrored-indexGrounds: SB-VENUES-2609 - [arxiv_co_siggraph2026] arXiv API venue-tag scan: comment field phrase SIGGRAPH 2026, newest first, 85 entries, newest 2026-08-31, digested by nx_arxiv_digest: HairLRM 2606.15238, a delighting prior for facial appearance capture 2605.05636, EchoAvatar 2605.28272, HairGPT 2605.08824, BodyReLux full-body video relighting 2605.21766, Pixel Cube portrait relighting 2606.02919, FIT try-on dataset 2604.08526, sewing patterns via latent flow matching 2601.17740, MotionBricks 2604.24833, ARDY 2607.08741, YASPS IPC 2605.23088, penetration-free elastodynamics 2512.12151, faster 3DGS densification 2604.28016, CAGS 2605.09279, Gabor Fields 2602.05081 publisher · read in our library
knowledge/fetched/cmp_graphics_arxiv_co_siggraph2026.xml· pinha5eae48d6f540067918910978e53b083ba123cf01265d4b600c1ad1e52c40211· accessed 2026-09-27 · mirrored-indexGrounds: SB-VENUES-2609 - [arxiv_co_siga2026] arXiv API venue-tag scan: comment field phrase SIGGRAPH Asia 2026, newest first, 55 entries, newest 2026-09-24: Heartian physiology-aware relightable Gaussian head avatar 2609.28539, OmniFabric UV texture synthesis for garment reconstruction 2609.30234, Gaussian Light Transport 2609.11430, RelightFormer 2609.07414, Shedding Light lighting benchmark 2609.10787, InstantMimic 2609.09821, CHIP 2609.06591, S4R interpenetration resolution 2609.20524, real-time neural hair G-buffer anti-aliasing 2605.17557, spectral rendering without a spectral renderer 2602.18741, an avatar-distortion intent study 2609.27560 publisher · read in our library
knowledge/fetched/cmp_graphics_arxiv_co_siga2026.xml· pinh51d49183aafe7aa1b0dbf264bc0c029b4716c07ea617c03a35fa2b58286599df· accessed 2026-09-27 · mirrored-indexGrounds: SB-VENUES-2609 - [arxiv_co_eg_family_2026] arXiv API venue-tag scan over the Eurographics 2026, EGSR 2026, HPG 2026, I3D 2026 and SCA 2026 comment phrases, 22 entries, newest 2026-08-19: the digital-humans STAR 2607.04341, GTAvatar 2512.09162, multi-view face and gesture animation with dynamic Gaussians 2608.04722, GRay ray-traced Gaussians 2606.30869, exact reduced Coulomb friction 2607.19599, a neural irradiance volume 2602.12949, VkSplat 2605.00219, Robo-Saber VR playtesting 2602.18319 publisher · read in our library
knowledge/fetched/cmp_graphics_arxiv_co_eg_egsr_hpg_i3d_sca_2026.xml· pinh90cec84d63d1566d2f462a0cad702bd0c4e3e47a82e01a8ecf9f5878c09732ba· accessed 2026-09-27 · mirrored-indexGrounds: SB-VENUES-2609 - [arxiv_co_cvpr2026_human] arXiv API venue-tag scan: CVPR 2026 comment phrase AND avatar or human or garment or hair or skin or relighting, 120 entries at the fetch cap, newest 2026-09-03: HumanNOVA single-image photoreal avatar 2606.02573, Archon unified digital human generation 2605.30311, PhysHead simulation-ready Gaussian head avatars 2604.06467, AvatarPointillist 4D Gaussian avatars 2604.04787, MeshLAM 2604.22865, GeoRelight 2604.20715, TokenLight 2604.15310, generalizable human Gaussian splatting 2604.25466, high-fidelity mobile avatars with pruned local blendshapes 2605.01854, AvatarMix outfit personalization 2606.03506, FreeForm reduced-order deformables 2605.29318, a complex 4D markerless capture dataset 2604.12765 publisher · read in our library
knowledge/fetched/cmp_graphics_arxiv_co_cvpr2026_human.xml· pinhb4b8b9bbbb66992d6344463bdc7fe37983b97af5724709621f31c110b9e96e5c· accessed 2026-09-27 · mirrored-indexGrounds: SB-VENUES-2609 - [advances_rtr_2026] SIGGRAPH 2026 Advances in Real-Time Rendering in Games course page, read from the mirrored bytes: seven talks, ORCA online radiance cache acceleration (EA SEED), the PSSR upgrade on PlayStation 5 Pro (Sony), variable rate ray tracing (Activision), Smolder volumetrics (IO Interactive), SLIM device-adaptive worlds (Roblox), screen-adaptive Catmull-Clark tessellation (Meta), the 2026 trends talk (Activision); the token subsurface occurs zero times on the page publisher · read in our library
knowledge/fetched/cmp_graphics_advances_rtr_2026.html· pinh60e636e3eacf804dcad58014215527a21a903a441775eb46a176f6af7c0b66bd· accessed 2026-09-27 · published-courseGrounds: SB-ADVANCES-2607 - [nvidia_siggraph2026_blog] NVIDIA's SIGGRAPH 2026 announcement: MotionBricks real-time motion model, GPC generative controller, ArtiFixer capture cleanup with photoreal GI prediction, VideoNeuMat relightable materials from generative video, ARDY, the Newton physics engine with new solvers for snow, sand and elastic solids, Cosmos 3, Earth-2, and a synthetic-video detector NIM publisher · read in our library
knowledge/fetched/cmp_graphics_nvidia_siggraph2026.html· pinhfe0809948d5a5de467f3e88fdd7da457e01f254131164537a6c733365d0e1a45· accessed 2026-09-27 · vendor-docGrounds: SB-VENUES-2609 - [siggraph2026_awards] ACM SIGGRAPH blog, May 2026: the SIGGRAPH 2026 Technical Papers awards, best paper Robust Planar Maps for 3D Vectorization (Carnegie Mellon), honorable mention Spatiotemporal FLIP (TU Munich and RWTH Aachen), four test-of-time awards from the 2014-2016 conferences publisher · read in our library
knowledge/fetched/cmp_graphics_siggraph2026_awards.html· pinh69c08802dec950605b4f2bf2f1f8412ec28ed4ebf5c4c00d2eed0bc67e11c65e· accessed 2026-09-27 · vendor-docGrounds: SB-VENUES-2609 - [arxiv_siga_rtgi_2026_09_27] arXiv API venue-tag listing fetched and read 2026-09-27 by the paced nx_frontier_scan fan-out: comment field SIGGRAPH Asia 2026 AND real AND time AND global AND illumination, 1 result, Gaussian Light Transport (Attimont, Subr, Soler, 2609.11430, published 2026-09-10, SIGGRAPH Asia 2026 conference track): global illumination expressed as a 13D Gaussian mixture over positions, directions, normals and material properties, parameters estimated by minimising the rendering-equation residual instead of a Neumann series, a culling strategy keeps optimisation tractable, view-independent solutions render in milliseconds at a fraction of the memory of neural rendering publisher · read in our library
knowledge/fetched/cmp_graphics_siga_rtgi_202609.xml· pinh4a6d43bf080dccf577e199f3e26705cb9f0dfbdf039cd62032d13f1151a31ed7· accessed 2026-09-27 · mirrored-indexGrounds: SB-RTGI - [itu_bt2124_2019] ITU-R Recommendation BT.2124-0 (01/2019), Objective metric for the assessment of the potential visibility of colour differences in television, mirrored and read 2026-10-01 through nx_pdf_text: defines DeltaE ITP from display-referred ICtCp (Recommendation BT.2100) for small colour differences in television images and signals, scaled so that 1 indicates the potential of a just noticeable difference; it presumes the most sensitive state of viewer adaptation so it will not under-predict perceived differences but may over-predict them; its considerings state that existing colour difference metrics assume a state of viewer adaptation (white point, illumination level, viewing environment) while in television that state is variable and often unknown; for scene-referred signals it applies only when a nominal peak display luminance is assumed, otherwise it gives an ordinal measure of distortion publisher · read in our library
knowledge/fetched/cmp_graphics_itu_bt2124_2019.pdf· pinhcad1e382516da511811c5205e5abe1b0b97999cf4722d4854a439389eff54470· accessed 2026-10-01 · published-standardGrounds: GR62 - [sharma_ciede2000_2005] Sharma, Wu and Dalal 2005, The CIEDE2000 color-difference formula: implementation notes, supplementary test data, and mathematical observations, Color Research and Application 30(1) 21-30, mirrored from the first author's Rochester page and read 2026-10-01 through nx_pdf_text: CIEDE2000 was published by the CIE in 2001, developed by Technical Committee 1-47 as an improved procedure for the computation of industrial color differences; several implementation errors pass the CIE worked examples and are caught only by the supplementary test pairs this paper publishes; the formula has three independent sources of mathematical discontinuity, the largest from the mean-hue computation; the paper states it does not address psychophysical evaluation, uniformity or the formula's domain of applicability publisher · read in our library
knowledge/fetched/cmp_graphics_sharma_ciede2000_2005.pdf· pinh903c2838342097fb130a7fdd66a8f5f17d5448d070f9c15cae7e67c358e34410· accessed 2026-10-01 · published-paperGrounds: GR62 - [cdflow_cvpr2023] Chen, Wang, Yang, Sun and Ma 2023, Learning a Deep Color Difference Metric for Photographic Images (CD-Flow), CVPR 2023 (City University of Hong Kong, Shenzhen MSU-BIT University, Transsion, CUHK Shenzhen), mirrored from CVF open access and read 2026-10-01 through nx_pdf_text: handcrafted color difference metrics are subject-calibrated against uniformly colored patches and do not generalize to photographic images; on the SPCD photographic set CIE94 and CIEDE2000 show no noticeable improvement over the original CIELAB formula; pixel-by-pixel CIELAB formulae are particularly sensitive to slight geometric misregistration; on the TID2013 colour subset CD-Flow scores STRESS 14.110 and PLCC 0.837 against CIEDE2000 at 18.203 and 0.730, while on the uniform-patch COM data used to develop CIEDE2000 it underperforms CIE94 and CIEDE2000 publisher · read in our library
knowledge/fetched/cmp_graphics_cdflow_cvpr2023.pdf· pinhe36d30c49c302ab6ffd5ba63ea431eaa51cc0562eb7f3bfdf470b4c525153999· accessed 2026-10-01 · published-paperGrounds: GR62 - [beyondskintone_iccv2023] Thong, Joniak and Xiang 2023, Beyond Skin Tone: A Multidimensional Measure of Apparent Skin Color, ICCV 2023 (Sony AI and The University of Tokyo), mirrored from CVF open access and read 2026-10-01 through nx_pdf_text: measures apparent skin colour in images as perceptual lightness L* and hue angle h* instead of a single tone scale; ITA was originally developed for the suntanning pathways of Caucasian skin, its low correlation with expert annotations makes it less reliable, and because ITA and h* both contain b* the paper measures L* to keep the two axes uncorrelated; apparent colour as imaged differs from true constitutive colour by illumination and colour cast; skin-colour extraction builds on the Merler et al. Diversity in Faces algorithm generalised to any pose, with an illumination-robustness analysis publisher · read in our library
knowledge/fetched/cmp_graphics_beyondskintone_iccv2023.pdf· pinhc4fde64115f61ceae48e48e6c926168eccf62228d3314a64ee7d01868f5d3c63· accessed 2026-10-01 · published-paperGrounds: GR62 - [alipour_ita_limits_2026] Alipour, Burke and Courtney 2026, Limitations of ITA for Skin Type Estimation Under Uncontrolled Imaging Conditions, Skin Research and Technology 32(8) e70374, published 2026-08-11 under CC BY 4.0 (Technological University Dublin), full text mirrored from Europe PMC (PMC13459198) and read 2026-10-01: on a skin-patch set derived from the dermatologist-labelled PAD-UFES-20 images, fixed-threshold ITA classification reached 22.9 percent accuracy against a 51.3 percent majority-class baseline and failed the darkest skin type (FST VI); lighting variation, threshold misalignment, overlapping ITA distributions and dataset imbalance made ITA unstable, and the authors conclude ITA is highly sensitive to lighting variation and should not be read as a validated skin-type measurement in image datasets publisher · read in our library
knowledge/fetched/cmp_graphics_alipour_ita_limits_2026.xml· pinh49ca50d744476e6b959687116da13fe2843ac81efa5e406bc35a65e84c3afc4a· accessed 2026-10-01 · published-paperGrounds: GR62 - [epmc_tanning_202609] Europe PMC REST search for individual typology angle or facultative or constitutive pigmentation or delayed tanning or persistent pigment darkening, sorted by publication date descending, first 25 of 762 hits, read 2026-10-01 (UTC): the hits first published in September 2026 are off the dose-to-tan axis (an AI-discovered depigmenting saponin, blue-light fibroblast senescence, a multilayer head phantom, chest reflectance oximetry, a visible-light photodamage review, solar-simulated UV and the skin microbiome); the newest on-subject entries are August 2026: Raynaud et al. (L'Oreal Advanced Research, Skin Pharmacology and Physiology, first published 2026-08-01), a year-long real-life Paris study of 59 phototype II-III women in which in vivo multiphoton melanin density and skin colour rose from winter to summer in the vehicle arm and not under daily SPF and PPD photoprotection, and Alipour et al. on ITA limits under uncontrolled imaging publisher · read in our library
knowledge/fetched/cmp_graphics_epmc_tanning_202609.json· pinh2c30f16dbf2d4def7c08e6bb7b0c1646d19b85680cad3a9db0cf959a1c97877b· accessed 2026-10-01 · mirrored-indexGrounds: GR59 - [ppg2017_t1001] Mueller, Gross, Novak (ETH Zurich and Disney Research, as printed on the paper). Practical Path Guiding for Efficient Light-Transport Simulation. Computer Graphics Forum 36(4), EGSR 2017. Read 2026-10-01 from nx_pdf_text over the mirrored PDF: a robust unbiased technique that learns the scene spatio-directional radiance field iteratively in an SD-tree (a binary tree over space above a quadtree over direction), budgets training against rendering automatically, needs no hyperparameter tuning and is production compatible; on its TORUS figure at equal time it reaches MSE 0.018 against 0.104 for Vorba et al. publisher · read in our library
knowledge/fetched/cmp_graphics_mueller2017_ppg.pdf· pinh6cd7505646afb2baed5928168854a1cec946263112009910bfd33f5a127b055c· accessed 2026-10-01 · published-paperGrounds: GR42 - [schmalwieser_tan_2025] Schmalwieser, Schwabel, Philipsen, Heydenreich and Young 2025, Seasonal and lifelong changes in skin colour and pigmentation of Austrian farming families: an exploratory study, Photochemical and Photobiological Sciences (University of Veterinary Medicine Vienna and Copenhagen University Hospital dermatology), open full text mirrored from Springer and read 2026-10-01: tristimulus skin colour every 6 to 7 weeks from April 2011 to February 2012 at buttock and inner upper arm (constitutive), forehead (continuous exposure) and shoulder and outer upper arm (intermittent), with personal electronic UV dosimeters; degree-of-tan TAN is the ITA of buttock minus the ITA of the exposed site; the change in TAN between visits regresses linearly on daily mean erythemal radiant exposure and crosses zero at 1.4 SED per day in both farmer age groups, so above it pigmentation increases and below it fades; an obvious TAN persists through late winter at exposed sites, outer-arm TAN peaks July to September, the inner arm is not truly unexposed, and buttock skin matches inner arm in L*, b* and ITA but is redder by a* 1.9 publisher · read in our library
knowledge/fetched/cmp_graphics_austrian_farmers_tan_2025.html· pinh118e1a179c2912edd0b8a4122e63ed5a2f8b0109203adede899eb9a886617298· accessed 2026-10-01 · published-paperGrounds: GR59 - [arxiv_colordiff_202609] arXiv search listing, abstract phrase color difference, ordered newest submission first, first 50 of 189 results, read 2026-10-01 (UTC): the newest is 2609.39130 Perceptual Color Difference Modeling Using Machine Learning and Human Similarity Judgments (Kadyrgali, Muratbekova, Yerkin, Toganas, Igali, Ziyada, Burambekova, Hasanov, Shamoi, submitted 2026-09-30, submitted to IEEE Access), which states that metrics such as CIEDE2000 may still be inconsistent with perceptual judgments and fits regressors to 2000 generated colour pairs each rated by seven observers on a four-point scale (LightGBM R2 0.703 with fuzzy colour categories added); then 2606.12588v2 A conformally-Euclidean Line Element for evaluating color differences (De Visschere, Candry, Neyts, revised 2026-09-23), a Euclidean difference scaled by a luminance-only factor fitted to surface-colour datasets with CIELab-like coordinates; the other September hits are a web-opacity contrast paper and astronomy homonyms; no October 2026 entry publisher · read in our library
knowledge/fetched/cmp_graphics_arxiv_colordiff_202609.html· pinh74a91607a9540b22a3bb42f1fed0a13630fb0af17f50b0c7cfdebf1b41a84f1c· accessed 2026-10-01 · mirrored-indexGrounds: GR62 - [openpgl071_t1001] Intel Open Path Guiding Library README, release v0.7.1 (reference entry Herholz and Dittebrandt 2022), read 2026-10-01; a vendor claim: implementations of current state-of-the-art path guiding methods for production renderers, a guiding field learned during rendering and updated per frame from radiance samples, guiding directional sampling on surfaces and inside volumes from a learned incident radiance distribution or its product with BSDF cosine lobes or single-lobe HG phase functions, VMM-based guiding models, Apache 2.0, still pre v1.0 with an API in flux publisher · read in our library
knowledge/fetched/cmp_graphics_openpgl_readme_20261001.md· pinh47aac7d994b999ca20498d3daf09f0a45c1f73cca50629b86021ef6093b7c048· accessed 2026-10-01 · vendor-docGrounds: GR42 - [arxiv_skintanning_202609] arXiv search listing, abstract terms skin tanning, ordered newest submission first, read 2026-10-01 (UTC): only 3 results and none on the UV-dose-to-tan axis (a skin-lesion colour-histogram classifier, a non-Hermitian skin-localisation homonym, and leather-defect segmentation where tanning means leather processing), so this arXiv search surfaces no tanning-kinetics work and the current-month listing for this axis is the biomedical index (epmc_tanning_202609) publisher · read in our library
knowledge/fetched/cmp_graphics_arxiv_skintanning_202609.html· pinh21de0c0c20696a3043cddbd24679ea401e9d5d245c1cf3a28d3d4946dc1f2a32· accessed 2026-10-01 · mirrored-indexGrounds: GR59 - [restir_search_t1001] arXiv search, ReSTIR in all fields, newest announcement first, read 2026-10-01: only 3 results -- ToF ReSTIR: Time-of-Flight Rendering with Spatio-temporal Reservoir Resampling (announced May 2026, DOI 10.1145/3811299), Neural Visibility Cache for Real-Time Light Sampling (June 2025) and Decorrelating ReSTIR Samplers via MCMC Mutations (November 2022); the 2020 ReSTIR paper itself is not among them, so the family is read on its publishers' pages as well as on arXiv publisher · read in our library
knowledge/fetched/cmp_graphics_arxiv_search_restir_20261001.html· pinha8209042867b58d7010df6cfbc351237d31993a345ca75bb8932a0117821da4f· accessed 2026-10-01 · mirrored-indexGrounds: GR42 - [amano_naturaltan_2020] Amano K, Xiao K, Wuerger S, Meyer G 2020, A colorimetric comparison of sunless with natural skin tan, PLoS ONE 15(12):e0233816, CC BY 4.0, JATS XML read 2026-09-30: control group n=60, inner forearm (infrequently sun-exposed) against outer forearm (frequently sun-exposed) defined as the naturally occurring tan; Delta E out-inn mean 6.03 SD 2.64; outer-forearm ITA mean 38.8; spectral RMSE inn-out 0.050 over 400-700 nm; a natural tan is a decrease in lightness without an increase in yellowness, and its ITAs span 10 (outer) to 70 (inner) publisher · read in our library
knowledge/fetched/rf_803ff2ff.raw· pinhd24c54fcc23d79b3cf224bf228b222c32e1669bb1a5064a2622730b53e5a9515· accessed 2026-09-30 · published-paperGrounds: GR59 - [jacques_skinoptics_1998] Jacques S L 1998, Skin Optics Summary, Oregon Medical Laser Center News, read 2026-09-30: mua.skinbaseline 0.244 + 85.3 exp(-(nm - 154)/66.2) per cm; melanosome mua.mel 6.6e11 nm^-3.33 per cm; mua.epi = f.mel mua.mel + (1 - f.mel) mua.skinbaseline; f.mel 1.3-6.3 percent for light-skinned adults, derived assuming a 60 um epidermis and a photon path of twice its thickness; musp 2e12 nm^-4 (Rayleigh) + 2e5 nm^-1.5 (Mie, collagen fibres), musp.epi = musp.derm to first approximation; typical average f.blood 0.2 percent publisher · read in our library
knowledge/fetched/rf_bc9db52f.raw· pinh551d1255c089c85b7c780f258e9ad44c475677d816da2f529e3d6c0e587bb083· accessed 2026-09-30 · published-courseGrounds: GR59 - [cie_cmf1931_2deg] CIE 2019, CIE 1931 colour-matching functions, 2 degree observer, dataset CIE_xyz_1931_2deg.csv from CIE 018:2019 Table 6, 1 nm, 360-830 nm, fetched 2026-09-30 from the CIE data table page's own link (the bare files.cie.co.at path answered 404) publisher · read in our library
knowledge/fetched/rf_81b67599.raw· pinhfa663e3535a7e0763a745993a1f0a192eb0275ac46ad2d1befd7626841e713c1· accessed 2026-09-30 · datasetGrounds: GR59 - [cie_illum_d65] CIE, CIE standard illuminant D65, dataset CIE_std_illum_D65.csv, relative spectral power at 1 nm, 300-830 nm, fetched 2026-09-30 publisher · read in our library
knowledge/fetched/rf_15ee391b.raw· pinhe76f210bffff3d552ef7113025da5f325d5dfec200dd4b878b1a2f3a507032cb· accessed 2026-09-30 · datasetGrounds: GR59 - [cie_illum_a] CIE, CIE standard illuminant A, dataset CIE_std_illum_A_1nm.csv, relative spectral power at 1 nm, 300-830 nm, fetched 2026-09-30 publisher · read in our library
knowledge/fetched/rf_2566c378.raw· pinh61ef23fe146b8b665c74706717ab28cec7db6c9022993490bdc71991f43cb59b· accessed 2026-09-30 · datasetGrounds: GR59 - [csgr202609_gfx_t1001] arXiv cs.GR monthly listing for September 2026 (show 2000), read 2026-10-01: Total of 170 entries, titles read by term; the on-subject light-transport titles are RenderFormer-V2: Neural Rendering with Heterogeneous Scene Primitives (2609.05738) and Gaussian Light Transport (2609.11430, Attimont, Subr, Soler, SIGGRAPH Asia 2026); no title names ReSTIR, path guiding, path tracing or radiance caching; the October 2026 listing is cited separately publisher · read in our library
knowledge/fetched/cmp_env_arxiv_csgr_list_202609.html· pinha3a5f9952ccd7ebd84a5d21176eb8aa6619e7bd3aaa6265e6a3e748cb8dce9ef· accessed 2026-10-01 · mirrored-indexGrounds: SB-RTGI - [ravnbak_tankinetics_2008] Ravnbak MH, Philipsen PA, Wiegell SR, Wulf HC 2008, Skin pigmentation kinetics after UVB exposure, Acta Derm Venereol 88:223-228, doi 10.2340/00015555-0431, open access, read 2026-09-30 through nx_pdf_text: Table I dose to minimal pigmentation after ONE exposure per Fitzpatrick type (solar simulator 5.3, 7.5, 9.6, 10.5 SED for types II to V) with mean pre-exposure pigmentation 22.9, 26.7, 31.2, 46.3; Table II absolute steady-state pigmentation increase at 0.4, 0.6 and 0.8 MMD after 6 exposures (6.9, 6.9, 7.6) and 12 exposures (7.2, 9.0, 9.1), independent of skin type; the dose fraction to minimal pigmentation lowered by a factor 2 after 6 and 3 after 12 exposures; no tanning at 0.2 MMD; mean days to total fading 132 and 134 after 6, 137 and 175 after 12 exposures, fitted by linear regression; the highest pigmentation expected one week after the last exposure publisher · read in our library
knowledge/fetched/rf_e2ec0d52.raw· pinhacd1bc8a51fcaf8d2de21ea7950bfb0c7a08abb75a098aebfd0bd1d4d53f3450· accessed 2026-09-30 · published-paperGrounds: GR59 - [csgr202610_gfx_t1001] arXiv cs.GR monthly listing for October 2026 (show 2000), read 2026-10-01 at 02:56 UTC (fetch epoch 1790823400): the page reads No updates for this time period, so September 2026 is the newest month with entries publisher · read in our library
knowledge/fetched/cmp_env_arxiv_csgr_list_202610.html· pinh01a04d5de95fd9dc53e21708920ecfe9cd67ab44a532dcd7860903b1908e764f· accessed 2026-10-01 · mirrored-indexGrounds: SB-RTGI - [arxiv_skinseg_202609] arXiv search listing, abstract phrase skin segmentation, ordered newest submission first, 21 results, read 2026-10-01 (UTC): the newest is SkinMap: Weighted Full-Body Skin Segmentation for Robust Remote Photoplethysmography 2510.05296 (Maleki, Akbari, Binesh, Khalaj, submitted 2025-10-06), which selects and weights skin regions all over the body before extracting the colour signal; the rest are older learned or colour-rule skin segmenters (3D medical skin segmentation, weakly supervised and colour-invariant human skin segmentation, real-time facial skin-tone grading) plus lesion and off-subject hits; the listing holds no 2026 entry publisher · read in our library
knowledge/fetched/cmp_graphics_arxiv_skinseg_202609.html· pinha3fc59593cd55ad2ecdc2ac450f1ad9d5d1da377d5f44724e7c7db78ff7c1464· accessed 2026-10-01 · mirrored-indexGrounds: GR62 - [sharma_ciede2000_testdata] Sharma, Wu and Dalal 2005, supplementary test data for The CIEDE2000 color-difference formula (Color Research and Application 30(1) 21-30): the first author's data file ciede2000testdata.txt from his Rochester page, mirrored 2026-10-01 through nx_research_fetch and read from the mirrored bytes: 34 rows of seven tab-separated columns, L1 a1 b1 L2 a2 b2 and the reference dE00 to four decimals, with the signs printed (the paper's PDF text layer drops the minus glyph, so this file is the signed witness for its Table I) publisher · read in our library
knowledge/fetched/cmp_graphics_sharma_ciede2000_testdata.txt· pinh44aebb39107128328add54fbef5ac8ee89909e50508f448a1580adea2058a4b8· accessed 2026-10-01 · datasetGrounds: GR62 - [restirgi2021_bf1001] Ouyang, Liu, Kettunen, Pharr, Pantaleoni (all NVIDIA Corporation). ReSTIR GI: Path Resampling for Real-Time Path Tracing. Computer Graphics Forum 40(8), High-Performance Graphics 2021. Paper PDF linked from NVIDIA Research's publication page, mirrored 2026-10-01 (165,410,530 B; the page itself is knowledge/fetched/cmp_graphics_restirgi2021_page.html) and its title block, teaser and abstract read through nx_pdf_text: screen-space spatiotemporal resampling of multi-bounce indirect lighting paths obtained by path tracing, built on the ReSTIR principles of Bitterli et al. 2020, sharing important paths across time and pixels; at one sample per pixel every frame it improves MSE by 9.3x to 166x over path tracing in the test scenes; the teaser is an equal-time comparison at 1080p on an RTX 3090 without denoising, path tracing 8.0 ms against ReSTIR GI 8.9 ms with MSE improved 15:1; with a denoiser it reaches real-time path-traced global illumination publisher · read in our library
knowledge/fetched/cmp_graphics_restirgi2021_paper.pdf· pinh52a748cb107850036633e7e948adf364e174fc4e423f7329f5a850986464dac5· accessed 2026-10-01 · published-paperGrounds: GR42 - [gris2022_bf1001] Lin (University of Utah) and Kettunen (NVIDIA) as joint first authors, Bitterli, Pantaleoni (NVIDIA), Yuksel (University of Utah), Wyman (NVIDIA). Generalized Resampled Importance Sampling: Foundations of ReSTIR. ACM TOG 41(4) Article 75, July 2022, DOI 10.1145/3528223.3530158. Preprint PDF linked from NVIDIA Research's publication page, mirrored 2026-10-01 (53,532,327 B; the page is knowledge/fetched/cmp_graphics_gris2022_page.html) and its abstract, introduction and Fig. 2 read through nx_pdf_text: ReSTIR's iterative reuse introduces correlation, so it loses most convergence guarantees RIS theory provides; GRIS extends RIS to correlated samples with unknown PDFs taken from varied domains, giving variance bounds and convergence conditions for ReSTIR-based samplers; Fig. 2 shows unbiased ReSTIR reuse converging to a wrong result because reused samples are duplicates; ReSTIR PT reuses paths between pixels through complex shift mappings, interactive on complex scenes with many-bounce diffuse and specular lighting while shading one path per pixel, and can be modified to guarantee convergence for offline renderers publisher · read in our library
knowledge/fetched/cmp_graphics_gris2022_paper.pdf· pinh38b86bffc20f61d6984b868be3f5a630b9d3d93a92c106896133799188a33f00· accessed 2026-10-01 · published-paperGrounds: GR42 - [arearestir2024_bf1001] Zhang (University of Utah) and Lin (NVIDIA) as joint first authors, Kettunen (NVIDIA), Yuksel (University of Utah), Wyman (NVIDIA). Area ReSTIR: Resampling for Real-Time Defocus and Antialiasing. ACM, DOI 10.1145/3658210 as linked from NVIDIA Research's publication page (knowledge/fetched/cmp_graphics_arearestir2024_page.html). Paper PDF mirrored 2026-10-01 (217,015,288 B); the abstract read on the publication page and the title block and Fig. 1 read through nx_pdf_text, which recovered little of the body from this file, so the body is unread: ReSTIR methods implicitly assume a pinhole camera and evaluate a pixel through one predetermined subpixel location, which prevents efficient path reuse in and near pixels with high-frequency detail; Area ReSTIR extends reservoirs to integrate each pixel's 4D ray space over film and lens, with subpixel-tracking temporal reuse and new shift mappings, importance sampling subpixel and lens coordinates for antialiasing and depth of field; Fig. 1 times path tracing at 1 spp 4.2 ms, ReSTIR PT 9.5 ms and Area ReSTIR 14.0 ms publisher · read in our library
knowledge/fetched/cmp_graphics_arearestir2024_paper.pdf· pinhf204154cc16a236cbafb5a1b73d167c953476282c71bb62f58381dc2e37cdb1f· accessed 2026-10-01 · published-paperGrounds: GR42 - [gr59_ma_fudan_2024_uvtime] Ma Y, Tan Y, Hu Y, Pu W, Xu J, Jin L and Wang J (Fudan University School of Life Sciences and Human Phenome Institute; Huashan Hospital; Shanghai Skin Disease Hospital; CAMS research unit 2019RU058), Quantitative Assessment of Ultraviolet-Induced Erythema and Tanning Responses in the Han Chinese Population, Phenomics 2024 4(2):138-145 (PMC11169260, CC BY), FULL TEXT read 2026-10-01 from the Europe PMC JATS XML: six gradient UV doses on the back skin of 31 volunteers, erythema index and melanin index measured for 14 consecutive days; the erythema index peaks on D1 and the melanin index on D7 regardless of dose (p below 0.001), the melanin-index variance at D7 more than twice baseline; no blisters at 45 mJ/cm2; the D1 and D7 increments at 45 mJ/cm2 track the erythema and melanin dose-response slopes at R2 above 0.9 and were confirmed in a 664-person cohort. A tanning time course from a group outside Copenhagen publisher · read in our library
knowledge/fetched/cmp_graphics_bfh_ma2024_hanchinese_tanning.xml· pinhbf16a7b572f6b21849ca12b9a7c671649818f9d22ad8982bcaefca0f8985e3cb· accessed 2026-10-01 · published-paperGrounds: GR59 - [gr59_coelho_nihfda_2009_llp] Coelho SG, Zhou Y, Bushar HF, Miller SA, Zmudzka BZ, Hearing VJ and Beer JZ (Laboratory of Cell Biology, National Cancer Institute, NIH, Bethesda; Center for Devices and Radiological Health, US FDA, Silver Spring), Long-Lasting Pigmentation of Human Skin, a New Look at an Overlooked Response to UV, Pigment Cell Melanoma Res 2009 22(2) (PMID 19226313, PMC2709973, NIH author manuscript), FULL TEXT read 2026-10-01 from the PMC JATS XML: subjects of six phototypes exposed to multiples of their MED; pigmentation strong within days stays robust on post-exposure day 16, and a subset keeps a visible grade of at least 1 for 9 months or more (long-lasting pigmentation); 32 percent of subjects showed it after 2 MED or more and 61 percent after 4 MED; high MED and MMD values were its best predictors. The slow tail of tanning kinetics from a group outside Copenhagen publisher · read in our library
knowledge/fetched/cmp_graphics_bfh_coelho2009_longlasting_pigm.xml· pinh24e4508982b94f9898a1398965491569d8f77969dacd648855a9cad0ea033bcf· accessed 2026-10-01 · published-paperGrounds: GR59 - [gr59_cie_eilv_17_26_069_sed] CIE, International Lighting Vocabulary (CIE S 017) term 17-26-069 standard erythema dose (SED), the CIE's own published definition, read 2026-10-01 from the mirrored e-ILV page: standardized unit of measure of erythemal radiation; Note 1 to entry: one standard erythema dose (1 SED) is equivalent to an erythemal radiant exposure of 100 J per square metre; Note 3 refers to ISO/CIE 17166 Erythema Reference Action Spectrum and Standard Erythema Dose; Note 4: the entry was numbered 17-1255 and 17-1151 in CIE S 017:2011 publisher · read in our library
knowledge/fetched/cmp_graphics_bfh_cie_eilv_17_26_069_sed.html· pinh5e809d7c6d3b8622894c32a9437d96e58f4aa1416db3580918056ce85fa421d9· accessed 2026-10-01 · published-standardGrounds: GR59 - [gr59_cie_pub_17166_2019] CIE, ISO/CIE 17166:2019 Erythema reference action spectrum and standard erythema dose, the CIE publication page read 2026-10-01: prepared by the CIE with ISO/TC 274; this first edition cancels and replaces ISO 17166:1999 and CIE S 007-1998 as a minor revision; the document specifies the erythema reference action spectrum s_er(lambda) and the standard erythema dose (SED), proposes SED as the standardized measure of erythemogenic UV radiation and reserves MED for observational studies; 12 pages sold through the CIE Webshop, so the standard text itself is NOT mirrored; the ISO catalogue page (Edition 1, 2019-05, status Published) is mirrored beside it at knowledge/fetched/cmp_graphics_bfh_iso_17166_page.html publisher · read in our library
knowledge/fetched/cmp_graphics_bfh_cie_sed_page_b.html· pinha7f034245966e23086939bfe2d71f63129ad84e8832d3acf0efbdec325dd2ab6· accessed 2026-10-01 · published-standardGrounds: GR59 - [worldspace_restir_amd2021_bf1001] Boisse (single author). World-Space Spatiotemporal Reservoir Reuse for Ray-Traced Global Illumination. SIGGRAPH Asia 2021 Technical Communications, Tokyo, DOI 10.1145/3478512.3488613. PDF from AMD's GPUOpen publications site, mirrored 2026-10-01 (33,559,971 B); the affiliation line was not in the extracted text, so the group is identified by the host and by an author set disjoint from the NVIDIA papers. Abstract and opening introduction read through nx_pdf_text: image-space reservoir reuse does not scale to light sampling at further bounces, so reservoirs of path vertices are cached in the cells of a hash grid built entirely on the GPU, allowing stochastic reuse of neighbouring reservoirs across space and time and efficient light sampling at arbitrary vertices along the eye path; it cites Bitterli et al. 2020 and Wyman and Panteleev 2021 as the screen-space line it extends publisher · read in our library
knowledge/fetched/cmp_graphics_worldspace_restir_amd2021.pdf· pinhb71e275b81e0b88c77646955d46a27cf761663528677f8c08bfb9d8c36ac0c1f· accessed 2026-10-01 · published-paperGrounds: GR42
generated by nx_swcompare_matrix (sovereign NishiLang organ) from knowledge/compare/graphics.matrix · source checks show implementation presence; runtime and user-outcome evidence are reported separately · JavaScript supports page controls