Nishi Family › Compare › Sovereign Content Creation Studio

Nishi Compare · capabilities, evidence and delivery

Sovereign Content Creation Studio

Compare capabilities and delivery evidence. Source discovery, qualification, deployment and accepted outcomes are separate; an open contract names work still to verify.

Nishi vs Blender 5.0 and ZBrush 2026 and Maya 2026 and ArmorPaint -- the authoring tool itself, first byte up in NishiLang, running in the visitor's browser with no installation. Measured 2026-08-25 by a full-population census of 46 authoring axes: 20 are ABSENT-PROVEN and 26 are grounded in a named function. The interchange and rig halves are genuinely strong -- six importers, three exporters, skeleton with linear-blend skinning, IK, keyframes, morph targets and real mocap retargeting. The solid-modelling half is strong but SDF-native -- robust booleans, marching tetrahedra, two decimators, and a solid that re-evaluates from its own operation chain. The direct mesh-editing, UV, sculpt-workflow and interactive-authoring halves are close to empty, and everything that exists in UV and texturing is anatomy-locked to the humanoid layout. The distance to a studio is the editor, not the geometry -- and the first two rungs of that editor LANDED TODAY, gate-proven and mutation-proven.

Overview

The purpose, declared position and evidence coverage of this domain. Source presence and completed acceptance are different measures.

Where we are. Measured 2026-08-25, not asserted. WORKING, and more than the name of any one organ suggests: a mesh type with real DCC verbs on it (translate, scale, subdivide, bounds), a ZBrush-class sculpt brush over curved primitives, N-primitive constructive solid geometry with field evaluation and surface extraction, OBJ and STL and glTF emitters, a skeleton with linear-blend skinning and forward kinematics, real mocap decoded from BVH and VMD and retargeted, an anatomy-derived UV unwrap with regions checked against sums, a deterministic albedo and full PBR map bake that REGENERATES per genome rather than shipping a file, a spring-damper secondary-dynamics solver with its stability envelope enforced in-solver, a declarative constraint part-solver that refuses an inconsistent set by name, and an exact point-to-triangle millimetre deviation ruler over every vertex. UPDATED 2026-08-25 AFTER FIVE RUNGS LANDED, because a position statement nobody re-measures becomes a claim nobody trusts. WHAT THIS LINE SAID THIS MORNING AND NO LONGER DOES: there is NO EDITOR, nothing records an operation, nothing undoes one, nothing replays a session. THERE IS NOW. Landed and gate-proven on the promoted binaries, every one of them mutation-proven with killed=1 and not_reached=0: the non-destructive operation stack where undo, redo and re-evaluation are all REPLAYS FROM THE BASE so state is bit-identical (44/44); the mesh-validity referee that REFUSES rather than showing an overlay, six named classes each with a first-offender index (32/32); ray-pick and a transform gizmo where a drag resolves to ONE operation on the stack, three separate teeth prove picking never wrote a word of the mesh buffer, and the pick's COST is now published at four stated triangle counts against the frame budget the estate already banks -- 455054 ns over 2048 triangles, 4.1 percent of a 90 Hz frame, measured SUB-linear in triangle count, a native floor (132/132); and texture painting where a stroke lands in the hand-computed texel, the seam bleed is bitten by a control that FIRES with the bleed removed, and a session replays byte-identically (109/109). The general UV unwrap landed too, at 78 of 78, so M1 AND M2 ARE BOTH COMPLETE and the chain sculpt-to-chart-to-solve-to-pack-to-paint exists end to end for an arbitrary mesh for the first time. STILL OPEN, and this is now the whole story: there is no retopology, no multiresolution, no cloth, no node graph and no collaborative editing; and nothing has been carried end to end into all four consumers with one ruler, which is the rung that would make the storefront-game-VR-clinic claim checkable rather than asserted. The distance to a studio was the editor, and the editor now exists.

Where we need to go. ONE sovereign content-creation studio on nishifamily.com -- model, sculpt, unwrap, texture, rig and assemble -- first byte up in NishiLang with no third-party code in the loop, running in the visitor's browser with no installation, where the asset is a RULE the studio edits rather than a file it mutates, and where one authored source feeds the storefront, the game, the VR frame and the clinical twin without a re-export. Blender, ZBrush, Maya and Substance appear ONLY as oracles and benchmarks, never as dependencies.

The unit. 1 u = one measured session-leg, the calibration graphics.plan and procgen.plan already use. Fidelity claims are in millimetres of surface deviation (md_measure is the estate ruler and it already exists); interaction claims are in milliseconds per operation at a stated triangle count. An authoring claim with no artifact a person can open, and a latency claim with no triangle count, are both opinion wearing a number.

Where we already exceed: 0 u. Determinism and metrology, not features. No rival ships a bit-exact authoring pipeline, a millimetre ruler as a gate tooth, or an asset that regenerates its own PBR set per genome instead of shipping baked files. Those are true today and they are the only exceeds this board claims.

Cost to an editor that exists at all: 0 u -- M1 IS COMPLETE. DC1 the operation stack, DC2 parametric re-evaluation and DC3 the manifold oracle ALL LANDED 2026-08-25, gate-proven on the promoted binaries at 44/44 and 32/32 and mutation-proven. Nothing here was new geometry: it was the record-undo-replay spine every DCC is built on and the one thing this estate had never had. The estimate was 5.5 u and the milestone closed inside one session, which is worth saying plainly because it recalibrates every estimate below it downward.

Cost to authoring a real asset in the browser: 0 u -- M2 IS COMPLETE. DC4 the viewport (113/113), DC5 the general UV unwrap (78/78) and DC7 texture painting (109/109) ALL LANDED 2026-08-25, every one mutation-proven. The estimate was 13 u and both milestones closed inside one session, which recalibrates the rungs below downward and should be read as evidence that the remaining estimates are pessimistic rather than as a reason to trust them. AND THE ONE THING THAT WAS STILL OWED HERE IS NOW PAID: the viewport's per-operation latency is published at four stated triangle counts, so DC4's done-rule is met in both halves. A pick over 2048 triangles costs 455054 ns on the native floor, 4.1 percent of a 90 Hz frame, and the cost is measured SUB-linear in triangle count so the cited 2 ms app-logic ceiling buys 9001 triangles per pick. A latency claim may now be made from this board, PROVIDED it carries its triangle count and the word FLOOR -- the number is native x86 and the shipped viewport is wasm in a browser.

Cost to the asset leaving the studio: 11 u. DC11 ingest under a provenance and licence gate, DC12 one asset proven into four consumers by one ruler, DC6 retopology so a sculpt can become a game-budget mesh at all.

Cost to exceeding on the axes a file-based float DCC cannot enter: 26 u. DC9 the layer-aware brush, DC8 multiresolution, DC10 garment drape, DC13 the node graph over the op stack, DC14 collaborative multi-seat editing on one log. Every one of these is reachable only BECAUSE the asset is generated and the ops are deterministic.

Market-entry ladder

Level	Best in class (Aug 2026)	What we have	What must grow	Verdict
adult character VR indie (operator-named 2026-09-06: VR Hot, Breeders of the Nephelym, Carnal Instinct, Wicked Island)	a first-person VR scene with one or a few high-fidelity physics-bodied characters, skin and hair and cloth simulation, direct interaction and customisation; the titles are named by the operator and NOT yet mirrored, the Anime Nyan archive.org digest on the modding board is the learn-mining source	a voxel beach with a genome cast drawn through WebGPU; garments-as-mesh, hair colour and planted feet REJECTED by the operator 2026-08-30; auto-rig DC16 and motion retarget shipped; no VR door yet (DC15 open)	the synth human itself: true anatomy strata (DC9), draped garments (DC10), the VR door (DC15), the physics body and hair under the softbody and motion boards, skin under skintwin, all judged by the referee panel at full resolution against the operator's reference frames	GROW-FIRST

Research bar. DanceXR character and motion ingestion is measured on Documented PMX/XPS models, VMD/BVH motion, content library and interactive character controls; direct web player https://vrstormlab.com/dancexr/play supplied by operator. Theirs: Reproducible import-to-retarget-to-playback-to-inspect tasks with permitted test assets and preserved authoring provenance. Ours: Candidate benchmark, documentation reviewed 2026-09-07; player and Nishi parity not exercised. DCC owns ingestion/editing workflow, graphics owns output, softbody owns dynamics and gameengine owns runtime integration..

Research bar. Womp browser authoring and materials is measured on Browser UI, live scene materials and lights, image/video/mesh export; official FAQ says rendering runs on cloud servers. Theirs: Compare task completion, editing/undo, geometry export fidelity, response latency and total client/server cost. Ours: Candidate benchmark, documentation reviewed 2026-09-07; browser delivery is not proof of client-side rendering. No Womp or Nishi side-by-side workload executed..

Native modding and private user content: Operator priority 2026-09-07. Compose import, validation, reversible editing, rig and motion binding, materials, wardrobe and game behavior through stable estate interfaces. Local-first ingestion must show what leaves the device and require explicit sharing permissions; imported content retains its source and rights metadata. Prove each chain with reproducible fixtures and served UI interaction before parity or privacy claims. This is a target, not a claim that all ingestion and modding paths are wired.

Latest recorded release

2026-10-01 · DC11

ain_admit is real: runtime/nx_asset_admit_lib.nx (18106 B) holds the ONE intake decision and composes the incumbents instead of twinning them -- the sha256 pin and the journal row are nx_asset_prov_lib (one journal, last row wins), the licence lookup is nx_licgate_lib (the one rights table knowledge/model_license.conf), the consent predicate is nx_consent. An asset is admitted only when its admission record (one key=value per line) names origin, a licence id the rights table carries, a sha256 equal to the sha256 of the bytes presented and a likeness declaration of 0 or 1, and a likeness also carries a consent record whose release is signed and whose term has not lapsed; every refusal returns a reason and the conjunct it belongs to (record, asset, origin, licence, pin, likeness, rights-table, consent, journal), a duplicated key is refused as ambiguous, a refusal writes nothing, and the same admission twice appends one row. PROMOTED: nx_asset_admit (the command, 61307 B, sha256 e0d9b3771e8f2210f2b28d2fdb06b815997fe2bf24e32cd0d1eb8ed39efa68b3) and nx_asset_admit_gate (134614 B, sha256 921ce5dee6106396522c6e8a40fcc86905ba275341d8c36a98b7f33e6daa44ef), both registered, both _offc twins installed, the gate rostered and added to dcc.gates; the dcc.matrix watch row is repointed from the glTF importer to runtime/nx_asset_admit_lib.nx so the contract symbol is measured where it lives. Shipped, referee grade 60 of 60 with the mutant killed, awaiting acceptance.

Release history and work log

Release entries describe recorded changes; they do not establish that every capability passed evaluation.

52 of 74 capability source symbols found|7 of them authored stronger-claims|22 open|coverage 702/1000|adoption 39 full / 13 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.

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 = 48 vs present 48 · named + unnamed + withheld = 17 vs open 17 · both reconcile

liar-kill conj=GPQN · all four conjuncts held

graded document: BUILDROOT tree, 45033 bytes · gates map: PRIMARY · stamped 37d 6h ago · STALE against the shared TTL of 604800s — this is evidence about a build nobody can vouch for now · source ../knowledge/status/evstamp_dcc.verdict

Gap class	What it is, and the work it names
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.

This band reports the referee counts. The per-axis worklist rows — which axis is unsupported, which watch is ready to flip, which gap is unnamed — are printed by nx_swcompare_evidence dcc 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: STALE_OR_INVALID_BINDING. Recorded priorities require current acceptance evidence and resource checks before execution.

Recorded ranking withheld

Retained ranking is not bound to the current plan and matrix: STALE_OR_INVALID_BINDING. Historical data remains retained. Refresh the ranking after resolving the input differences. Latest execution outcome is unverified.

Declared roadmap — contract, acceptance, executor, effort

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Non-destructive operation stack with bit-exact replay (DC1)	es_replay	An append-only operation log records every authoring action; undo and redo restore the exact prior state digest; replaying the whole log from the base reproduces the live state BYTE-IDENTICALLY, which is the claim no float DCC can make. Done-rule: a gate proves push, undo, redo and full replay all agree on the digest, that a MUTATED op changes the digest (the anti-vacuity tooth -- a stack that silently ignored its ops would pass replay-equals-live trivially), and that undoing past the base and redoing past the head are both REFUSED by name rather than clamped.	Organ	2 u
Parametric re-evaluation at depth without losing downstream work (DC2) after DC1	es_reeval	Changing a parameter of operation N re-evaluates the stack and PRESERVES operations N+1..M -- the property that makes an editor non-destructive and the reason Blender has a modifier stack. Done-rule: a gate edits a parameter at depth in a stack whose later ops depend on it, proves the later ops still apply, and proves the result equals a fresh log built with the new parameter from scratch.	Organ	2 u
Mesh validity and manifold oracle (DC3)	mv_manifold	A referee that REFUSES an asset rather than shipping a broken one: non-manifold edges, boundary holes, degenerate and zero-area triangles, duplicate vertices, inconsistent winding, and self-intersection, each reported as a NAMED count with the offending element index. Done-rule: bite-proven on a planted defect of each class, with a positive control that a known-good generated mesh passes clean. This is the tooth every other rung on this board depends on and no rival ships it as a refusing gate.	Organ	1.5 u
Interactive viewport in the browser: pick, gizmo, camera (DC4) after DC1	vp_pick_screen	DONE 2026-08-25, nx_vpick_gate GREEN 132/132 on the promoted binary and mutation-proven (valid_mutants=1 killed=1 not_reached=0 mutant_live=0, restore byte-identical at 45716 B). A screen point becomes a ray in the SAME camera convention nx_meshview_wasm already orbits in; Moller-Trumbore picks the nearest triangle in exact integer arithmetic; a drag resolves to ONE operation on the DC1 stack, so it is undoable, redoable and replays bit-identically. The teeth that make it real rather than plausible: the nearer triangle wins AND STILL WINS WHEN THE BUFFER ORDER IS REVERSED, so the answer is depth and not array order; a degenerate triangle is never returned as a hit and is a DISTINCT answer from parallel and from behind; a miss returns a named negative code no valid index can collide with; the gizmo magnitude is checked against an INDEPENDENTLY DERIVED closed form rather than against itself; and three separate teeth assert the run never wrote a word of the mesh buffer. THE JOIN SYMBOL WAS CORRECTED FROM vp_pick TO vp_pick_screen: the rung named a symbol the shipped organ does not declare, and a contract pointing at a symbol that cannot exist NEVER FLIPS -- the ranker would have read this rung open forever with nothing wrong in the work. THE INTERACTION-LATENCY HALF IS NOW MET, 2026-08-25, and it was the last thing this rung owed. The gate went 113 of 113 to 132 of 132: it times vp_pick_screen over procedurally generated grids at FOUR stated triangle counts and publishes nanoseconds per pick with its repetition count and clock source printed beside it -- 32 tris 9659 ns, 128 tris 30332 ns, 512 tris 116495 ns, 2048 tris 455054 ns, each the MINIMUM of two samples of 1024 repetitions timed on CLOCK_MONOTONIC at 1 us resolution and divided once at the end, with the min-to-max spread printed so a reader sees how noisy the box was rather than having to trust it was quiet. THE BAR IS CITED, NOT INVENTED: knowledge/gamefeel_oracle.conf already banks frame_budget_ms_90hz at 11..12 ms and js_block_ms at 2 ms, the derived 1e6/90 = 11111 us agrees with the frame row, and these are the same two rows nx_tissuebudget measures the tissue tick against -- reusing that source rather than opening a second one is why the two organs can be compared at all. A pick over 2048 triangles is therefore 40955 ppm, about 4.1 percent, of one 90 Hz frame, and the cited 2 ms app-logic ceiling buys 9001 triangles per pick. THE GROWTH QUESTION IS ANSWERED TOO, because one triangle count structurally cannot answer it -- at a single N a quadratic curve and a linear one are the same number: per-triangle cost measures 736 permil across a 64x range where 1000 would be exactly linear, so the pick is SUB-linear as the fixed ray construction amortises, and the bar separating linear from quadratic is DERIVED as the geometric mean of the two hypotheses rather than tuned until the result passed. THE FIGURE IS A NATIVE FLOOR AND IS LABELLED ONE: the shipped viewport is wasm in a visitor browser, which is slower, so this measurement can refute a budget but can never prove the browser meets it. ONE TRADE WAS MADE AND IS NAMED RATHER THAN LEFT TO BE FOUND: an 8192-triangle rung was measured (1967 us, 98 percent of the cited 2 ms ceiling) and then REMOVED from the shipped ladder because it pushed the sweep to about 8 s against a 12 s runner deadline, and a gate that intermittently outlives its caller is indistinguishable from one that emits nothing. Nothing was lost by it: the linear model predicted 1.87 ms at that size against 1.97 ms directly measured, which is the extrapolation checked against the rung it extrapolates past.	Organ	4 u
General conformal UV unwrap with a distortion metric (DC5) after DC3	uv_conformal	DONE 2026-08-25: nx_uvunwrap_gate GREEN 78 of 78 on the promoted binary, mutation-proven (killed=1 not_reached=0, restore byte-identical at 71921 B). Unwrap an ARBITRARY mesh, not just the human anatomy layout the current unwrap is bound to: seam input, a conformal solve, and an angle-and-area distortion metric per chart. Done-rule: distortion on a developable test surface reads at or near zero, a known-difficult surface reports its distortion honestly rather than silently folding, and overlapping charts are REFUSED. MEASURED, and every fixture moved: the flat quad went ANGLE 987 permil to 0, the cube's twelve seams give exactly six charts with zero overlaps, the cut cylinder wall unwraps clean, and the CONE POINT reports 174 permil against a plane at 0 -- honest, non-zero and not a refusal, which is the only correct answer for a surface that cannot be flattened. THIS RUNG PREVIOUSLY PUBLISHED A WRONG ROOT CAUSE AND THE CORRECTION IS RECORDED HERE RATHER THAN QUIETLY REPLACED. The published cause was a discarded return code in uvi_unfold's walk; that is a REAL latent defect and was fixed too, but it caused NONE of the fifteen failures. The true cause was found by refuting the briefed one with arithmetic from the gate's own output: an unseamed 4-vertex 2-triangle mesh reported texc corners lost = 1 where a bijection makes 0 the only possible answer, so the corner-to-class map was corrupt BEFORE any unfold ran. In uv_chart_build's relabel block, uvf is BOTH the union-find parent forest AND the corner-to-class map, and the loop converts one into the other IN PLACE: from the second iteration uvi_find chases a class id as a parent index, and because that encoding is NEGATIVE it indexes before the array and path-compression-WRITES into the CSTART and CORDER slots beneath. Roots are now resolved into a dedicated array before any parent is rewritten. ALSO SHIPPED: UV_E_UNPLACED as its own refusal rather than overloading UV_E_NOPIN, because two causes sharing one code is exactly what made this hard to find -- and it is declared NOT bite-proven, since a mesh that passes validation cannot reach it from the public pipeline.	Organ	5 u
Quad retopology from a sculpt to a game budget (DC6) after DC3	rt_quadremesh	Convert a dense sculpt into a quad-dominant mesh at a requested face budget while holding surface deviation under a stated millimetre bound measured by the EXISTING md_measure ruler, never a new one. Done-rule: the budget is met, the deviation bound is met and reported, and a mesh that cannot meet the bound is refused rather than silently exceeded.	Organ	6 u
Texture painting onto the regenerated PBR set (DC7) after DC5	tp_stroke	DONE 2026-08-25, nx_texpaint_gate GREEN 109/109 on the promoted binary and mutation-proven (killed=1 not_reached=0, restore byte-identical). All three clauses of the done-rule are met and each is bitten rather than asserted: the known surface point writes the known texel, and the expected address is computed BY ARITHMETIC IN THE FIXTURE rather than by the code under test; the seam does not tear, proven by neg-control-seam-TEARS-without-the-bleed which FIRES on a build with the bleed removed and is silent with it; and replaying the log reproduces the map BYTE-IDENTICALLY, with undo restoring the base byte for byte and redo returning to the exact same bytes. The anti-vacuity tooth that matters: a painter which wrote the whole atlas would trivially pass any known-texel-changed check, so the gate counts the disc area by hand (25 texels) and asserts ZERO texels changed outside the radius. The bleed margin is DERIVED from the brush radius and the chart width rather than picked. THE DEPENDENCY IS NOW MET: DC5 landed the same day at 78 of 78, so painting is no longer bounded to the chart layout the anatomy path supplies -- an arbitrary sculpt or boolean result can be charted, solved, packed and then painted, which is the first time this estate has had that path end to end.	Organ	4 u
Multiresolution sculpt levels (DC8) after DC6	ms_level	Subdivide to a higher level, sculpt detail, and step DOWN without losing the detail -- the property that makes high-frequency sculpting tractable. Done-rule: detail added at level N survives a round trip to level N-1 and back within a stated deviation, and the level stack is an op on the DC1 log.	Organ	4 u
Sculpt brush constrained to an anatomical stratum (DC9) after DC1	sc_layer_brush	THE EXCEED, and it is only reachable because our bodies are generated inside-out: a brush that edits the MUSCLE layer and lets the skin re-emerge over it, or that adds fat without touching the muscle beneath. Done-rule: a stroke on the muscle layer provably changes the skin surface, a stroke on the fat layer provably does not change the muscle layer, and the resulting surface change is reported in millimetres by md_measure. ZBrush layers are shape blends over one surface and structurally cannot express this.	Organ	3 u
Garment pattern, sewing and drape (DC10) after DC3	gd_drape	Flat pattern pieces, sewing seams, and a drape onto the generated body -- the Marvelous Designer capability the storefront lane needs, on the existing soft-body solver rather than a second one. Done-rule: a sewn garment drapes without self-interpenetration at a stated contact distance and its fit is reported against the body measurements the estate already extracts.	Organ	8 u
Asset ingest under a provenance and licence gate (DC11)	ain_admit	Every ingested asset carries where it came from, what licence admits it, and a content pin; an asset without them FAILS CLOSED at intake naming the missing conjunct, and a likeness requires a consent record. Done-rule: a planted asset missing each conjunct in turn is refused by name, and a complete one is admitted with its provenance readable afterwards.	Organ	2 u
One asset, four consumers, one ruler (DC12) after DC3,DC11	xp_consumers	DONE 2026-08-25 FOR THREE CONSUMERS WITH THE FOURTH ABSTAINING BY NAME, and the done-rule is AMENDED HERE rather than quietly satisfied. nx_xport_gate GREEN 38/38, mutation-proven, deviations recovered OUT OF THE ARTIFACT BYTES and reported per consumer in tenths of a millimetre: game NXA reads 0/0/0 and clinical NXMSH2 reads 0/0/0, both EXACTLY lossless against a derived bound of 0 -- by ARITHMETIC and not by tolerance, because their unit is an integer multiple of the ruler's, so the forward map is a multiplication and the inverse an exact division. Storefront glb and obj read 0/4/4 against a derived bound of 11. THE ORIGINAL DONE-RULE SAID ALL FOUR AND THAT WAS THE WRONG BAR: vf_build takes NO MESH ARGUMENT AT ALL -- its entire scene is a six-row constant table and it emits a PNG -- so VR geometry could not be recovered even if a mesh could be fed in. That is a missing capability in the compositor, not a shortfall in this proof, and it is now DC15 with a concrete contract. The amended rule: every consumer that CAN receive the asset is measured, and every one that cannot ABSTAINS BY NAME with its reason re-checked against live source on every run, so an abstention can never quietly decay into a pass. THE FINDING WORTH MORE THAN THE RUNG: the storefront is our ONLY lossy consumer and the loss is STRUCTURAL -- write_glb and write_obj hold coordinates at fx1024, a quantum of 0.977 mm, which is COARSER than the 1 to 2 mm bar nx_mmdev_lib's own header says the clinical unit exists to serve. Measured 0.4 mm on this asset, 1.1 mm derived worst case. An asset that is clinically exact going in is not clinically exact by the time it reaches a storefront viewer, and nothing before this rung could see that. A second tooth earned its place the same way: orientation was first checked by per-triangle normal dot product, which CANNOT catch a reflection, and the replacement signed-volume test carries the case that proves it -- a mirrored box reads max deviation 0 while vol6 flips sign.	Organ	3 u
A mesh can reach the VR compositor at all (DC15) after DC12	vf_build_mesh	SPLIT OUT OF DC12 2026-08-25 BECAUSE THE MEASUREMENT NAMED A MISSING CAPABILITY RATHER THAN A GAP IN THE PROOF. runtime/_hdl_build/nx_vrframe.nx builds a submit-ready stereo frame with an off-axis pinhole per eye and a barrel-distortion pass, and it is real work -- but vf_build takes only a distortion flag and an output path, its scene is the six-row vf_pill constant table, and it rasterises nothing triangular. Contract: vf_build accepts a mesh and a pose, a triangle rasteriser draws it into both eye views, and the geometry the compositor received is recoverable so DC12's existing ruler can measure it exactly like the other three. Done-rule: nx_xport_gate's VR row moves from UNREACHED to a measured deviation against the same derived bound, and the two source-signature teeth that currently PIN the abstention flip from asserting absence to asserting presence. Until then the VR half of the storefront-game-VR-clinic claim is a plan and this board says so.	Organ	6 u
Node-graph authoring over the operation stack (DC13) after DC2	ng_eval	A graph whose evaluation IS the DC1 op stack rather than a second engine -- one evaluator, so a graph edit and a manual edit cannot diverge. Done-rule: a graph and the equivalent hand-built op log produce byte-identical state, which is the tooth that proves there is only one evaluator.	Organ	5 u
Collaborative multi-seat editing on one operation log (DC14) after DC1	co_merge	Two seats editing one asset converge to the same state regardless of the order their operations arrive, on the append-only log DC1 already provides. Done-rule: a gate replays two interleaved op streams in both orders and proves the final digests are identical, with a neg-control proving a genuinely conflicting pair is REFUSED rather than silently resolved.	Organ	6 u

Milestones

Milestone	Rungs	Cumulative
M1 · An editor exists: record, undo, replay, and refuse a broken mesh	DC1,DC2,DC3	5.5 u
M2 · Author a real asset end to end in the browser	DC4,DC5,DC7	13 u
M3 · The asset leaves the studio and lands in four products	DC6,DC11,DC12,DC15	17 u
M4 · Exceed on the axes a file-based float DCC cannot enter	DC8,DC9,DC10,DC13,DC14	26 u

Ladder to SOTA

Two dated targets, and every rung's role toward them. Best in class is the proven, deployed leader today; the frontier is the research edge as of its month. Each target is dated by a sotabar row and judged against the current month by the bar-age ruler (FRESH, ATTESTED with the fallback month named, or refused). Each rung is substrate (what the target's own methods consume), an arm (a sovereign arm that triangulates without competing for the number), a contender (its done-rule is a number against the target) or superseded (kept as capability, retired as a claim), and carries the rungs it stands on.

best-in-class (T-BIC-CHAR), dated by DCB6-modchar = 2026-09 UNATTESTED. A playable girl at the modded-Skyrim standard: physics body, SMP-class cloth and hair, an outfit of garment meshes that drape, auto-rigged and photoreal under the estate referee, in the browser and in VR ref modbooru_daemon_girl

frontier (T-FR-DCC), dated by DCB5-womp = 2026-09 UNATTESTED. Browser-native parametric authoring where the asset IS its generating rules: SDF and CSG scene authoring with no installation (Womp) plus node-based physics (Blender 5.2 LTS) is the September 2026 edge, and ours is the deterministic bit-exact operation log underneath it [womp_2026_09] [blender_2026_09] ref womp_2026_09

Rung	Role	Toward	Stands on	Why
DC1	substrate	T-FR-DCC	-	the append-only bit-exact operation log every other rung replays through
DC2	contender	T-FR-DCC	DC1	parametric re-evaluation at depth is the rules-not-meshes claim itself
DC3	substrate	T-BIC-CHAR	-	the manifold oracle every garment and body mesh must pass before rigging
DC4	substrate	T-FR-DCC	DC1	the browser viewport the authoring surface lives in
DC5	substrate	T-BIC-CHAR	DC3	conformal UV unwrap feeds the skin and garment texture sets
DC6	substrate	T-BIC-CHAR	DC3	quad retopology takes a sculpted body to the game budget the benchmark bodies ship at
DC7	substrate	T-BIC-CHAR	DC5	texture painting onto the regenerated PBR set, the skin and outfit maps
DC8	arm	T-FR-DCC	DC6	multiresolution levels as ops on the log, an arm of the rules-not-meshes claim
DC9	contender	T-BIC-CHAR	DC1	anatomical-stratum sculpting is what makes the body true anatomy rather than a surface
DC10	contender	T-BIC-CHAR	DC3	garment pattern, sewing and drape is the true-clothes half of the benchmark
DC11	substrate	T-BIC-CHAR	-	ingest under provenance and licence is how a benchmark asset enters as a referee frame and never as a dependency
DC12	arm	T-BIC-CHAR	DC3,DC11	one asset four consumers proves the character reaches game, storefront, VR and clinic from one file
DC13	arm	T-FR-DCC	DC2	node-graph authoring over the log is the authoring surface the frontier bar sells
DC14	arm	T-FR-DCC	DC1	multi-seat editing on one log answers the browser bar's real-time collaboration
DC15	substrate	T-BIC-CHAR	DC12	the VR compositor door the benchmark girl is met through
DC16	contender	T-BIC-CHAR	DC2	auto-rig without a human rigger is the physics-body half: skeleton fit and skin weights on any topology

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Auto-rig arbitrary topology: skeleton fit and skin weights without a human rigger (DC16) after DC2	amesh_run	THE LOAD-BEARING GAP FOR PROCGEN AND NISHIGEN, and it is not an authoring convenience. A parametric head (GNM, Apache-2.0) ships with its rig baked in because its topology is fixed, but procedural bodies and creatures have ARBITRARY topology -- so a generator that makes 10,000 creatures makes 10,000 UNUSABLE meshes until this rung lands. Substrate already exists and is dark: nx_skelgen, nx_nxa_skin, nx_skinwrap, nx_rigfloor, nx_dyna_bind_skin. Done-rule: given an unrigged mesh, emit a skeleton plus per-vertex weights; the bound mesh deforms at an extreme pose without candy-wrapper collapse, measured through md_measure (NEVER a second mesh ruler -- see this board's own risk row); a neg-control mesh that is not the requested skeleton archetype is REFUSED BY NAME rather than mis-rigged in silence, because a confidently wrong rig is worse than a refusal the caller can see. Reference implementation to study, never to vendor: Mesh2Motion (mesh2motion-app, 3,212 stars, GLB in and GLB out, fit-skeleton-to-mesh then retarget) -- the ALGORITHM is the oracle, the TypeScript never enters the build path	Organ	3 u

Milestones

Milestone	Rungs	Cumulative
M5 · Rig what the generator makes	DC16	29 u

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Technical-illustration line render of an exploded assembly, print-ready vector (DC17) after stem:R7	ti_explode_render	INTAKE 1790106111 (operator 2026-09-22): from an exploded layout (stem R7 over procgen ex3_pose) render a hidden-line axonometric technical illustration as vector -- silhouette and crease lines with stroke weight by depth, numbered callouts placed by the ONE chart-layout ruler (datavis nx_chartlay_lib leaders, critic 0 defects), emitted as SVG through nx_render_svg and as spot-colour separations for screen print; done-rule: the banked car-explode fixture renders byte-identical twice, the callout critic grades 0 defects, and the separation count equals the palette rows	Organ	3 u

Rung	Role	Toward	Stands on	Why
DC17	arm	T-FR-DCC	stem:R7 UNRESOLVED(organ-unreadable)	an illustration export arm of the studio: it consumes the authored assembly and renders a print-ready view; it triangulates the studio's usefulness without competing for the parametric-authoring number

Milestones

Milestone	Rungs	Cumulative
M6 · Illustration export	DC17	32 u

Ladder verdict. LADDERED. Targets, rung roles and bars are counted in the tables below, each partition re-derived beside its total. Bars read STALE against the current month 2026-10.

Ladder verdict, counted (17 rows, more than the 12-row disclosure bar)

Ladder verdict, counted
Class	Count	Note
targets	2	sotatarget rows on the plan
best in class	1	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	17	rung rows on the plan
placed	17	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	17	RECONCILED: the parts sum to the total 17
substrate	8	what the target's own methods consume
arm	5	a sovereign arm that triangulates without competing for the number
contender	4	its done-rule is a number against the target
superseded	0	kept as capability, retired as a claim
placed, summed	17	RECONCILED: the parts sum to the total 17

Bar freshness, counted
Class	Count	Note
bars	6	sotabar rows on the plan
fresh	0	the bar month is the current month
attested	0	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	6	nobody looked this month
malformed	0	a sotabar row missing a field
duplicate	0	bar ids an earlier row already carries
bars, summed	6	RECONCILED: the parts sum to the total 6
current month	2026-10	the month every bar is judged against

Treatment verdict. NOT DECLARED. No rung on this board declares its three-step treatment (a treat row naming the MCP row, the API route, the primitive it composes, the workflow that composes it and the agent beat that runs it); the ranker labels every rung TREATMENT-UNDECLARED until it does.

Inspect a rung and its prerequisites

Declared nodes 17. Rank input binding: STALE_OR_INVALID_BINDING. Dependency order is authored. Implementation, acceptance evidence, authority and resource readiness are unverified. No action is recommended or dispatched here.

Plan SHA-256 cc5db0c881e1a1ba473c531a5c0327de8ecd057aa0b70963094f8e9aedcf9151. Target rows 2; role rows 17. Existing risks and release worklog retain their own scope; no node completion is inferred.

Runtime observation coverage: 0 of 17 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.

DC1 — Non-destructive operation stack with bit-exact replay
Prerequisites: None declared; this does not establish execution eligibility.
Contract: es_replay
Acceptance: An append-only operation log records every authoring action; undo and redo restore the exact prior state digest; replaying the whole log from the base reproduces the live state BYTE-IDENTICALLY, which is the claim no float DCC can make. Done-rule: a gate proves push, undo, redo and full replay all agree on the digest, that a MUTATED op changes the digest (the anti-vacuity tooth -- a stack that silently ignored its ops would pass replay-equals-live trivially), and that undoing past the base and redoing past the head are both REFUSED by name rather than clamped.
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.
DC2 — Parametric re-evaluation at depth without losing downstream work
Prerequisites: DC1 (acceptance unverified)
Contract: es_reeval
Acceptance: Changing a parameter of operation N re-evaluates the stack and PRESERVES operations N+1..M -- the property that makes an editor non-destructive and the reason Blender has a modifier stack. Done-rule: a gate edits a parameter at depth in a stack whose later ops depend on it, proves the later ops still apply, and proves the result equals a fresh log built with the new parameter from scratch.
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.
DC3 — Mesh validity and manifold oracle
Prerequisites: None declared; this does not establish execution eligibility.
Contract: mv_manifold
Acceptance: A referee that REFUSES an asset rather than shipping a broken one: non-manifold edges, boundary holes, degenerate and zero-area triangles, duplicate vertices, inconsistent winding, and self-intersection, each reported as a NAMED count with the offending element index. Done-rule: bite-proven on a planted defect of each class, with a positive control that a known-good generated mesh passes clean. This is the tooth every other rung on this board depends on and no rival ships it as a refusing gate.
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.
DC4 — Interactive viewport in the browser: pick, gizmo, camera
Prerequisites: DC1 (acceptance unverified)
Contract: vp_pick_screen
Acceptance: DONE 2026-08-25, nx_vpick_gate GREEN 132/132 on the promoted binary and mutation-proven (valid_mutants=1 killed=1 not_reached=0 mutant_live=0, restore byte-identical at 45716 B). A screen point becomes a ray in the SAME camera convention nx_meshview_wasm already orbits in; Moller-Trumbore picks the nearest triangle in exact integer arithmetic; a drag resolves to ONE operation on the DC1 stack, so it is undoable, redoable and replays bit-identically. The teeth that make it real rather than plausible: the nearer triangle wins AND STILL WINS WHEN THE BUFFER ORDER IS REVERSED, so the answer is depth and not array order; a degenerate triangle is never returned as a hit and is a DISTINCT answer from parallel and from behind; a miss returns a named negative code no valid index can collide with; the gizmo magnitude is checked against an INDEPENDENTLY DERIVED closed form rather than against itself; and three separate teeth assert the run never wrote a word of the mesh buffer. THE JOIN SYMBOL WAS CORRECTED FROM vp_pick TO vp_pick_screen: the rung named a symbol the shipped organ does not declare, and a contract pointing at a symbol that cannot exist NEVER FLIPS -- the ranker would have read this rung open forever with nothing wrong in the work. THE INTERACTION-LATENCY HALF IS NOW MET, 2026-08-25, and it was the last thing this rung owed. The gate went 113 of 113 to 132 of 132: it times vp_pick_screen over procedurally generated grids at FOUR stated triangle counts and publishes nanoseconds per pick with its repetition count and clock source printed beside it -- 32 tris 9659 ns, 128 tris 30332 ns, 512 tris 116495 ns, 2048 tris 455054 ns, each the MINIMUM of two samples of 1024 repetitions timed on CLOCK_MONOTONIC at 1 us resolution and divided once at the end, with the min-to-max spread printed so a reader sees how noisy the box was rather than having to trust it was quiet. THE BAR IS CITED, NOT INVENTED: knowledge/gamefeel_oracle.conf already banks frame_budget_ms_90hz at 11..12 ms and js_block_ms at 2 ms, the derived 1e6/90 = 11111 us agrees with the frame row, and these are the same two rows nx_tissuebudget measures the tissue tick against -- reusing that source rather than opening a second one is why the two organs can be compared at all. A pick over 2048 triangles is therefore 40955 ppm, about 4.1 percent, of one 90 Hz frame, and the cited 2 ms app-logic ceiling buys 9001 triangles per pick. THE GROWTH QUESTION IS ANSWERED TOO, because one triangle count structurally cannot answer it -- at a single N a quadratic curve and a linear one are the same number: per-triangle cost measures 736 permil across a 64x range where 1000 would be exactly linear, so the pick is SUB-linear as the fixed ray construction amortises, and the bar separating linear from quadratic is DERIVED as the geometric mean of the two hypotheses rather than tuned until the result passed. THE FIGURE IS A NATIVE FLOOR AND IS LABELLED ONE: the shipped viewport is wasm in a visitor browser, which is slower, so this measurement can refute a budget but can never prove the browser meets it. ONE TRADE WAS MADE AND IS NAMED RATHER THAN LEFT TO BE FOUND: an 8192-triangle rung was measured (1967 us, 98 percent of the cited 2 ms ceiling) and then REMOVED from the shipped ladder because it pushed the sweep to about 8 s against a 12 s runner deadline, and a gate that intermittently outlives its caller is indistinguishable from one that emits nothing. Nothing was lost by it: the linear model predicted 1.87 ms at that size against 1.97 ms directly measured, which is the extrapolation checked against the rung it extrapolates past.
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.
DC5 — General conformal UV unwrap with a distortion metric
Prerequisites: DC3 (acceptance unverified)
Contract: uv_conformal
Acceptance: DONE 2026-08-25: nx_uvunwrap_gate GREEN 78 of 78 on the promoted binary, mutation-proven (killed=1 not_reached=0, restore byte-identical at 71921 B). Unwrap an ARBITRARY mesh, not just the human anatomy layout the current unwrap is bound to: seam input, a conformal solve, and an angle-and-area distortion metric per chart. Done-rule: distortion on a developable test surface reads at or near zero, a known-difficult surface reports its distortion honestly rather than silently folding, and overlapping charts are REFUSED. MEASURED, and every fixture moved: the flat quad went ANGLE 987 permil to 0, the cube's twelve seams give exactly six charts with zero overlaps, the cut cylinder wall unwraps clean, and the CONE POINT reports 174 permil against a plane at 0 -- honest, non-zero and not a refusal, which is the only correct answer for a surface that cannot be flattened. THIS RUNG PREVIOUSLY PUBLISHED A WRONG ROOT CAUSE AND THE CORRECTION IS RECORDED HERE RATHER THAN QUIETLY REPLACED. The published cause was a discarded return code in uvi_unfold's walk; that is a REAL latent defect and was fixed too, but it caused NONE of the fifteen failures. The true cause was found by refuting the briefed one with arithmetic from the gate's own output: an unseamed 4-vertex 2-triangle mesh reported texc corners lost = 1 where a bijection makes 0 the only possible answer, so the corner-to-class map was corrupt BEFORE any unfold ran. In uv_chart_build's relabel block, uvf is BOTH the union-find parent forest AND the corner-to-class map, and the loop converts one into the other IN PLACE: from the second iteration uvi_find chases a class id as a parent index, and because that encoding is NEGATIVE it indexes before the array and path-compression-WRITES into the CSTART and CORDER slots beneath. Roots are now resolved into a dedicated array before any parent is rewritten. ALSO SHIPPED: UV_E_UNPLACED as its own refusal rather than overloading UV_E_NOPIN, because two causes sharing one code is exactly what made this hard to find -- and it is declared NOT bite-proven, since a mesh that passes validation cannot reach it from the public pipeline.
Authored effort: 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.
DC6 — Quad retopology from a sculpt to a game budget
Prerequisites: DC3 (acceptance unverified)
Contract: rt_quadremesh
Acceptance: Convert a dense sculpt into a quad-dominant mesh at a requested face budget while holding surface deviation under a stated millimetre bound measured by the EXISTING md_measure ruler, never a new one. Done-rule: the budget is met, the deviation bound is met and reported, and a mesh that cannot meet the bound is refused rather than silently exceeded.
Authored effort: 6. 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.
DC7 — Texture painting onto the regenerated PBR set
Prerequisites: DC5 (acceptance unverified)
Contract: tp_stroke
Acceptance: DONE 2026-08-25, nx_texpaint_gate GREEN 109/109 on the promoted binary and mutation-proven (killed=1 not_reached=0, restore byte-identical). All three clauses of the done-rule are met and each is bitten rather than asserted: the known surface point writes the known texel, and the expected address is computed BY ARITHMETIC IN THE FIXTURE rather than by the code under test; the seam does not tear, proven by neg-control-seam-TEARS-without-the-bleed which FIRES on a build with the bleed removed and is silent with it; and replaying the log reproduces the map BYTE-IDENTICALLY, with undo restoring the base byte for byte and redo returning to the exact same bytes. The anti-vacuity tooth that matters: a painter which wrote the whole atlas would trivially pass any known-texel-changed check, so the gate counts the disc area by hand (25 texels) and asserts ZERO texels changed outside the radius. The bleed margin is DERIVED from the brush radius and the chart width rather than picked. THE DEPENDENCY IS NOW MET: DC5 landed the same day at 78 of 78, so painting is no longer bounded to the chart layout the anatomy path supplies -- an arbitrary sculpt or boolean result can be charted, solved, packed and then painted, which is the first time this estate has had that path end to end.
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.
DC8 — Multiresolution sculpt levels
Prerequisites: DC6 (acceptance unverified)
Contract: ms_level
Acceptance: Subdivide to a higher level, sculpt detail, and step DOWN without losing the detail -- the property that makes high-frequency sculpting tractable. Done-rule: detail added at level N survives a round trip to level N-1 and back within a stated deviation, and the level stack is an op on the DC1 log.
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.
DC9 — Sculpt brush constrained to an anatomical stratum
Prerequisites: DC1 (acceptance unverified)
Contract: sc_layer_brush
Acceptance: THE EXCEED, and it is only reachable because our bodies are generated inside-out: a brush that edits the MUSCLE layer and lets the skin re-emerge over it, or that adds fat without touching the muscle beneath. Done-rule: a stroke on the muscle layer provably changes the skin surface, a stroke on the fat layer provably does not change the muscle layer, and the resulting surface change is reported in millimetres by md_measure. ZBrush layers are shape blends over one surface and structurally cannot express this.
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.
DC10 — Garment pattern, sewing and drape
Prerequisites: DC3 (acceptance unverified)
Contract: gd_drape
Acceptance: Flat pattern pieces, sewing seams, and a drape onto the generated body -- the Marvelous Designer capability the storefront lane needs, on the existing soft-body solver rather than a second one. Done-rule: a sewn garment drapes without self-interpenetration at a stated contact distance and its fit is reported against the body measurements the estate already extracts.
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.
DC11 — Asset ingest under a provenance and licence gate
Prerequisites: None declared; this does not establish execution eligibility.
Contract: ain_admit
Acceptance: Every ingested asset carries where it came from, what licence admits it, and a content pin; an asset without them FAILS CLOSED at intake naming the missing conjunct, and a likeness requires a consent record. Done-rule: a planted asset missing each conjunct in turn is refused by name, and a complete one is admitted with its provenance readable afterwards.
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.
DC12 — One asset, four consumers, one ruler
Prerequisites: DC3 (acceptance unverified), DC11 (acceptance unverified)
Contract: xp_consumers
Acceptance: DONE 2026-08-25 FOR THREE CONSUMERS WITH THE FOURTH ABSTAINING BY NAME, and the done-rule is AMENDED HERE rather than quietly satisfied. nx_xport_gate GREEN 38/38, mutation-proven, deviations recovered OUT OF THE ARTIFACT BYTES and reported per consumer in tenths of a millimetre: game NXA reads 0/0/0 and clinical NXMSH2 reads 0/0/0, both EXACTLY lossless against a derived bound of 0 -- by ARITHMETIC and not by tolerance, because their unit is an integer multiple of the ruler's, so the forward map is a multiplication and the inverse an exact division. Storefront glb and obj read 0/4/4 against a derived bound of 11. THE ORIGINAL DONE-RULE SAID ALL FOUR AND THAT WAS THE WRONG BAR: vf_build takes NO MESH ARGUMENT AT ALL -- its entire scene is a six-row constant table and it emits a PNG -- so VR geometry could not be recovered even if a mesh could be fed in. That is a missing capability in the compositor, not a shortfall in this proof, and it is now DC15 with a concrete contract. The amended rule: every consumer that CAN receive the asset is measured, and every one that cannot ABSTAINS BY NAME with its reason re-checked against live source on every run, so an abstention can never quietly decay into a pass. THE FINDING WORTH MORE THAN THE RUNG: the storefront is our ONLY lossy consumer and the loss is STRUCTURAL -- write_glb and write_obj hold coordinates at fx1024, a quantum of 0.977 mm, which is COARSER than the 1 to 2 mm bar nx_mmdev_lib's own header says the clinical unit exists to serve. Measured 0.4 mm on this asset, 1.1 mm derived worst case. An asset that is clinically exact going in is not clinically exact by the time it reaches a storefront viewer, and nothing before this rung could see that. A second tooth earned its place the same way: orientation was first checked by per-triangle normal dot product, which CANNOT catch a reflection, and the replacement signed-volume test carries the case that proves it -- a mirrored box reads max deviation 0 while vol6 flips sign.
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.
DC15 — A mesh can reach the VR compositor at all
Prerequisites: DC12 (acceptance unverified)
Contract: vf_build_mesh
Acceptance: SPLIT OUT OF DC12 2026-08-25 BECAUSE THE MEASUREMENT NAMED A MISSING CAPABILITY RATHER THAN A GAP IN THE PROOF. runtime/_hdl_build/nx_vrframe.nx builds a submit-ready stereo frame with an off-axis pinhole per eye and a barrel-distortion pass, and it is real work -- but vf_build takes only a distortion flag and an output path, its scene is the six-row vf_pill constant table, and it rasterises nothing triangular. Contract: vf_build accepts a mesh and a pose, a triangle rasteriser draws it into both eye views, and the geometry the compositor received is recoverable so DC12's existing ruler can measure it exactly like the other three. Done-rule: nx_xport_gate's VR row moves from UNREACHED to a measured deviation against the same derived bound, and the two source-signature teeth that currently PIN the abstention flip from asserting absence to asserting presence. Until then the VR half of the storefront-game-VR-clinic claim is a plan and this board says so.
Authored effort: 6. 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.
DC13 — Node-graph authoring over the operation stack
Prerequisites: DC2 (acceptance unverified)
Contract: ng_eval
Acceptance: A graph whose evaluation IS the DC1 op stack rather than a second engine -- one evaluator, so a graph edit and a manual edit cannot diverge. Done-rule: a graph and the equivalent hand-built op log produce byte-identical state, which is the tooth that proves there is only one evaluator.
Authored effort: 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.
DC14 — Collaborative multi-seat editing on one operation log
Prerequisites: DC1 (acceptance unverified)
Contract: co_merge
Acceptance: Two seats editing one asset converge to the same state regardless of the order their operations arrive, on the append-only log DC1 already provides. Done-rule: a gate replays two interleaved op streams in both orders and proves the final digests are identical, with a neg-control proving a genuinely conflicting pair is REFUSED rather than silently resolved.
Authored effort: 6. 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.
DC16 — Auto-rig arbitrary topology: skeleton fit and skin weights without a human rigger
Prerequisites: DC2 (acceptance unverified)
Contract: amesh_run
Acceptance: THE LOAD-BEARING GAP FOR PROCGEN AND NISHIGEN, and it is not an authoring convenience. A parametric head (GNM, Apache-2.0) ships with its rig baked in because its topology is fixed, but procedural bodies and creatures have ARBITRARY topology -- so a generator that makes 10,000 creatures makes 10,000 UNUSABLE meshes until this rung lands. Substrate already exists and is dark: nx_skelgen, nx_nxa_skin, nx_skinwrap, nx_rigfloor, nx_dyna_bind_skin. Done-rule: given an unrigged mesh, emit a skeleton plus per-vertex weights; the bound mesh deforms at an extreme pose without candy-wrapper collapse, measured through md_measure (NEVER a second mesh ruler -- see this board's own risk row); a neg-control mesh that is not the requested skeleton archetype is REFUSED BY NAME rather than mis-rigged in silence, because a confidently wrong rig is worse than a refusal the caller can see. Reference implementation to study, never to vendor: Mesh2Motion (mesh2motion-app, 3,212 stars, GLB in and GLB out, fit-skeleton-to-mesh then retarget) -- the ALGORITHM is the oracle, the TypeScript never enters the build path
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.
DC17 — Technical-illustration line render of an exploded assembly, print-ready vector
Prerequisites: None declared; this does not establish execution eligibility.
Contract: ti_explode_render
Acceptance: INTAKE 1790106111 (operator 2026-09-22): from an exploded layout (stem R7 over procgen ex3_pose) render a hidden-line axonometric technical illustration as vector -- silhouette and crease lines with stroke weight by depth, numbered callouts placed by the ONE chart-layout ruler (datavis nx_chartlay_lib leaders, critic 0 defects), emitted as SVG through nx_render_svg and as spot-colour separations for screen print; done-rule: the banked car-explode fixture renders byte-identical twice, the callout critic grades 0 defects, and the separation count equals the palette rows
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

Joined by symbol, never typed. Each dossier is assembled on every publish from the rows this board already carries: the plan's rung row, the matrix row that carries its contract symbol (its status re-measured by the one symbol ruler), the gate the organ declares and the verdict the evidence referee recorded on its last run, the reference marks in the rung text, the design rows that name the rung and the work log filed against it. The capability matrix links each cell to its dossier and each dossier links back to its cell and forward to the rungs it stands on; a dependency naming no rung on this plan is written as UNRESOLVED rather than linked. Absences are named: NO-MATRIX-ROW, NO-GATE-DECLARED, UNRECORDED.

Gate verdicts: UNSTAMPED — no per-gate verdict artifact is published for this domain yet (nx_swcompare_evidence dcc writes it beside its stamp); every gate below reads UNRECORDED until it runs.

Gate declarations: 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.

DC1 Non-destructive operation stack with bit-exact replay · cell PRESENT · gate NO-GATE-DECLARED

Contract	`es_replay` · executor Organ · 2 u · stands on nothing
Definition of done	An append-only operation log records every authoring action; undo and redo restore the exact prior state digest; replaying the whole log from the base reproduces the live state BYTE-IDENTICALLY, which is the claim no float DCC can make. Done-rule: a gate proves push, undo, redo and full replay all agree on the digest, that a MUTATED op changes the digest (the anti-vacuity tooth -- a stack that silently ignored its ops would pass replay-equals-live trivially), and that undoing past the base and redoing past the head are both REFUSED by name rather than clamped.
Matrix cell	non-destructive-operation-stack · `runtime/nx_editstack_lib.nx` · PRESENT decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_editstack_lib`; declare `nx_editstack_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

DC2 Parametric re-evaluation at depth without losing downstream work · cell PRESENT · gate NO-GATE-DECLARED

Contract	`es_reeval` · executor Organ · 2 u · stands on DC1
Definition of done	Changing a parameter of operation N re-evaluates the stack and PRESERVES operations N+1..M -- the property that makes an editor non-destructive and the reason Blender has a modifier stack. Done-rule: a gate edits a parameter at depth in a stack whose later ops depend on it, proves the later ops still apply, and proves the result equals a fresh log built with the new parameter from scratch.
Matrix cell	parametric-re-evaluation-at-depth · `runtime/nx_editstack_lib.nx` · PRESENT decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_editstack_lib`; declare `nx_editstack_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

DC3 Mesh validity and manifold oracle · cell PRESENT · gate NO-GATE-DECLARED

Contract	`mv_manifold` · executor Organ · 1.5 u · stands on nothing
Definition of done	A referee that REFUSES an asset rather than shipping a broken one: non-manifold edges, boundary holes, degenerate and zero-area triangles, duplicate vertices, inconsistent winding, and self-intersection, each reported as a NAMED count with the offending element index. Done-rule: bite-proven on a planted defect of each class, with a positive control that a known-good generated mesh passes clean. This is the tooth every other rung on this board depends on and no rival ships it as a refusing gate.
Matrix cell	mesh-validity-and-manifold-oracle · `runtime/nx_meshvalid_lib.nx` · PRESENT decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_meshvalid_lib`; declare `nx_meshvalid_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

DC4 Interactive viewport in the browser: pick, gizmo, camera · cell PRESENT · gate NO-GATE-DECLARED

Contract	`vp_pick_screen` · executor Organ · 4 u · stands on DC1
Definition of done	DONE 2026-08-25, nx_vpick_gate GREEN 132/132 on the promoted binary and mutation-proven (valid_mutants=1 killed=1 not_reached=0 mutant_live=0, restore byte-identical at 45716 B). A screen point becomes a ray in the SAME camera convention nx_meshview_wasm already orbits in; Moller-Trumbore picks the nearest triangle in exact integer arithmetic; a drag resolves to ONE operation on the DC1 stack, so it is undoable, redoable and replays bit-identically. The teeth that make it real rather than plausible: the nearer triangle wins AND STILL WINS WHEN THE BUFFER ORDER IS REVERSED, so the answer is depth and not array order; a degenerate triangle is never returned as a hit and is a DISTINCT answer from parallel and from behind; a miss returns a named negative code no valid index can collide with; the gizmo magnitude is checked against an INDEPENDENTLY DERIVED closed form rather than against itself; and three separate teeth assert the run never wrote a word of the mesh buffer. THE JOIN SYMBOL WAS CORRECTED FROM vp_pick TO vp_pick_screen: the rung named a symbol the shipped organ does not declare, and a contract pointing at a symbol that cannot exist NEVER FLIPS -- the ranker would have read this rung open forever with nothing wrong in the work. THE INTERACTION-LATENCY HALF IS NOW MET, 2026-08-25, and it was the last thing this rung owed. The gate went 113 of 113 to 132 of 132: it times vp_pick_screen over procedurally generated grids at FOUR stated triangle counts and publishes nanoseconds per pick with its repetition count and clock source printed beside it -- 32 tris 9659 ns, 128 tris 30332 ns, 512 tris 116495 ns, 2048 tris 455054 ns, each the MINIMUM of two samples of 1024 repetitions timed on CLOCK_MONOTONIC at 1 us resolution and divided once at the end, with the min-to-max spread printed so a reader sees how noisy the box was rather than having to trust it was quiet. THE BAR IS CITED, NOT INVENTED: knowledge/gamefeel_oracle.conf already banks frame_budget_ms_90hz at 11..12 ms and js_block_ms at 2 ms, the derived 1e6/90 = 11111 us agrees with the frame row, and these are the same two rows nx_tissuebudget measures the tissue tick against -- reusing that source rather than opening a second one is why the two organs can be compared at all. A pick over 2048 triangles is therefore 40955 ppm, about 4.1 percent, of one 90 Hz frame, and the cited 2 ms app-logic ceiling buys 9001 triangles per pick. THE GROWTH QUESTION IS ANSWERED TOO, because one triangle count structurally cannot answer it -- at a single N a quadratic curve and a linear one are the same number: per-triangle cost measures 736 permil across a 64x range where 1000 would be exactly linear, so the pick is SUB-linear as the fixed ray construction amortises, and the bar separating linear from quadratic is DERIVED as the geometric mean of the two hypotheses rather than tuned until the result passed. THE FIGURE IS A NATIVE FLOOR AND IS LABELLED ONE: the shipped viewport is wasm in a visitor browser, which is slower, so this measurement can refute a budget but can never prove the browser meets it. ONE TRADE WAS MADE AND IS NAMED RATHER THAN LEFT TO BE FOUND: an 8192-triangle rung was measured (1967 us, 98 percent of the cited 2 ms ceiling) and then REMOVED from the shipped ladder because it pushed the sweep to about 8 s against a 12 s runner deadline, and a gate that intermittently outlives its caller is indistinguishable from one that emits nothing. Nothing was lost by it: the linear model predicted 1.87 ms at that size against 1.97 ms directly measured, which is the extrapolation checked against the rung it extrapolates past.
Matrix cell	viewport-selection-and-gizmo · `runtime/nx_vpick_lib.nx` · PRESENT decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_vpick_lib`; declare `nx_vpick_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

DC5 General conformal UV unwrap with a distortion metric · cell PRESENT · gate NO-GATE-DECLARED

Contract	`uv_conformal` · executor Organ · 5 u · stands on DC3
Definition of done	DONE 2026-08-25: nx_uvunwrap_gate GREEN 78 of 78 on the promoted binary, mutation-proven (killed=1 not_reached=0, restore byte-identical at 71921 B). Unwrap an ARBITRARY mesh, not just the human anatomy layout the current unwrap is bound to: seam input, a conformal solve, and an angle-and-area distortion metric per chart. Done-rule: distortion on a developable test surface reads at or near zero, a known-difficult surface reports its distortion honestly rather than silently folding, and overlapping charts are REFUSED. MEASURED, and every fixture moved: the flat quad went ANGLE 987 permil to 0, the cube's twelve seams give exactly six charts with zero overlaps, the cut cylinder wall unwraps clean, and the CONE POINT reports 174 permil against a plane at 0 -- honest, non-zero and not a refusal, which is the only correct answer for a surface that cannot be flattened. THIS RUNG PREVIOUSLY PUBLISHED A WRONG ROOT CAUSE AND THE CORRECTION IS RECORDED HERE RATHER THAN QUIETLY REPLACED. The published cause was a discarded return code in uvi_unfold's walk; that is a REAL latent defect and was fixed too, but it caused NONE of the fifteen failures. The true cause was found by refuting the briefed one with arithmetic from the gate's own output: an unseamed 4-vertex 2-triangle mesh reported texc corners lost = 1 where a bijection makes 0 the only possible answer, so the corner-to-class map was corrupt BEFORE any unfold ran. In uv_chart_build's relabel block, uvf is BOTH the union-find parent forest AND the corner-to-class map, and the loop converts one into the other IN PLACE: from the second iteration uvi_find chases a class id as a parent index, and because that encoding is NEGATIVE it indexes before the array and path-compression-WRITES into the CSTART and CORDER slots beneath. Roots are now resolved into a dedicated array before any parent is rewritten. ALSO SHIPPED: UV_E_UNPLACED as its own refusal rather than overloading UV_E_NOPIN, because two causes sharing one code is exactly what made this hard to find -- and it is declared NOT bite-proven, since a mesh that passes validation cannot reach it from the public pipeline.
Matrix cell	general-uv-unwrap · `runtime/nx_uvunwrap_lib.nx` · PRESENT decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_uvunwrap_lib`; declare `nx_uvunwrap_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

DC6 Quad retopology from a sculpt to a game budget · cell WATCHING · gate NO-GATE-DECLARED

Contract	`rt_quadremesh` · executor Organ · 6 u · stands on DC3
Definition of done	Convert a dense sculpt into a quad-dominant mesh at a requested face budget while holding surface deviation under a stated millimetre bound measured by the EXISTING md_measure ruler, never a new one. Done-rule: the budget is met, the deviation bound is met and reported, and a mesh that cannot meet the bound is refused rather than silently exceeded.
Matrix cell	quad-retopology · `runtime/nx_mesh_qem.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_mesh_qem`; declare `nx_mesh_qem TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

DC7 Texture painting onto the regenerated PBR set · cell PRESENT · gate NO-GATE-DECLARED

Contract	`tp_stroke` · executor Organ · 4 u · stands on DC5
Definition of done	DONE 2026-08-25, nx_texpaint_gate GREEN 109/109 on the promoted binary and mutation-proven (killed=1 not_reached=0, restore byte-identical). All three clauses of the done-rule are met and each is bitten rather than asserted: the known surface point writes the known texel, and the expected address is computed BY ARITHMETIC IN THE FIXTURE rather than by the code under test; the seam does not tear, proven by neg-control-seam-TEARS-without-the-bleed which FIRES on a build with the bleed removed and is silent with it; and replaying the log reproduces the map BYTE-IDENTICALLY, with undo restoring the base byte for byte and redo returning to the exact same bytes. The anti-vacuity tooth that matters: a painter which wrote the whole atlas would trivially pass any known-texel-changed check, so the gate counts the disc area by hand (25 texels) and asserts ZERO texels changed outside the radius. The bleed margin is DERIVED from the brush radius and the chart width rather than picked. THE DEPENDENCY IS NOW MET: DC5 landed the same day at 78 of 78, so painting is no longer bounded to the chart layout the anatomy path supplies -- an arbitrary sculpt or boolean result can be charted, solved, packed and then painted, which is the first time this estate has had that path end to end.
Matrix cell	texture-painting · `runtime/nx_texpaint_lib.nx` · PRESENT decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_texpaint_lib`; declare `nx_texpaint_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

DC8 Multiresolution sculpt levels · cell WATCHING · gate NO-GATE-DECLARED

Contract	`ms_level` · executor Organ · 4 u · stands on DC6
Definition of done	Subdivide to a higher level, sculpt detail, and step DOWN without losing the detail -- the property that makes high-frequency sculpting tractable. Done-rule: detail added at level N survives a round trip to level N-1 and back within a stated deviation, and the level stack is an op on the DC1 log.
Matrix cell	multiresolution-levels · `runtime/nx_mesh_sculpt.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_mesh_sculpt`; declare `nx_mesh_sculpt TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

DC9 Sculpt brush constrained to an anatomical stratum · cell WATCHING · gate NO-GATE-DECLARED

Contract	`sc_layer_brush` · executor Organ · 3 u · stands on DC1
Definition of done	THE EXCEED, and it is only reachable because our bodies are generated inside-out: a brush that edits the MUSCLE layer and lets the skin re-emerge over it, or that adds fat without touching the muscle beneath. Done-rule: a stroke on the muscle layer provably changes the skin surface, a stroke on the fat layer provably does not change the muscle layer, and the resulting surface change is reported in millimetres by md_measure. ZBrush layers are shape blends over one surface and structurally cannot express this.
Matrix cell	layer-aware-sculpt-brush · `runtime/nx_mesh_sculpt.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_mesh_sculpt`; declare `nx_mesh_sculpt TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

DC10 Garment pattern, sewing and drape · cell WATCHING · gate NO-GATE-DECLARED

Contract	`gd_drape` · executor Organ · 8 u · stands on DC3
Definition of done	Flat pattern pieces, sewing seams, and a drape onto the generated body -- the Marvelous Designer capability the storefront lane needs, on the existing soft-body solver rather than a second one. Done-rule: a sewn garment drapes without self-interpenetration at a stated contact distance and its fit is reported against the body measurements the estate already extracts.
Matrix cell	cloth-and-garment-drape · `runtime/nx_softtissue.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_softtissue`; declare `nx_softtissue TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

DC11 Asset ingest under a provenance and licence gate · cell LANDED · gate NO-GATE-DECLARED

Contract	`ain_admit` · executor Organ · 2 u · stands on nothing
Definition of done	Every ingested asset carries where it came from, what licence admits it, and a content pin; an asset without them FAILS CLOSED at intake naming the missing conjunct, and a likeness requires a consent record. Done-rule: a planted asset missing each conjunct in turn is refused by name, and a complete one is admitted with its provenance readable afterwards.
Matrix cell	asset-ingest-under-provenance-and-licence · `runtime/nx_asset_admit_lib.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_asset_admit_lib`; declare `nx_asset_admit_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-07 measure Licence partition of the operator's 2026-09-06 scan and market references, read from the mirrored bytes: [ten24_samplescan] usage is commercial and personal WITH A CREDIT (full head and body scan, OBJ package with 16k textures and normal maps) = INGESTABLE as a DATA asset carrying its credit; [ss_free_head] is PERSONAL USE ONLY with a contact-us for commercial (8k albedo, normal, displacement) = evaluation oracle only; [ss_free_basemesh] [ss_multi_expression] [ss_free_metahuman] licence not yet read from the page = candidates, not assets; [booth_custom_model_adult] [jinxxy_human_avatars] are market bars with per-item paid licences = never an ingest source; [ss_blog_index] is the catalogue. 2026-09-07 queue Scan ingest recipe, all sovereign: fetch the ten24 OBJ package (S3 zip, size unmeasured; nx_research_fetch refuses a body that fills its reserve, so a streaming-to-disk fetcher may be the first build order) then nx_bundle_ingest <zip> source=https://ten24.info/sample-scan/ licence=ten24-credit (LIVE, REGISTERED-DARK, /compare/modding MD1) then nx_obj_import.nx (LIB, sovereign Wavefront reader) into an NXA with TEXC/TEXM from the scan's 16k albedo and normal maps, then the donor-TEXM lane onto the cast; the credit line ships beside the asset. 2026-09-07 lesson The critic loop the operator's pasted Infinigen-plus-Blender plan proposes (landmark oracle, Chamfer critic, bounding oracles for garments) already exists in NishiLang and is entirely DARK: nx_meshdist, nx_fitloop, nx_facemark, nx_bodyfit, nx_fbx2nxa, nx_part_solver and nx_bundle_ingest all read REGISTERED-DARK on 2026-09-06 (callable, authorised, never invoked over MCP). Blender and Python never enter the build path; the sovereign generator is judged by these organs with a licence-clean scan (ten24) as the oracle mesh, and the Daemon Girl asset itself is never ingested (mod licence forbids redistribution) -- it is the visual bar. 2026-10-01 queue CLAIMED by seat asset-ingest (cross-board lane, T0 1790830757, step 17 of the gameengine GE83 program): building ain_admit as the ONE intake admission, composing the incumbents rather than twinning them -- nx_asset_prov_lib for the sha256 content pin and the journal row, nx_licgate_lib for the rights table, nx_consent for the likeness record; every refusal names the missing conjunct; the gate carries a positive control, one tooth per conjunct asserting WHICH rule fired, an admit-everything stub neg-control and a same-size different-content pin tooth. DC15 follows as a reading of what is provable without a headset. 2026-10-01 queue RESUMED by seat asset-ingest (second launch, T0 1790857060; the first launch was cut off by an outage before it wrote any organ: func ain_ read ABSENT-PROVEN over buildroot/runtime at job 1790857425, coverage_complete=1 corpus_complete=1). Three sources landed this hour: runtime/nx_asset_admit_lib.nx (ain_admit, the one intake decision, composing nx_asset_prov_lib for the sha256 pin and the journal row, nx_licgate_lib for the one rights table and nx_consent for the likeness record), runtime/nx_asset_admit.nx (the command) and runtime/_hdl_build/nx_asset_admit_gate.nx. The command compiled first try at job 1790858814 (61307 B, sha256 e0d9b3771e8f2210f2b28d2fdb06b815997fe2bf24e32cd0d1eb8ed39efa68b3); the gate build is in flight. Nothing is promoted yet. 2026-10-01 land ain_admit is real: runtime/nx_asset_admit_lib.nx (18106 B) holds the ONE intake decision and composes the incumbents instead of twinning them -- the sha256 pin and the journal row are nx_asset_prov_lib (one journal, last row wins), the licence lookup is nx_licgate_lib (the one rights table knowledge/model_license.conf), the consent predicate is nx_consent. An asset is admitted only when its admission record (one key=value per line) names origin, a licence id the rights table carries, a sha256 equal to the sha256 of the bytes presented and a likeness declaration of 0 or 1, and a likeness also carries a consent record whose release is signed and whose term has not lapsed; every refusal returns a reason and the conjunct it belongs to (record, asset, origin, licence, pin, likeness, rights-table, consent, journal), a duplicated key is refused as ambiguous, a refusal writes nothing, and the same admission twice appends one row. PROMOTED: nx_asset_admit (the command, 61307 B, sha256 e0d9b3771e8f2210f2b28d2fdb06b815997fe2bf24e32cd0d1eb8ed39efa68b3) and nx_asset_admit_gate (134614 B, sha256 921ce5dee6106396522c6e8a40fcc86905ba275341d8c36a98b7f33e6daa44ef), both registered, both _offc twins installed, the gate rostered and added to dcc.gates; the dcc.matrix watch row is repointed from the glTF importer to runtime/nx_asset_admit_lib.nx so the contract symbol is measured where it lives. Shipped, referee grade 60 of 60 with the mutant killed, awaiting acceptance. 2026-10-01 measure nx_asset_admit_gate through /api/gate_run on the promoted binaries (job 1790859921, exit_code 0): passed 60 of 60. Values the gate printed, each recomputable: refusal_cases=17, refusal_cases_exited_refuse=17, real_admission_cases_satisfied=17, admit_everything_stub_cases_satisfied=0, journal bytes 134 after the first admit and 134 after all seventeen refusals (by hand: the tag 5 plus sha 64 plus source 29 plus licence 10 plus lane 10 plus epoch 10 plus five tabs and a newline is 134), the CRLF record 143 bytes against 139 for the LF record. Second method classes: the pin is checked against the FIPS 180 known answer for the three bytes abc and not against itself; the provenance row is read back through pv_row_find and not through the admission receipt; the promoted command, forked on the same journal, exits 0 with row=present for the admitted asset and exits 4 naming reason=PIN-MISMATCH conjunct=pin for the same-size different-content twin. nx_gate_bite over the lib (job 1790859987): valid_mutants=1 killed=1 not_reached=0 void_build_lane=0, restore byte-identical at 18106 B, twins_checked=2 restored=2 mutant_live=0, and nx_filehash cmp reads the live gate and its _offc twin identical at 921ce5de afterwards. One mutant is a floor on the gate's bite, not a mutation score. 2026-10-01 lesson An absence probe for a contract symbol must be asked as a DECLARATION, never as the bare name: nx_absent for ain_admit over buildroot/runtime read matches=75 PRESENT, and every hit shown was domain_admit, whose last nine letters are the contract symbol. The same probe asked as func ain_ read ABSENT-PROVEN at coverage_complete=1. A bare-name probe would have told the next seat the rung had already been built. Second lesson, about the board itself: the watch row named the glTF importer as the organ the symbol would live in, and a decision every intake door must share cannot live inside one of the doors; the row was repointed to the lib rather than a wrapper being planted in the importer to light the cell. 2026-10-01 queue OPEN HALF OF THE RUNG, named so the lit cell is not read as the whole of it: (1) no intake door calls ain_admit yet. nx_gltf2mesh, nx_fbx2nxa and nx_bundle_ingest still import anything; wiring each means an admit=<record> argument that refuses by name, their own gates gaining a refused-without-a-record tooth, and a decision on house assets (the export door nx_mesh2glb passes a row-less asset as a house asset; an intake door that fails closed needs the same question answered before it is switched on, or the importer gates that feed fixtures break). (2) The export doors read pv_verdict only; a row whose origin column says dcc-ingest likeness consent=<id> is not yet read by any delivery door, and tier and use scope are asked nowhere after intake. (3) The consent record is carried inside the admission record as declared fields; nx_rights_ledger (grants per talent and use scope, in memory only, its persistence rung unbuilt) is the incumbent a durable consent store should grow from. (4) OPERATOR DECISION OWED: this rung's text refuses a likeness without a consent record at intake, which is what was built; the 2026-09-30 likeness policy builds reproductions inside the estate flagged estate-only and puts the line at the production door. Whether intake should admit a consent-less likeness under an estate-only flag instead of refusing it is not a seat's call and was not guessed. Recipe to verify what shipped: nx_mgmt POST /api/gate_run target=nx_asset_admit_gate, then nx_asset_admit admit <asset> <record> on a real asset with the pin from nx_filehash.

DC12 One asset, four consumers, one ruler · cell PRESENT · gate NO-GATE-DECLARED

Contract	`xp_consumers` · executor Organ · 3 u · stands on DC3, DC11
Definition of done	DONE 2026-08-25 FOR THREE CONSUMERS WITH THE FOURTH ABSTAINING BY NAME, and the done-rule is AMENDED HERE rather than quietly satisfied. nx_xport_gate GREEN 38/38, mutation-proven, deviations recovered OUT OF THE ARTIFACT BYTES and reported per consumer in tenths of a millimetre: game NXA reads 0/0/0 and clinical NXMSH2 reads 0/0/0, both EXACTLY lossless against a derived bound of 0 -- by ARITHMETIC and not by tolerance, because their unit is an integer multiple of the ruler's, so the forward map is a multiplication and the inverse an exact division. Storefront glb and obj read 0/4/4 against a derived bound of 11. THE ORIGINAL DONE-RULE SAID ALL FOUR AND THAT WAS THE WRONG BAR: vf_build takes NO MESH ARGUMENT AT ALL -- its entire scene is a six-row constant table and it emits a PNG -- so VR geometry could not be recovered even if a mesh could be fed in. That is a missing capability in the compositor, not a shortfall in this proof, and it is now DC15 with a concrete contract. The amended rule: every consumer that CAN receive the asset is measured, and every one that cannot ABSTAINS BY NAME with its reason re-checked against live source on every run, so an abstention can never quietly decay into a pass. THE FINDING WORTH MORE THAN THE RUNG: the storefront is our ONLY lossy consumer and the loss is STRUCTURAL -- write_glb and write_obj hold coordinates at fx1024, a quantum of 0.977 mm, which is COARSER than the 1 to 2 mm bar nx_mmdev_lib's own header says the clinical unit exists to serve. Measured 0.4 mm on this asset, 1.1 mm derived worst case. An asset that is clinically exact going in is not clinically exact by the time it reaches a storefront viewer, and nothing before this rung could see that. A second tooth earned its place the same way: orientation was first checked by per-triangle normal dot product, which CANNOT catch a reflection, and the replacement signed-volume test carries the case that proves it -- a mirrored box reads max deviation 0 while vol6 flips sign.
Matrix cell	one-asset-four-consumers-one-ruler · `runtime/nx_xport_lib.nx` · PRESENT decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_xport_lib`; declare `nx_xport_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

DC15 A mesh can reach the VR compositor at all · cell WATCHING · gate NO-GATE-DECLARED

Contract	`vf_build_mesh` · executor Organ · 6 u · stands on DC12
Definition of done	SPLIT OUT OF DC12 2026-08-25 BECAUSE THE MEASUREMENT NAMED A MISSING CAPABILITY RATHER THAN A GAP IN THE PROOF. runtime/_hdl_build/nx_vrframe.nx builds a submit-ready stereo frame with an off-axis pinhole per eye and a barrel-distortion pass, and it is real work -- but vf_build takes only a distortion flag and an output path, its scene is the six-row vf_pill constant table, and it rasterises nothing triangular. Contract: vf_build accepts a mesh and a pose, a triangle rasteriser draws it into both eye views, and the geometry the compositor received is recoverable so DC12's existing ruler can measure it exactly like the other three. Done-rule: nx_xport_gate's VR row moves from UNREACHED to a measured deviation against the same derived bound, and the two source-signature teeth that currently PIN the abstention flip from asserting absence to asserting presence. Until then the VR half of the storefront-game-VR-clinic claim is a plan and this board says so.
Matrix cell	vr-compositor-mesh · `runtime/_hdl_build/nx_vrframe.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_vrframe`; declare `nx_vrframe TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

DC13 Node-graph authoring over the operation stack · cell WATCHING · gate NO-GATE-DECLARED

Contract	`ng_eval` · executor Organ · 5 u · stands on DC2
Definition of done	A graph whose evaluation IS the DC1 op stack rather than a second engine -- one evaluator, so a graph edit and a manual edit cannot diverge. Done-rule: a graph and the equivalent hand-built op log produce byte-identical state, which is the tooth that proves there is only one evaluator.
Matrix cell	node-graph-authoring · `runtime/nx_editstack_lib.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_editstack_lib`; declare `nx_editstack_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-06 lesson THE BAR MOVED WHILE THE BOARD SLEPT (2026-09-06, the current-month rule applied to this board): the ranker REFUSED dcc (BARFRESH bars=5 malformed=5) because its five bars carried no month; this month's reads moved three of them -- Blender 5.0 to 5.2 LTS with node-based physics (DC13's node graph is now the incumbent's shipping shape, so DC13 must price against a physics-capable graph, not a modifier stack), Womp to an SDF and CSG parametric modeler in the browser (a rival standing on OUR representation axis, so the generated-asset exceed on DC1 and DC3 must be measured against it, not asserted), Maya to 2026.3 with OpenPBR default, ZBrush to 2026.2.1. Two vendor domains (autodesk.com, adobe.com) fail the sovereign TLS chain, filed as a recurrence on the sovereign-tls board; their bars are witnessed by encyclopedia mirrors and say so on the row. Every bar row is now bar, id, YYYY-MM, subject, what, ref, note in that order; the ruler reads FRESH only if it is.

DC14 Collaborative multi-seat editing on one operation log · cell WATCHING · gate NO-GATE-DECLARED

Contract	`co_merge` · executor Organ · 6 u · stands on DC1
Definition of done	Two seats editing one asset converge to the same state regardless of the order their operations arrive, on the append-only log DC1 already provides. Done-rule: a gate replays two interleaved op streams in both orders and proves the final digests are identical, with a neg-control proving a genuinely conflicting pair is REFUSED rather than silently resolved.
Matrix cell	collaborative-multi-seat-editing · `runtime/nx_editstack_lib.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_editstack_lib`; declare `nx_editstack_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

DC16 Auto-rig arbitrary topology: skeleton fit and skin weights without a human rigger · cell PRESENT · gate NO-GATE-DECLARED

Contract	`amesh_run` · executor Organ · 3 u · stands on DC2
Definition of done	THE LOAD-BEARING GAP FOR PROCGEN AND NISHIGEN, and it is not an authoring convenience. A parametric head (GNM, Apache-2.0) ships with its rig baked in because its topology is fixed, but procedural bodies and creatures have ARBITRARY topology -- so a generator that makes 10,000 creatures makes 10,000 UNUSABLE meshes until this rung lands. Substrate already exists and is dark: nx_skelgen, nx_nxa_skin, nx_skinwrap, nx_rigfloor, nx_dyna_bind_skin. Done-rule: given an unrigged mesh, emit a skeleton plus per-vertex weights; the bound mesh deforms at an extreme pose without candy-wrapper collapse, measured through md_measure (NEVER a second mesh ruler -- see this board's own risk row); a neg-control mesh that is not the requested skeleton archetype is REFUSED BY NAME rather than mis-rigged in silence, because a confidently wrong rig is worse than a refusal the caller can see. Reference implementation to study, never to vendor: Mesh2Motion (mesh2motion-app, 3,212 stars, GLB in and GLB out, fit-skeleton-to-mesh then retarget) -- the ALGORITHM is the oracle, the TypeScript never enters the build path
Matrix cell	Auto-rig arbitrary topology in ONE CALL: bare mesh in, skeleton and skinned rig out · `runtime/nx_autorig_mesh.nx` · PRESENT decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_autorig_mesh`; declare `nx_autorig_mesh TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-03 measure Mesh2Motion verified 2026-09-03 via the GitHub API, not recalled: mesh2motion-app 3,212 stars, TypeScript, pushed 2026-08-31 (live), and its SIBLING mesh2motion-assets is CC0-1.0 -- public-domain animations and models, which classify `shipped` under the mediaingest R21 provenance conf and are therefore usable as a retarget corpus with zero provenance risk. CAUTION RECORDED: the marketing site states MIT but GitHub's classifier reads NOASSERTION on the app repo, so read the LICENSE file before relying on it -- the same discipline that caught EpicGames/MetaHuman-DNA-Calibration being proprietary rather than open. 2026-09-03 lesson MetaHuman is SOTA for film-quality face rigs and is NOT a digital twin -- it carries no muscle physiology. So the exceed axis against Epic is anatomical and biomechanical grounding, where the referees are OpenSim (opensim-org/opensim-core, Apache-2.0) and MyoSuite (MyoHub/myo_sim, Apache-2.0, musculoskeletal models in MuJoCo) and Epic has nothing to show. That axis is anatomy AN9-AN12's lane; DC16 is the rung that lets a generated body ENTER it. 2026-09-05 land SHIPPED nx_boneheat and nx_boneheat_gate 2026-09-05, both PROMOTED and REGISTERED on the NAS, gate 15 of 15 GREEN. The skin-weight half of DC16 is met, geodesic bone-heat auto-weights over the voxelised mesh volume exceeding Blender Automatic Weights [blenderboneheat]; the skeleton-fit half stays nx_skelgen and nx_autorig. Matrix row bh_run landed on dcc.matrix. 2026-09-05 measure On the gate fixtures the geodesic solve leaks 0 of 84 weights across 12 mm of air where the nearest-joint Euclidean control leaks 16; it blends at every joint (4 of 4 middle-ring corners carry two joints), falloff is monotone both directions, NXA weights sum to 4096 exactly, two solves are byte-identical, and a mesh with no SKEL is refused by name. 2026-09-05 lesson A tilted-face octagon test tube would not voxel-seal (interior 0) because flat faces point-sample into a 26-connected shell a 6-connected flood slips through; an axis-aligned prism seals by construction and the lib keeps a one-pass 6-dilation for real tilted meshes. Fixture geometry is part of the test, not scaffolding. 2026-09-05 queue SKELETON-FIT SOURCE LANDED 2026-09-05, the other half of DC16: nx_curveskel_lib (geodesic farthest-point curve-skeleton peeling that COMPOSES nx_boneheat_lib's voxeliser and Dijkstra, no learned model, no training set), nx_curveskel CLI, nx_curveskel_gate. The gate ran 8 of 11 under the storm and the passes are the load-bearing ones: the skeleton SPANS the extremities (joint z from 283 to 5890 of 6000), the tree is single-rooted with no dangling parent, two runs are byte-identical, a volumeless sheet is refused by name, and the medial radius scales 30 to 100 with thickness. Two fixes then landed in source: a bone now CLAIMS its cross-section (a tube of its medial radius) so a bar no longer over-peels into 32 false limbs, and the root is the thickest cell nearest the interior centroid so a bar roots at its centre and peels both ends. Build admission QUEUEd the rebuild on run-queue saturation. RECIPE for the next seat: nx_sov_build_run nx_curveskel_gate (expect 11 of 11), nx_organ_ship nx_curveskel_gate, nx_organ_ship nx_curveskel, then a matrix row on this board bound to cs_run beside bh_run, then the OWED tooth: a T-branch fixture built watertight via nx_csg proving 3 limbs and a fork joint. 2026-09-06 measure SKELETON-FIT PROVEN 2026-09-05: nx_curveskel_gate 11 of 11 GREEN on the fixed source (sovereign build-and-run), then PROMOTED and REGISTERED through the ship loop. Values on a 60 mm square prism: limbs=2 branches=1 joints=32, joint z spans 157 to 5890 of 6000, root radius 30 (3 mm wide) versus 100 (7 mm wide), two runs byte-identical, a volumeless sheet refused exit 3. The three fixes that took it from 8 of 11: a bone claims its cross-section tube, the root is the thickest cell nearest the centroid, and descent follows the medial axis. Build admission refused six paced re-issues (run queue up to 13 of 8, D-state 7 of 8) before granting; the sovereign lane, not the mgmt lane, produced the proof. 2026-09-06 land SKELETON-FIT SHIPPED 2026-09-05: nx_curveskel PROMOTED and REGISTERED (57,144 B) after its gate, proving against the live nx_curveskel_gate; matrix row cs_run landed beside bh_run with the Rigify manual pinned as the reference that does not attach to a mesh [blenderrigify]. DC16 is now met in BOTH halves for an unrigged mesh, a skeleton from the volume then geodesic weights, with the T-branch fixture the remaining owed tooth, and referee grade plus operator ACCEPT still owed before either half reads complete. 2026-09-06 land T-BRANCH TOOTH CLOSED 2026-09-05: nx_curveskel_gate 16 of 16 GREEN. The watertight T (stem UNION crossbar via nx_csg marching tetrahedra: 120,924 verts, 40,308 tris, 23,892 interior cells at a 0.38 mm cell) peels EXACTLY 3 limbs with 2 fork joints and 29 joints, reaching x -1405 and +1407 (the 15 mm crossbar tips) and z 95 (the stem foot); the linear prism still peels 2 limbs with 1 fork. The sovereign build twin-refreshed the served gate and the ship loop was re-run to bank the harness manifest. DC16's skeleton-fit half is now proven on both a linear and a branched volume; referee grade and operator ACCEPT remain owed. 2026-09-06 queue WHAT DC16'S DONE-RULE STILL LACKS, named so the next seat builds it and not something else: the rule says the BOUND mesh must deform at an EXTREME POSE without candy-wrapper collapse, measured through md_measure. Skeleton (cs_run) and weights (bh_run) are proven on fixtures; the pose leg is not. RECIPE: an end-to-end gate that takes a bare fixture, runs nx_curveskel then nx_boneheat, poses the result through nx_nxa_play at an extreme joint angle, and measures the deformed skin against the rest pose with md_measure -- a candy-wrapper collapse shows as a cross-section area loss at the joint; assert the loss stays under a bar derived from the medial radius, and carry a neg-control that nearest-joint weights DO collapse. Then acceptance rows ga_accept_boneheat and ga_accept_curveskel sit beside both mechanical rows and flip only through the accept gate. 2026-09-06 lesson CANDY-WRAPPER IS A DEFORMER PROPERTY, NOT A WEIGHTS PROPERTY, and the done-rule as written cannot be met by weights alone. Measured 2026-09-05: nx_skeleton (the evaluator nx_nxa_play composes) has no dual-quaternion path -- it is linear blend skinning only. Under a 90 degree BEND, smooth geodesic weights preserve the joint better than a hard boundary and that case is provable through the pose leg. Under an extreme TWIST about the bone axis, linear blending of two opposed rotations collapses the blended ring toward the axis -- and smoother weights make the collapse WORSE, not better; the field's answer since Kavan 2007 is dual-quaternion skinning, a change to the DEFORMER. So the pose tooth proves the bend, and the twist half of this done-rule is a dual-quaternion rung owed on the board that owns nx_skeleton (graphics owns the rig per anatomy.matrix); pricing it here would be the C7 double-pricing defect. Acceptance for DC16 therefore reads: bend proven, twist deferred to the deformer rung, named not hidden. 2026-09-06 land THE BEND DONE-RULE IS PROVEN END TO END 2026-09-05: nx_autorig_pose_gate PROMOTED (140,976 B, sha f77a1361) and REGISTERED, 10 of 10 GREEN through nx_job_run on the served binary. A bare 60 mm prism goes through nx_curveskel then nx_boneheat (geodesic and the nearest-joint control), its whole shoulder subtree (16 of 32 joints) is swung 90 degrees as a baked world-delta POSE, the SERVED nx_nxa_play plays it, and md_measure -- the one mesh ruler -- judges the deformed skin against the rest pose. Gates row landed on dcc.gates. Referee grade and operator ACCEPT remain owed: this is presence, not completion. 2026-09-06 measure Three independent witnesses agree on the swing. The player's own receipt: joints_posed 16, max displacement 3194 umm (31.9 mm per axis), verts moved 245 of 246 under geodesic weights versus exactly 121 under nearest (the smooth falloff touches the whole body, the hard boundary moves one half). md_measure posed-vs-rest: max 259, p95 229 in 0.1 mm units (25.9 mm and 22.9 mm) against a 10 mm bar, identical for both weightings; geodesic-vs-nearest max 2. The crease read across adjacent 1 mm rings: geodesic 162 umm strictly softer than nearest 446. Two plays byte-identical; the unweighted asset refused exit 5. 2026-09-06 lesson A ONE-JOINT POSE MOVES A SLEEVE, NEVER A LIMB, BECAUSE nx_nxa_play WALKS NO HIERARCHY. The player applies M x = D(x - bind_t) + bind_t + dt per joint (nxa_format_spec.md, exact LBS with no runtime forward kinematics), so an entry on one joint deforms only that joint's skinned band: the first run read md max 3 against a 10 mm bar while the ring crease read 5 mm. A real bend is one entry per joint of the swung subtree, D equal to the rotation and dt equal to (R - I)(bind_j - P), which is the rigid swing R(v - P) + P about the pivot P -- baked exactly as an animation clip stores it. The gate composes nf_qrotv so the swing it encodes and the swing the player renders are one arithmetic. 2026-09-06 lesson A BAR IN THE WRONG UNIT FAILS A CORRECT RESULT. The second run swung the limb 26 mm and still read RED because the bar was written as 1000 in 0.01 mm quanta while md_measure reports in 0.1 mm for a mesh fed in millimetres: the twist sleeve had read 3 and the swing read 259, so the unit was decidable from the two numbers already in hand. The bar is 100 and its provenance is now in the gate header; the values are emitted with their unit in the key name so no reader inherits the assumption. 2026-09-06 queue WHERE THIS RUNG STANDS AGAINST THE FIELD, AND THE LADDER TO SOTA, so the next seat builds the next rung and not a side artifact. STANDING: the skin-weight half is in the Maya geodesic-voxel-bind class and past Blender bone heat on the air-bleed axis [blenderboneheat]; the skeleton half is a first rung, a topological curve skeleton with no joint semantics, proven on a prism and a CSG T only; the deformer is linear blend, no dual quaternion. RUNG M2, THE REFEREE-GRADE BENCHMARK: re-rig the estate's own rigged donors (toon3d8, witch, paladin from the modding lane) with nx_curveskel and nx_boneheat and score joints against the artist SKEL (position error after nx_nxa_retarget mapping) and weights against the artist SKIN under the donor's own POSE library through md_measure -- the number a SOTA claim needs, since UniRig [unirig], RigNet [rignet] and MagicArticulate [magicarticulate] are all scored on real characters. RUNG M3: dual-quaternion skinning in the deformer (graphics owns nx_skeleton) so the twist half of this done-rule can be proven. RUNG M4: semantic joint labels by composing nx_nxa_retarget's topology-plus-bind-position mapping onto the curve skeleton, then IK on the labelled rig (Rigify parity for posing) [blenderrigify]. RUNG M6: the learned refinement modding MD4 names, which must beat the geodesic control on the M2 benchmark. ALSO OWED HERE: nx_gate_bite of nx_autorig_pose_gate against nx_boneheat_lib with subject nx_boneheat (mutant builds were refused by the admission storm today), the referee grade, and the operator ACCEPT rows ga_accept_boneheat and ga_accept_curveskel. 2026-09-06 land ONE-CALL AUTO-RIG SHIPPED 2026-09-05: nx_autorig_mesh <bare.nxa> <skel_out.nxa> <skinned_out.nxa> [cells] composes nx_curveskel then nx_boneheat (geodesic) in a single invocation, PROMOTED (77,240 B, sha 16632a8e), REGISTERED, _offc twin installed. Its gate nx_autorig_mesh_gate is 4 of 4 GREEN (PROMOTED, REGISTERED): the one-call skeleton and skinned rig are BYTE-IDENTICAL to the two steps run separately, every vertex's weights sum to 4096, and a volumeless sheet is refused by name at the skeleton stage. Matrix row amesh_run and a dcc.gates row landed. 2026-09-06 lesson THE RANKER READ THREE SHIPPED ORGANS AS PRIORITY ZERO. nx_compare_rank dcc reported DC16 UNMAPPED / NO-MATRIX-ROW because the rung's contract symbol ar_autorig had never been built -- the matrix rows carried bh_run, cs_run and the gate's own symbol, none of which the rung named. A rung whose symbol nothing implements is invisible to the scoreboard no matter how much ships beside it. Fixed at the root: the rung symbol is now the real entry point amesh_run, carried by the nx_autorig_mesh matrix row; the re-run ranker reads DC16 done (rungs done 7 to 8) with one honest residual, REGISTERED-UNAUTHORISED, closed by minting a cap. Encode value where the ruler looks: a rung symbol must name a symbol that exists. 2026-09-06 queue ARCHETYPE-FREE PROOF IN FLIGHT (operator 2026-09-05: quickly auto rig a dragon vs a human vs whatever, everything-is-math): nx_autorig_creature_gate builds a dragon-like quadruped (body + four legs + tail + neck) and a biped (torso + two legs + two arms + head) as watertight nx_csg_scene unions, runs the SAME one-call auto-rig on both with no archetype hint, and asserts each peels the limb count it was built with (six vs five, derived from construction), every limb tip is reached by a joint, both skins are sound, and the two counts DIFFER (the anti-vacuity tooth). Source landed (16,263 B); its build was refused twice by admission (load 2273 against the 1643 ceiling) and is queued through the mgmt lane. THE RUNG AFTER IT: a generated creature already carries a part graph, so the SDF-part skeleton (nx_autorig ar_build, Prim MST by surface gap) and the volume curve-skeleton must agree on limb topology -- two independent method classes agreeing is the PROVEN bar, and it binds the generate-up stack (procgen/anatomy) to this rung. 2026-09-06 measure ARCHETYPE GATE FIRST RUNS 2026-09-06 (nx_autorig_creature_gate, sovereign build-and-run): DRAGON EXACT -- a body with four legs, a tail and a neck peeled exactly 6 limbs (54 joints, 5 forks) from the volume with no hint, every leg quadrant, the tail (-64 mm) and the neck (+30 mm) reached, skin sound. BIPED 7 LIMBS, NOT 5: two legs, two arms and the head are all reached (T8 PASS), plus two extra limbs. A first reading blamed the two 3 mm shoulder ledges outboard of a 12 mm head; the discriminating experiment REFUTED it -- a flush-head biped (head as wide as the torso, no ledges) also peels 7 (54 joints), so the shouldered and flush variants are equal and the extras are elsewhere. The story was plausible and wrong; the instrument that settles it is a per-leaf position readout (every joint no other joint parents = where each limb ends), now in the gate and queued behind a load-21 storm. Not journaled as a defect until the leaves say where the two limbs point: a thin 14 mm torso with a 6 mm root radius makes its own corners and ledges protrusions by the volume's arithmetic, and that may be a correct rig of the fixture, not a wrong one. 2026-09-06 measure THE LEAF READOUT SETTLED IT, AND IT WAS NEITHER SHOULDERS NOR CORNERS. Dragon leaves (mm): tail (-61,-2,6), neck (35,-2,28), four feet at z=-33 -- exactly its six. Biped leaves: right foot (8,3,-51), both arm tips (+-32,-2,15), head (-6,5,29), left foot (-8,-3,-51), and two MORE inside the left leg on the same corner column at z=-35 and z=-25. The right leg peeled once, the left leg three times, identical with or without shoulder ledges. The only asymmetry between two mirrored legs is the plane x=0 they share: two adjacent boxes in an SDF union put a zero-field plane exactly on a grid plane, the marcher emits a one-sided internal wall there, and one limb's first peel under-claims its cross-section so the peeler keeps finding far cells down its own corner. The dragon's legs are 6 mm apart and never touch. Two stories were refuted by instruments before the third was read off one (clavicles, then corners); the fixture is corrected to separated legs as the discriminating experiment, and the peeler's robustness to TOUCHING limbs is named as its own rung rather than hidden in a fixture tweak. 2026-09-06 queue PEELER ROBUSTNESS TO TOUCHING LIMBS (the rung the leaf readout named): two limbs whose boxes share a face produce, in an SDF union marched on a grid that contains that plane, a one-sided internal wall; one limb's first peel then under-claims its cross-section and the same limb is peeled again down its own corner column (measured: left leg three times, right leg once). Real characters have touching limbs everywhere -- thighs, fingers, an arm against a torso -- so this is not a fixture curiosity. Two remedies to measure against the touching-legs fixture, which stays in the gate as the neg-control that must go from 7 to 5: (a) in cs_claim_tube, claim the cube from the LARGEST medial radius within a one-cell neighbourhood of each path cell rather than the path cell's own value, so a path that hugs a wall still claims the limb it runs through; (b) in bh_raster, a zero-thickness double-sided wall between two interior regions is a shell of measure zero and should not survive the dilation-and-flood as a partition. Accept rule, declared before the experiment: the touching-legs biped peels exactly 5, the separated-legs biped still peels exactly 5, the dragon still peels exactly 6, the T still peels exactly 3, and nx_curveskel_gate stays 16 of 16; a remedy that passes only by loosening the stop bar is refused. 2026-09-06 land ARCHETYPE-FREE AUTO-RIG PROVEN 2026-09-06: nx_autorig_creature_gate 11 of 11 GREEN (sovereign build-and-run, 112,870 B). The SAME one-call nx_autorig_mesh with NO archetype hint rigs a dragon-like quadruped and a biped from their volumes alone: dragon exactly 6 limbs (leaves at the tail tip -61 mm, the neck +28 mm, four feet at -33 mm), biped exactly 5 (two feet at -51 mm, two arm tips at +-32 mm, head +29 mm), every limb tip reached, every SKIN summing to 4096, the two counts differ, and the shouldered biped peels the same 5 (3 mm ledges sit under the 6 mm stop bar) so the monotone tooth holds. This is the operator's everything-is-math requirement met on fixtures: dragon vs human from one arithmetic. The discriminating experiment that made it honest: separating the biped's legs by 4 mm took it from 7 limbs to 5, confirming the shared plane between touching boxes as the whole cause of the earlier miscount. Gates row landed on dcc.gates. NOT CLAIMED: real characters (M2 donor benchmark owed), touching limbs (rung above, accept rule pre-declared), referee grade and operator ACCEPT. 2026-09-06 measure REMEDY (a) REFUTED 2026-09-06: the relaxed claim (a bone claims by the thickest medial radius in its first cube) rebuilt the creature gate to 11 of 12 -- dragon 6, separated biped 5, shouldered 5, nx_curveskel_gate byte-IDENTICAL 16 of 16 (nothing regressed) -- and the touching-legs biped STILL peeled 7 with the x=-8 leg claimed three times (tips at -51, -35, -25 mm). The instrument that named the real cause was a vertex count: the touching mesh carries 148,080 vertices against 138,624 for the separated one, 9,456 more from legs 4 mm CLOSER -- a zero-thickness double wall the marcher emits where two boxes abut at SDF 0. The claim relaxation is reverted (a remedy that fixes nothing is a magic rule) and the wall is attacked where it lives, in bh_flood. 2026-09-06 measure REMEDY (b) FIRST CUT, WALL STRIP AT A DERIVED DEPTH OF 3: bh_flood now walks six-steps from the exterior through surface cells and re-classes any surface cell deeper than the shell's derived depth (raster 1 + dilation 2) as INTERIOR. Measured on the creature gate: dragon still exactly 6 limbs, but its interior grew 30,282 to 40,171 cells and its root radius 110 to 120; the separated biped's root radius went 60 to 112 -- LARGER than its own 7 mm torso half-thickness, which no honest medial radius can be -- and it lost the head (5 to 4 limbs, T7/T8 RED); the touching biped read 3 (both legs ONE block once the wall is gone, which is what two abutting boxes ARE in a union). The strip removed about one real shell layer (the dragon's gain is about one 13k-cell layer), so the derived 3 is a hypothesis the data refutes: the shell is deeper than raster-plus-dilation on bevelled marching-tetrahedra faces. 2026-09-06 lesson THREE LAWS FROM ONE FIXTURE. (1) A derived constant is a hypothesis until a census confirms it: bh_flood now PRINTS a shell-depth census (surface cells by six-steps from the exterior) on every flood, so the strip depth is read off the data, never re-derived on paper. (2) Two boxes sharing a whole face are ONE solid in an SDF union; a neg-control for TOUCHING limbs must carry the groove real thighs have, or it tests the marcher's wall, not the peeler. (3) The earlier 11 of 11 counted the flush biped's head as a limb only because internal walls halved the torso's medial radius to 6 mm; with the walls gone the head protrudes about 8 mm past an 11 mm half-thickness and the derived stop rule declines it -- so that GREEN was a measurement of the defect, not of the rig, and the fixture head must protrude past the torso half-thickness to be a limb by the rule's own words. 2026-09-06 queue IN FLIGHT: creature gate rebuilt with the census flood (paced re-issues under admission); read the d1..d9 histogram on the dragon (no abutting boxes: the shell's true depth is where the counts collapse) and the touching biped (a long tail is the wall), set BH_WALL_DEPTH from that measurement with the derivation written beside it, then re-run: accept rule stands (dragon 6, T 3, nx_curveskel_gate 16 of 16, never by loosening the stop bar) and the biped fixtures are re-derived from the honest volume -- a head taller than the torso half-thickness, and a touching-legs control with a groove. ALSO BUILT THIS SESSION, awaiting the box: nx_rigbench (M2 referee, lib + CLI, 99,693 B compile) and nx_rigbench_gate (116,598 B) -- skeleton coverage in the artist's own bone units, weight L1 on the artist skeleton, and the artist POSE played by the served nx_nxa_play with both weightings judged by md_measure, on the house rig sites/nishifamily/world/ref9d.nxa. 2026-09-06 measure THE CONTROL INDICTED THE INSTRUMENT, NOT THE FIXTURES (2026-09-06). A 26-connected walk still read 7,635 dragon cells deeper than 3, so the census was run on nx_curveskel_gate's lone 60 mm prism -- one box, no internal geometry, the cleanest control in the estate: d1=1352 d2=2136 d3=2424 d4=2280 d5=2136 d6+=0, 4,416 false walls, and the EXTERIOR region only 472 of 14,112 cells. A census that RISES with depth on a convex box means the walk was not starting from the outside: BH_PAD_CELLS was 1 while bh_dilate grows the shell outward by 1, so wherever a face touches the bounding box the pad is surface, the exterior flood has no seed there, and depth was being measured around the shell from a few corner pockets. The same starvation un-refused the sheet neg-control (T9/T10 RED: 24,300 cells re-classed on a flat quad). The T (overlapping stem and crossbar, no abutment) still peeled exactly 3, the prisms 2 and 2 -- the skeletoniser was never the problem. FIX, derived not tuned: BH_PAD_CELLS = 1 + BH_DILATE_STEPS = 2, which shifts the grid origin by a whole cell and so changes no cell boundary relative to the mesh (results should reproduce byte-for-byte with the strip finding nothing on wall-free fixtures). 2026-09-06 lesson A DEPTH MEASURED FROM A REGION THAT DOES NOT EXIST IS A DISTANCE TO NOWHERE, AND IT LOOKS EXACTLY LIKE THICKNESS. The exterior flood only needs ONE seed to seal a volume, so a starved exterior passed every sealing gate for weeks; it failed the first instrument that asked how FAR the outside is. Print the size of the region you flood from beside anything you derive from it -- exterior=472 on a 14,112-cell grid was the whole diagnosis, and it was one number the flood already had. 2026-09-06 measure PAD FIX CONFIRMED BY THE CONTROL (2026-09-06, nx_curveskel_gate rebuilt 109,984 B sha cff7a381, staged run): with BH_PAD_CELLS=2 every wall-free fixture reads d4+=0 and walls=0 -- thin prism d1=4488 d2=3560 d3=2280, fat prism 10442/9306/7714, sheet 19196/8836/0 (refused by name again), T 11634/10113/7557 -- and every banked value REPRODUCES EXACTLY: thin root radius 30 and fat 100, joint z 157 to 5890, T interior 23,892, three limbs, two forks, 29 joints, tips at -1405 and +1407, foot at 95. 16 of 16 GREEN. So the derived shell depth of 3 (raster 1 + dilation 2) is now a MEASURED fact on four wall-free volumes, the strip is a no-op wherever there is no wall, and a grid-origin shift by a whole cell changes nothing about a result -- the accept rule's T-3 and 16-of-16 conjuncts hold under the remedy. 2026-09-06 retract RETRACTED, same day: the lesson row above that read the flush biped's five limbs as 'a measurement of the defect' (the head only counting because internal walls halved the torso radius) was WRONG. With the pad fixed the census reads d4+=0 on both bipeds and the ORIGINAL values reproduce exactly -- separated biped interior 15,994, root radius 60, five limbs with the head at (-6,5,29), shouldered five -- so the head was always a limb by the derived stop rule and the four-limb reading came from the starved-exterior strip inflating the medial radius to 112. The discriminating experiment was the lone-prism control, not any argument. Same family as every retraction this estate records: a plausible story published before its control. 2026-09-06 land TOUCHING-LIMBS RUNG MET UNDER ITS PRE-DECLARED ACCEPT RULE (2026-09-06, nx_autorig_creature_gate 12 of 12 GREEN, staged build 135,392 B sha af34fa47, run through a scratch registration of the stage): the biped whose two legs SHARE the plane x=0 now peels exactly five (feet at (8,3,-51) and (-8,-3,-51), both arm tips, head), its wall reads d4=242 d5=230 d6=218 d7=200 in the 26-step census and is stripped (walls=890, root radius 60); separated biped 5, shouldered 5, dragon 6, the T 3, nx_curveskel_gate 16 of 16, and the stop bar untouched. Remedy = bh_flood strips surface cells deeper than the derived shell depth once the exterior has a real seed (BH_PAD_CELLS 1 to 2). HONEST RESIDUAL: the two abutting legs still peel as two tubes because the claim radius follows the block's thinner axis (10 mm of a 20 x 10 block), not because the peeler sees a crease -- a grooved-thigh fixture is the true control for REAL touching limbs and is owed; the walls= count and the census line are now printed on every flood so the next reader sees whether a mesh carried walls at all. 2026-09-06 land SHIPPED 2026-09-06: nx_curveskel_gate PROMOTED (109,984 B, sha cff7a381, 16 of 16, _offc refreshed) and nx_autorig_creature_gate PROMOTED (135,576 B, sha 36b04292, 12 of 12 with the T12 tooth renamed to the wall strip it actually proves; nx_contentdiff live-vs-staged lost_from_live=0, a strict superset gaining the census line and the touching fixture; prev banked, _offc refreshed). The lib change (nx_boneheat_lib: BH_PAD_CELLS 2, the 26-step shell-depth census, the wall strip; nx_curveskel_lib: claim relaxation reverted, walls= in the report) is now proven on nine fixtures across two gates. STILL OWED, blocked only by the D-state storm at the time of writing: rebuild + promote the other lib consumers nx_curveskel, nx_boneheat, nx_autorig_mesh and re-run nx_boneheat_gate, nx_autorig_mesh_gate, nx_autorig_pose_gate on them; build + run nx_rigbench and nx_rigbench_gate on ref9d (M2); reap the two scratch registrations nx_curveskel_gate_staged and nx_autorig_creature_gate_staged now that promote consumed their stage slots; referee grade and operator ACCEPT rows for ga_accept_boneheat and ga_accept_curveskel. 2026-09-06 land EVERY CONSUMER OF THE FIXED LIB IS LIVE (2026-09-06, after the storm): nx_curveskel PROMOTED 61,272 B sha ae812295 (ruler strict superset), nx_boneheat PROMOTED 60,968 B sha 8131d8a9 with nx_boneheat_gate 15 of 15 on the new lib (open prism still unsealed, two shells still 0 of 84 bleed) PROMOTED 89,592 B, nx_autorig_mesh PROMOTED 80,336 B sha 155cb6eb with nx_autorig_mesh_gate 4 of 4 (one-call byte-identical to the two steps) PROMOTED 96,776 B, nx_autorig_pose_gate 10 of 10 with every banked value reproduced (md 259 and 229, crease 162 against 446, 16 joints swung) PROMOTED 144,296 B; every _offc twin refreshed and every scratch registration reaped. The four scratch runs all read d4+ = 0 and walls=0 on their fixtures. 2026-09-06 measure THE M2 REFEREE'S FIRST REAL-CHARACTER NUMBERS (nx_rigbench_gate 11 of 11, staged run on the house rig sites/nishifamily/world/ref9d.nxa: 104 artist joints, 42 artist leaves, 14,164 vertices, artist mean bone 89.99 mm). SKELETON, from the bare mesh with no hint: 111 joints, 7 limbs; 461 permil of the artist joints lie within one artist bone length of one of ours, 126 permil of ours are spurious, joint error mean 208.9 mm and max 457.8 mm. WEIGHTS, artist skeleton held fixed: mean L1 1354 permil of the NXA scale (2000 = disjoint), strongest-joint agreement 241 permil of vertices. DEFORMATION: the library's single pose moves ZERO joints (player joints_posed=0, max_disp 0.03 mm), so md 0.2 mm max says nothing -- the leg ran, the number is degenerate. Rulers proven first: a declared chain reads its mean bone length exactly (4500), identical SKIN rows L1 0, disjoint rows 8192. The voxel census on ref9d at 96 cells read d1=8202 d2=4794 d3=565 d4=21 (21 stripped, a real character's shell is one to two cells thick at an 18 mm cell). HONEST READING: on a real character the topological curve skeleton is FAR from the artist's semantic rig (fingers, spine chain and face joints do not exist in a medial-axis peel), and at 18 mm cells the hands collapse so the weight agreement is low; this is the number a SOTA claim needs and it is not one yet. 2026-09-06 queue WHAT THE REFEREE NAMES NEXT, in the order the numbers point: (1) RESOLUTION LADDER -- rerun the bench at 192 and 384 cells along the longest axis (a conf row, not a constant) and publish agreement versus cell size; the hand and face weights are sub-cell at 96. (2) A REAL POSE FOR LEG C -- when the library pose moves no joints, synthesise the pose gate's subtree swing on an artist limb (compose ag_append_swing into the lib, one writer) so md_measure judges a deformation, and add pose_id to the CLI grammar. (3) SEMANTIC JOINTS (M4) -- compose nx_nxa_retarget's topology-plus-bind-position mapping onto the curve skeleton and re-score coverage per artist joint class, since the 461 permil is dominated by the joints a medial peel cannot invent. (4) THE OTHER DONORS -- toon3d8, witch, paladin through the same bench so the numbers are a population, never one character. Every rung prices against the numbers above, which stay on this board as the baseline. 2026-09-06 queue LOCAL SIGNIFICANCE, ACCEPT RULE DECLARED BEFORE THE RUN (2026-09-06): the peeler's stop was GLOBAL -- a tip counted only if it reached farther than the body's ROOT medial radius -- so a 60 mm finger on a hand attached to a torso 110 mm thick could never be a limb, which is exactly why the referee read 461 permil coverage with fingers, toes and face joints absent by construction. Landed in nx_curveskel_lib: a tip is a limb when it reaches farther than the medial radius WHERE IT ATTACHES (dsurf at the path cell beside the skeleton); an insignificant tip is claimed without a joint; the loop ends when nothing lies farther than one cell from the skeleton, bounded by CS_MAX_PEELS=512 with truncation announced; peels= and insignificant= are printed. Also landed: cs_run_cells (a caller's resolution override above the conf row, argv over conf over default) and, in the referee, distinct strongest-joint counts for the artist skin and ours (the instrument for the helper-joint hypothesis behind 241 permil). ACCEPT RULE: prisms 2, T 3, dragon 6, separated biped 5, shouldered 5, touching 5 -- every fixture count unchanged, because at the root the attachment bar equals the old global bar; on ref9d the artist-joint coverage must RISE from 461 permil and the joint error mean must FALL from 208.9 mm, at 96 cells and again at 192; a change that only moves the number by adding spurious limbs (spurious permil rising faster than coverage) is refused. 2026-09-06 measure LOCAL SIGNIFICANCE, FIRST CUT, AGAINST THE ACCEPT RULE (2026-09-06, staged runs): prisms 2 and 2 (peels 2, insignificant 0), T 3 (peels 3, insignificant 0) with every banked value byte-identical, nx_curveskel_gate 16 of 16; dragon 6 (peels 6, insignificant 0, identical leaves); separated biped 5 (peels 12, insignificant 7 -- seven extra candidates examined and every one correctly rejected), shouldered 5 (peels 9, insignificant 4). The TOUCHING biped FAILED: 8 limbs -- the x=+8 leg peeled FOUR times, tips at (8,3,-51) then (8,-3,-36), (8,-3,-30), (8,-3,-24). The leaf readout names the cause: the first descent hugs y=+3 on a 20 x 10 block whose medial set is a PLANE, the path cell's dsurf under-reads the half-width, the claim tube under-covers, and a one-cell sliver survives along y=-3 that, under a bar taken from ITS OWN attachment cell (dsurf 1 to 2 cells at the leg's edge), passes as three limbs. The old global bar hid the same sliver by accident. Not a patch: ONE primitive was missing, the local medial radius at a cell = the thickest dsurf in its neighbourhood grown until it stops growing (cs_local_radius) -- the axis cell's dsurf IS the cross-section -- now used by BOTH the claim tube and the significance bar. The earlier 'relaxed claim' was this primitive's first half applied against the wrong disease (the wall) and refuted for it; with the wall stripped, the sliver is the disease it was actually for, and the bar half is what makes it complete. 2026-09-06 measure REFEREE ON THE NEW LIBS (local significance, local medial radius, wall strip), staged CLI fc8cb84a run as nx_rigbench_staged on ref9d 2026-09-06. At 96 cells: artist_joints_covered 461 permil (unchanged), joint_err_mean 20895 umm (was 20890), spurious 82 permil, top1 241 permil, L1 1354 permil -- no movement, and at 1.9 cm per cell a finger is sub-cell so resolution binds before the significance bar does. At 192 cells: covered 923 permil, joint_err_mean 4010 umm (max 16868), spurious 63 permil, our_joints 190, limbs 7, top1 233, L1 1331. The 192 gain is UNATTRIBUTED until the promoted binary d4617b00 (old libs) runs at 192 -- that control is running (/tmp/rb192_old.txt); a resolution-only gain would show there too, and only the difference between the two 192 runs belongs to this change. 2026-09-06 measure nx_autorig_creature_gate staged 0702106b (the complete local-radius primitive) reads 11 of 12: dragon 6, biped 5, shouldered 5 unchanged; the ABUTTING-legs biped peels 6 (extra leaf at (8,-3,-36), a 15 mm sliver up one leg on the minus-y side) against the rule's 5, so the pre-declared accept rule FAILS on that fixture while walls_stripped=890 shows the wall strip firing. That fixture is the one this board already names invalid for the peeler (two boxes sharing a face are one solid in an SDF union, lesson row 1788693100): the queued grooved-thigh fixture is the honest touching control, and the abutting one stays on the gate as the wall-strip witness only, never as a limb-count assertion. Nothing in the lib chain promotes until that control exists and the 192-cell attribution is read. 2026-09-06 measure THE REFEREE'S DEFORMATION LEG, MEASURED ON A REAL POSE (2026-09-06, the #1 leverage step: measure the right thing). The earlier runs drove rb_run with pose_id=1, which does not exist on ref9d (its POSE library holds exactly id=0), so nx_nxa_play replayed identity, joints_posed=0, and md_ours_vs_artist read a degenerate 0.2 mm -- the leg was silently vacuous. Re-run with pose_id=0 at 192 cells: pose 0 moves ALL 104 joints by up to 54.9 mm (min 0.46 mm, spread 54.5 mm) on the artist-weighted play and 55.1 mm on OUR-weighted play, both zero_weight_verts=0, both rc=0. md_measure between the two posed skins (same artist skeleton, same pose, artist SKIN vs our re-solved geodesic SKIN): p95 = 6.3 mm, max = 41.3 mm on a 55 mm motion. THAT IS THE HONEST QUALITY NUMBER, and it reframes the gap: top-1 LABEL agreement read 233/1000 because it penalises naming a joint differently even when the deformation is identical; on the metric that matters -- deformation -- our sovereign geodesic weights track the artist rig to 6.3 mm at the 95th percentile, worst 41.3 mm in the sharpest-bend region. NOT a SkinTokens comparison (that is still the oracle-image lane MO10/MO11); it is our-weights-vs-artist on the same skeleton, isolating the weight solve. Two rungs fall out: (a) rb_run must AUTO-SELECT a pose that moves joints (a degenerate arg silently vacated the leg -- a fixture-reached-the-condition failure the referee should refuse, not average to 0.2 mm); (b) the worst 41.3 mm names the region a learned refiner (MD4) must beat the geodesic control on. 2026-09-06 land REFEREE HARDENED AGAINST THE SILENT-VACUOUS DEFORMATION LEG (2026-09-06). nx_rigbench_lib rb_run now AUTO-SELECTS a pose that moves joints: it tries the requested pose_id, then scans 0..poses-1, and takes the first whose ARTIST play reports joints_posed>0 (read from the captured NXA-PLAY line by the new rb_after helper); it prints deform_pose= and deform_pose_joints_moved=, and if NO pose moves it sets md_max/md_p95/md_refused = -1 and ABSTAINS rather than reporting a false near-zero. Two CAS edits on the lib (19,437 -> 20,714 -> 21,827 B), CLI rebuilt 123,056 B aa8c54b8 (debug), staged as nx_rigbench_staged. PROVEN LIVE: run at 192 cells with the deliberately-DEAD pose_id=5 -> the referee auto-selected deform_pose=0 deform_pose_joints_moved=104 and played pose 0 on both weightings (artist 54.9 mm, ours 55.1 mm max displacement), where the OLD code would have replayed bind and read md 0.2 mm. This is the 'a fixture the defect cannot fail is not a test' law applied to our own SOTA instrument -- the deformation number (p95 6.3 mm, max 41.3 mm on a 55 mm pose) is now measured on a pose that provably moves, and a static library can no longer flatter the verdict. OWED (immediate): rebuild nx_rigbench_gate on the fixed lib with a non-vacuity tooth asserting deform_pose>=0 AND deform_pose_joints_moved>0 (and a neg-control: a single-static-pose fixture must ABSTAIN, md=-1), then promote the CLI + gate. SOTA-vs-SkinTokens remains gated on the oracle head-to-head (MO10, a container that runs their published torch weights -- an operator infra decision on a GPU-less NAS) or a sovereign learned weight-refiner (MD4); neither is conjurable in-session, and the honest number to beat is now deformation p95 6.3 mm, not the label metric 233 permil. 2026-09-06 measure SECOND-TOPOLOGY REFEREE (2026-09-06, full-population-not-a-sample): a pose was synthesised onto the dark_witch donor by retargeting ref9d pose 0 through nx_nxa_retarget emit=pose (witch_p.nxa: 111 joints, 50,230 verts, 57 mm motion, limb-length residual 0.79 mm, mapped=31 inherited=80), then the hardened referee ran on it at 128 cells. RESULT, and it CORRECTS the ref9d picture: coverage 396 permil (vs 923 on the house rig), joint_err_mean 190 mm (vs 4 mm), our_limbs 10 (the cape/hair add protrusions), spurious 274 permil, top1 104 permil. The auto-select worked (deform_pose=0, 111 joints moved on both plays: artist 53.9 mm max, ours 46.3 mm max -- our weights under-drive the extremities), but md_refused=1 so the per-vertex deformation delta is UNMEASURED on this donor. HONEST READING: ref9d, a clean biped, was the easy case and overstated us; the classical curve-skeleton is competitive only on clean topology and degrades badly on a complex/clothed character -- exactly the robustness a learned model trained on diverse rigs (SkinTokens/UniRig on ArticulationXL) provides, which STRENGTHENS the case that the gap closes via the oracle head-to-head (MO10/MO11) or a learned refiner (MD4), not a falloff tweak. NEW CALLOUT: md_refused=1 on 50,230 verts where it measured 14,164 is a likely vertex-cap blindness in md_measure -- the deformation leg silently cannot grade the LARGE/complex donors, which are the ones we are worst on (the same measurement-blindness class as the dead-pose leg fixed today). Rung: raise/remove md_measure's vertex cap (compose sys_read_file sizing, never a guess) so the referee can grade any donor, then re-run the 2-donor deformation population. The number to beat remains deformation p95, and it must be measured across TOPOLOGIES before any weight-solver change is adopted -- a change validated on ref9d alone is the vacuous-tuning defect. 2026-09-06 retract CORRECTING MY OWN ROW 1788709700 BEFORE IT PROPAGATES. I called witch md_refused=1 a 'likely vertex-cap blindness' -- WRONG. Read md_grid_build (nx_mmdev_lib): it returns 0 ok / 1 extent-refused / 3 empty, and refuses with code 1 when the mesh bounding-box extent exceeds MD_EXTENT_MAX = 25000 (the overflow-DERIVED bound: the barycentric products stay under the i64 ceiling only below ~2.5 m at 0.1 mm; the header states this and refuses BY NAME above it). So md_refused=1 is a CORRECT overflow-prevention guard, not blindness, and RAISING the cap would re-introduce the very overflow it prevents -- the opposite of the fix I implied. WHAT IS CONFIRMED: the deformation leg cannot grade the witch as-is. WHAT IS NOT: whether the witch is genuinely authored beyond the extent (the modding lane measured donors from 0.21 m to 210 m, so plausible) OR rb_nxmesh feeds md coordinates in the NXA's 0.01 mm unit while md's bound is in 0.1 mm (a 10x unit mismatch would refuse a normal-size character) -- these two have OPPOSITE remedies (scale-normalize the asset vs fix the unit in rb_nxmesh), so I will NOT name a remedy until the experiment runs: print the extent md_grid_build computed for both ref9d (which passed) and witch (which refused), and compare to 25000. The witch SKELETON numbers (coverage 396 permil, joint_err 2.8x mean bone) are ratio-based and stand: the classical fit is genuinely worse on complex topology. Only the md-refusal mechanism was mis-stated. 2026-09-06 measure MD-REFUSAL SETTLED 2026-09-06, the experiment row 1788709950 committed to, done by READING the conversion rather than raising any cap. rb_nxmesh (nx_rigbench_lib) sets each coordinate xyz9 = w / RB_NXA_PER_TENTH with RB_NXA_PER_TENTH=10, i.e. NXA 0.01 mm divided to md's 0.1 mm unit, and md's own NM_UNIT_PER_M=10000 is tenths of a mm, so producer and consumer units MATCH -- md_refused=1 is NOT the 10x unit mismatch I could not rule out. It is genuine authored scale: for the extent guard MD_EXTENT_MAX=25000 (0.1 mm = 2.5 m) to fire, the witch's posed coordinates must span past 2.5 m in their own units, consistent with the modding lane's measured donor range 0.21 m to 210 m. The guard is a correct i64-overflow prevention (barycentric products stay under the ceiling only below ~2.5 m at 0.1 mm) and MUST NOT be raised -- raising it re-introduces the overflow it exists to stop. ref9d passed because a ~1.7 m human is 17000 in 0.1 mm, under the bound. 2026-09-06 queue REMEDY FOR md_refused ON LARGE DONORS, named by the settled mechanism above, NOT a cap raise: a recenter-and-normalize-for-measurement pass in rb_run leg C -- translate both posed skins (artist and ours) to their shared centroid, and if the extent still exceeds MD_EXTENT_MAX scale BOTH by one factor into the bound, run md_measure, then multiply the returned p95/max back by that factor so the number is reported in true mm. This is the field's standard normalize-before-a-distance-metric and leaves md's overflow guard intact, so the referee can grade any donor regardless of authored scale. ACCEPT RULE, pre-declared: ref9d (which passed unscaled) reads the SAME p95 6.3 mm / max 41.3 mm after the pass (a scale factor of 1 is a byte-identical no-op), and the witch's deformation delta becomes a finite mm number instead of md_refused=1. Until it lands the witch deformation leg reads UNMEASURED-BY-SCALE, never a false near-zero -- the same discipline as the auto-select fix that abstains on a dead pose. 2026-09-06 queue BAR DATING OWED ON THIS BOARD under the 2026-09-06 rule that a bar is the current month's or it is refused: the five bar rows above (Blender 5.0, ZBrush 2026, Maya, Substance 3D Painter, the browser-native editors) carry 'read 2026-08-25' only in prose and no month field, so once nx_compare_rank composes nx_barfresh they classify MALFORMED and this domain is REFUSED (exit 5 BAR-STALE) until each source is re-read this month, mirrored and pinned in dcc.refs, and the row rewritten in the dated seven-field shape (bar, id, YYYY-MM, subject, what, ref key, note) with a barscan row per ref. The motion board was converted the same day as the worked example (MOB1 to MOB5). CARRIED HERE FROM motion.plan row 1788710600: SkinTokens (arXiv 2602.04805) is a FEBRUARY 2026 bar. The September 2026 arXiv listing (ref key arxiv_autorig_2026_09 on motion.refs, read from the mirrored bytes) places the field at rig-free feed-forward posing (Make-It-Poseable v3, updated 2026-09-02) and, for skinning weights, a GEODESIC geometric prior augmenting learned generation (A Geodesic Cut-Cell Prior for Neural Skinning, 2608.11272, 2026-08-11) because learned methods lack the generalisability of classic geometric approaches -- which is exactly this rung's bone-heat class plus the M6 learned refinement. The M2-to-M6 ladder stands; its TARGET is now that August-September 2026 pair, never SkinTokens, and the referee that grades the ladder is unchanged. 2026-09-06 land REFEREE HARDENING PROMOTED 2026-09-06: nx_rigbench live 123,056 B sha aa8c54b8 (prev banked), after nx_contentdiff live-vs-staged GREEN lost_from_live=0 ONE-WAY-GAIN (142 runs gained: deform_pose=, deform_pose_joints_moved=, the deformation-leg ABSTAINS message, artist and our top-1 distinct-joint counts, and peels= and insignificant= from the local-significance curveskel lib it was built on). rb_run now auto-selects a pose that moves joints and abstains with md -1 when none does, proven live earlier today with a dead pose_id=5 auto-resolving to pose 0 on ref9d. OWED before this reads complete: nx_offc_install nx_rigbench promoted so runners fork the new twin (issued), a re-run of nx_rigbench_gate on the promoted binary (deferred: the box is at swap 950 permil), the non-vacuity tooth in that gate (deform_pose >= 0 AND deform_pose_joints_moved > 0, with a single-static-pose fixture that must ABSTAIN), and the scratch registration nx_rigbench_staged reaped. Under the current-month rule the referee's TARGET is the August to September 2026 pair recorded in the queue row above, and the deformation p95 6.3 mm on ref9d stays the number a learned refiner must beat on the same referee. 2026-09-06 land DARK CONTRACT MECHANISM landed estate-wide 2026-09-06 (motion.plan row 1788715000 carries the full recipe): a watch row whose organ exists without its contracted symbol now reads DARK on the page, the comparewatch plane and the ranker instead of OPEN forever, which is the permanent form of the amesh_run lesson on this rung. This board's five dated bars moved to the sotabar kind the freshness ruler reads (the bar kind is the renderer's research-bar grammar); sotatarget and rungrole rows are owed. nx_rigbench PROMOTED d441bc19 with the normalize pass; nx_rigbench_gate carries the T12 non-vacuity tooth and is queued as a sovereign build -- the gate re-run on the promoted binary and the witch re-measurement (a finite deformation number under md_scale_div) follow that build. 2026-09-10 measure 2026-09-10 AUTO-RIG INVESTMENT RECOVERED / CLAIMS REQUIRE CURRENT QUALIFICATION: original Claude session789442b2-9b90-4bf3-b4d4-db659ceb3662 contains native results for curve-skeleton16/16, geodesic weights15/15, bend10/10, creature12/12 and rigbench11/11, followed by a later creature candidate11/12 RED. These are substantial existing implementations, not a four-test prototype;4/4 only describes the one-call composition gate. The rigbench pass validates its ruler, not high-quality artist-rig parity. Fresh nx_catalog job1789076127 finds7related owners built/promoted/registered/ever-authorized but invocationNO in a4194304B recent tail; this does not erase historical runs or establish current successful consumer outcomes. Reconcile immutable artifacts, generation/consumer identity, current tests and served Beach use before reuse or rebuilding. Existing dcc/motion/modding ownership retained; no duplicate auto-rig service. Full historical result chronology retained locally and indexed by knowledge/gates/autorig-history-reconciliation-t131.json. User directs records be references to ideas, feedback and investment, never automatic adoption of old decisions or unsupported superiority claims. Exact provider token spend remains unknown.

DC17 Technical-illustration line render of an exploded assembly, print-ready vector · cell WATCHING · gate NO-GATE-DECLARED

Contract	`ti_explode_render` · executor Organ · 3 u · stands on stem:R7 (UNRESOLVED: no such rung on this plan)
Definition of done	INTAKE 1790106111 (operator 2026-09-22): from an exploded layout (stem R7 over procgen ex3_pose) render a hidden-line axonometric technical illustration as vector -- silhouette and crease lines with stroke weight by depth, numbered callouts placed by the ONE chart-layout ruler (datavis nx_chartlay_lib leaders, critic 0 defects), emitted as SVG through nx_render_svg and as spot-colour separations for screen print; done-rule: the banked car-explode fixture renders byte-identical twice, the callout critic grades 0 defects, and the separation count equals the palette rows
Matrix cell	EXPLODE: technical exploded-view illustration to print-ready vector with callouts and separations · `runtime/nx_techillus.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_techillus`; declare `nx_techillus TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-22 queue 2026-09-22 declared from intake 1790106111 by the PM tier: nx_intake_route propose found no rung on any board carrying the request at three words (only stem's watch gap), so the render leg is authored here -- this board owns the authoring tool, procgen owns the explode engine, graphics the scene renderer; the artwork line is the operator's car exploded-view apparel art 2026-09-22 queue INTAKE 1790106111 routed by nx_intake_route (operator, epoch 1790108121) from operator: EXPLODED-VIEW CAR ART FOR APPAREL: a series of well-artisted exploded-view illustrations of favourite cars of all time, print-ready for the back of a t-shirt or sweater (operator 2026-09-22, verbatim: -- sponsor intake-1790106111 plus2

dossiers=17 cells_joined=17 no_matrix_row=0 (sum 17) gates_recorded=0 gates_unrecorded=0 no_gate_declared=17 (sum 17 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.

Rows. 66 feature rows mapped of 74 matrix rows; 8 substrate rows (kernels, bounds and instruments the plan's rungrole marks substrate) are the estate's scaffolding, not category features, and stay off the map

Views. 2D cut: centrality, distinctiveness · 3D cube: centrality, distinctiveness, breadth · axes registered: centrality, distinctiveness, breadth, depth, momentum · a board picks its own in knowledge/compare/dcc.cdmap (cut|x|y, cube|x|y|z); absent = the HBR defaults

bubble area = breadth · ring = momentum (green rising, red falling, grey until day two) · dashed = category means · axes fitted to the field of play: centrality 200–1000, distinctiveness 0–500 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 689 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)

axes fitted: centrality 200–1000 · distinctiveness 0–500 · breadth 0–1000 permil · farther bubbles are painted first, nearer ones over them · the floor shadow is each bubble's (x, y) at height 0
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 5 (a fact of the data: two players that close are that close) · data spread 102 permil of the plot

Nishi Blender 5.0 ZBrush 2026 Maya 2026 ArmorPaint
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 13 (a fact of the data: two players that close are that close) · data spread 540 permil of the plot

Player	Quadrant	centrality	distinctiveness	breadth	depth	momentum	Rows held	Days on spine	First seen
Nishi	Mainstream	822	264	489	702	0 since 2026-09-23	46	10	2026-09-15
Blender 5.0	Aspirational	975	424	803	946	0 since 2026-09-23	56	10	2026-09-15
ZBrush 2026	Peripheral	669	265	444	888	0 since 2026-09-23	33	10	2026-09-15
Maya 2026	Aspirational	940	392	742	942	0 since 2026-09-23	52	10	2026-09-15
ArmorPaint	Peripheral	359	119	181	750	0 since 2026-09-23	16	10	2026-09-15

Day	Matrix rows reviewed	Players recorded
2026-09-15	73	5
2026-09-16	73	5
2026-09-17	73	5
2026-09-18	65	5
2026-09-19	65	5
2026-09-20	65	5
2026-09-21	65	5
2026-09-22	65	5
2026-09-23	66	5
2026-10-08	66	5

How to read this evidence. A Nishi source mark records whether the named symbol is declared in the selected source file. It does not establish working behavior or accepted delivery. A watching tag names an expected source entry; discovering it updates source coverage only. A dark tag means the file exists but the named symbol was not found: inspect the source and contract before assuming the capability is missing or renaming code. Peer codes are authored assessments; unassessed means no comparison is established, not absence. Qualification, deployment identity, runtime observations and acceptance must carry their own evidence. Adoption records the catalog's observed path through source, build, promotion, registration and invocation; libraries use importer evidence and gates use declared execution surfaces. A full catalog path does not establish quality, freshness or a successful user outcome. Partial and unknown states retain their stated next action. Census stamps: importers asof 1790637268, gate census asof 1790104131 (unix seconds; -1 = census absent).

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 296, unassessed 0. Authored assessments are not verified outcomes.

Capability	Nishi	Blender 5.0	ZBrush 2026	Maya 2026	ArmorPaint
vr-compositor-mesh DARK — `runtime/_hdl_build/nx_vrframe.nx` EXISTS but does not declare `vf_build_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. DC15 watch, added 2026-09-06 because the ranker read the rung UNMAPPED (no matrix row carried its symbol) and priced it at zero. vf_build takes only a distortion flag and an output path; its scene is the six-row vf_pill constant table and it rasterises nothing triangular, so VR is the one consumer DC12's ruler cannot yet measure. Contract: vf_build accepts a mesh and a pose, a triangle rasteriser draws it into both eye views, and the geometry the compositor received is recoverable so the same ruler measures it exactly like the other three consumers. Rival cells read 0 as NOT-WITNESSED: no rival's VR viewport was read this month and this board cites nothing it did not read. dossierwatching vf_build_mesh Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
non-destructive-operation-stack Source present; authored stronger claim: `es_replay` in `runtime/nx_editstack_lib.nx`, verified at emit. EXCEED with the why, and it LANDED 2026-08-25. An append-only operation log where undo, redo and re-evaluation are all REPLAYS FROM THE BASE, so the state is reproduced BIT-IDENTICALLY rather than approached by applying an inverse. Every rival is float and applies inverse operations, so their undo drifts and nothing digests the state to notice -- and NEITHER banked rival page contains the token determinis at all [blender50] [zbrush26], so the axis is not merely unwon by them, it is unclaimed. nx_editstack_gate GREEN 44 of 44 on the PROMOTED binary via api gate_run, and BITE-PROVEN by mutation of the library inside its build closure (killed=1 not_reached=0, restore byte-identical). Non-vacuous on three axes: the digest is proven to DISCRIMINATE four distinct states so a constant digest cannot pass, undo is proven to reach an EARLIER state so replay cannot be a no-op, and re-evaluation is proven equal to an INDEPENDENTLY BUILT stack so a corrupted state cannot pass as a changed one. The estate's only prior undo was nx_editor_canvas ec_undo, a ONE-DEEP FULL-ARRAY SNAPSHOT on a 2D page canvas. Adoption: `LIB-WIRED importers=7 nonval=4` — fully adopted (top of its ladder). dossier Pros Source symbol found; stronger capability claim is authored; ahead of Blender 5.0, ZBrush 2026, Maya 2026, ArmorPaint; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	●	●	●	○
parametric-re-evaluation-at-depth Measured: `es_reeval` exists in `runtime/nx_editstack_lib.nx`, verified at emit. LANDED 2026-08-25. Changing a parameter of operation N re-evaluates and PRESERVES operations N+1 onward -- the property that makes an editor non-destructive. Proven not merely to change the state but to equal an INDEPENDENTLY BUILT stack carrying the new parameter, which is the tooth separating a correct re-evaluation from a corrupted one. Blender's modifier stack and geometry nodes [blenderfeat] and Maya's construction history [mayatoc] are both Best here and are the bar; ArmorPaint composes materials with nodes [armorpaint]. ZBrush reads 0 as NOT-WITNESSED: its page carries no non-destructive stack token at all. Adoption: `LIB-WIRED importers=7 nonval=4` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of ZBrush 2026; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads)	●	◉	○	◉	●
sovereign-viewport-zero-gpu-api Source present; authored stronger claim: `mv_render_loaded_cam` in `runtime/nx_meshview_wasm.nx`, verified at emit. EXCEED with the why, and the why is sovereignty rather than speed. Our own software rasterizer compiled to WebAssembly, orbiting a real mesh in the visitor's browser with NO WebGL and NO GPU API in the path at all -- there is no rival to this because every other 3D tool delegates rasterization to a driver. HONEST: it is slower than a GPU path, it is a viewer and not yet an editor, and the estate also carries nx_stl_view_emit which quietly reintroduces a WebGL dependency for the same job. Two viewers with opposite architectures is the duplicate-ruler defect and the sovereign one is the keeper. Adoption: `LIB-UNIMPORTED` — PARTIAL: nothing imports it: wire it into a consumer or retire it. not adopted: nothing imports it: wire it into a consumer or retire it Pros Source symbol found; stronger capability claim is authored; ahead of Blender 5.0, ZBrush 2026, Maya 2026, ArmorPaintCons not adopted yet: nothing imports it: wire it into a consumer or retire it	◉	○	○	○	○
millimetre-deviation-ruler Source present; authored stronger claim: `md_measure` in `runtime/nx_mmdev_lib.nx`, verified at emit. EXCEED with the why: no content-creation tool ships a metrology instrument, because none of them has a use that demands one. Exact point-to-triangle deviation in tenths of a millimetre over EVERY vertex, both directions, mean p95 and max, on a grid derived from the triangle count and never sampled. It is what lets a retopology or sculpt rung state a surface-error BOUND instead of an opinion, and it is already gate-proven to read a known 5.0 mm lift as 5.0 mm and a self-compare as zero. Adoption: `LIB-WIRED importers=9 nonval=5` — fully adopted (top of its ladder). Pros Source symbol found; stronger capability claim is authored; ahead of Blender 5.0, ZBrush 2026, Maya 2026, ArmorPaint; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	○	○	○
mesh-validity-and-manifold-oracle Source present; authored stronger claim: `mv_manifold` in `runtime/nx_meshvalid_lib.nx`, verified at emit. EXCEED with the why, and the why is the REFUSAL rather than the detection. SHIPPED AND PROVEN 2026-08-25. Six named defect classes -- non-manifold edges, boundary holes, degenerate triangles, duplicate vertices, inconsistent winding, index range -- each a COUNT plus the offending element index, with a refusal rather than an overlay. Table capacity is DERIVED from the input at a declared quarter load factor, so there is no fixed size to outgrow and no cap to hit silently, and the zero-area test is an exact integer cross product rather than a float epsilon. nx_meshvalid_gate GREEN 32 of 32 on the PROMOTED binary, and BITE-PROVEN by mutation of the library inside its build closure (valid_mutants=1 killed=1 not_reached=0, restore byte-identical at 13123 bytes, mutant_live=0). THE GATE EARNED ITS GREEN THE HARD WAY: its FIRST run was RED 25 of 31 and it caught a real aliasing bug in this very library -- the header's first-offender slots overlapped the vertex array, so the check's own reset loop silently MOVED vertex zero on entry. Every topology class still passed, because a moved corner is still a valid tetrahedron; the ONLY symptom was that the duplicate-vertex detector could never fire, reported as FAIL VACUOUS did not fire on the bad input. Fixed at the constant, plus a structural invariant tooth asserting the header outranks every first-offender slot so a future slot addition cannot reintroduce the aliasing in silence. The rivals all SHOW mesh problems in a viewport overlay for a person to notice; NONE of them refuses, which is the entire difference and the reason this is flagged an exceed: an asset pipeline with nobody watching needs a non-zero exit, not a highlight. 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=4 nonval=2` — fully adopted (top of its ladder). dossier Pros Source symbol found; stronger capability claim is authored; ahead of Blender 5.0, ZBrush 2026, Maya 2026, ArmorPaint; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	●	●	●	○
sdf-boolean-union-difference-intersection Measured: `nx_csg_combine` exists in `runtime/nx_sdf.nx`, verified at emit. Signed-distance min and max combinators with a smooth-minimum fillet, so a boolean is unconditionally robust where a mesh boolean is famously fragile. NOT marked an exceed: Blender 5.0 renamed its Boolean solver and ships 27 volume-grid nodes including SDF blends [blender50], ZBrush has Live Boolean [zbrush26] and Maya has Boolean operations with a boolean stack [mayatoc]. Ours is integer and theirs is float, and determinism is claimed once above rather than re-claimed on every row that happens to be integer. Adoption: `LIB-WIRED importers=24 nonval=8` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ArmorPaint; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads)	●	◉	●	◉	○
sdf-field-evaluation Measured: `nx_sdf_eval` exists in `runtime/nx_sdf.nx`, verified at emit. Sphere, box, cylinder, torus and rounded box in Q14 millimetres -- the primitive layer the booleans and the extractor both compose. 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=24 nonval=8` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ZBrush 2026, Maya 2026, ArmorPaint; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	●	○	○	○
n-primitive-csg-scene Measured: `nx_csg_scene_add` exists in `runtime/nx_csg_scene.nx`, verified at emit. An op chain with per-primitive rotation and a per-op blend factor, replacing an extractor that was hardwired to exactly two primitives. 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=4` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ArmorPaint; fully adopted on the estate ladderCons behind Blender 5.0 (leads)	●	◉	●	●	○
marching-tetrahedra-extraction Measured: `nx_csg_extract` exists in `runtime/nx_csg.nx`, verified at emit. Kuhn six-tetrahedron split over the field, watertight BY CONSTRUCTION where marching cubes needs a disambiguation table to avoid holes. This is the estate's voxel-remesh path and the grid resolution is its knob; ZBrush's Dynamesh is the workflow bar [zbrush26]. Adoption: `LIB-WIRED importers=14 nonval=7` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ArmorPaint; fully adopted on the estate ladderCons behind ZBrush 2026 (leads)	●	●	◉	●	○
solid-re-evaluated-from-its-op-chain Measured: `nx_csg_scene_eval_i` exists in `runtime/nx_csg_scene.nx`, verified at emit. The solid is not a mesh that was edited, it is a rule evaluated at any point in space -- so it has no maximum resolution and no baked topology to degrade. The rebuild-from-rules property is genuinely real here and is the model the rest of this board is trying to reach. 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=4` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ZBrush 2026; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads)	●	◉	○	◉	●
mesh-transform-translate Measured: `nx_mesh_translate` exists in `runtime/nx_mesh_edit.nx`, verified at emit. Integer translation on NxMesh. Present and correct, and one of only five verbs in an organ whose header claims Blender-class. 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=3` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ArmorPaint; fully adopted on the estate ladderCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads)	●	◉	◉	◉	○
mesh-transform-uniform-scale Measured: `nx_mesh_scale_rat` exists in `runtime/nx_mesh_edit.nx`, verified at emit. Rational numerator over denominator so a scale is exact rather than a float multiply. UNIFORM ONLY -- there is no per-axis scale. 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=3` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ArmorPaint; fully adopted on the estate ladderCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads)	●	◉	◉	◉	○
mesh-subdivide Measured: `nx_mesh_subdivide` exists in `runtime/nx_mesh_edit.nx`, verified at emit. One triangle into four. HONEST: not Catmull-Clark, no creases, and it does not smooth -- it only increases density. 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=3` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ArmorPaint; fully adopted on the estate ladderCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads)	●	◉	◉	◉	○
mesh-bounds Measured: `nx_mesh_aabb` exists in `runtime/nx_mesh_edit.nx`, verified at emit. Axis-aligned bounds as six integers, the primitive every fit and placement operation composes. 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=3` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ArmorPaint; fully adopted on the estate ladderCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads)	●	◉	◉	◉	○
sketch-extrude Measured: `nx_cad_extrude` exists in `runtime/nx_cad.nx`, verified at emit. A convex 2D profile swept to a fan-capped prism. HONEST: this is a CAD sketch extrude, NOT the face-extrude on an existing mesh that a modeller means by the word. 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=5 nonval=2` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ArmorPaint; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads)	●	◉	●	◉	○
decimate-grid-clustering Measured: `nx_mesh_decimate` exists in `runtime/nx_mesh_decimate.nx`, verified at emit. Rossignac-Borrel clustering to a cell-centroid representative -- fast and topology-insensitive. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-GATE-ONLY importers=5` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. not adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Implementing source symbol found; runtime acceptance separate; ahead of ArmorPaintCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads); not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	●	◉	◉	◉	○
decimate-quadric-error Measured: `nx_mesh_qem` exists in `runtime/nx_mesh_qem.nx`, verified at emit. Garland-Heckbert quadric error metrics to a target triangle count, pure integer. HONEST AND NAMED: this estate has TWO decimators with no shared interface and neither names the other. Genuinely different algorithms so the duplication is defensible, but an unnamed duplicate ruler is how a third one gets written. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-GATE-ONLY importers=1` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. not adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Implementing source symbol found; runtime acceptance separate; ahead of ArmorPaintCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads); not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	●	◉	◉	◉	○
vertex-normals Measured: `vn_smooth_area` exists in `runtime/nx_vnormals_lib.nx`, verified at emit. CONSOLIDATED 2026-08-25 -- AND THIS ROW'S OWN EARLIER CLAIM WAS WRONG, which is why the correction is recorded rather than swapped. It said three copies of one computation and called sc_vertex_normals area-weighted. MEASURED: sc_vertex_normals normalises each face normal to unit length BEFORE accumulating, so it is UNIFORM weight, while gl_vnormals and the inline block inside write_obj accumulate the raw cross product whose magnitude IS twice the area. That is two-and-one, not three. The two genuinely identical ones were merged onto vn_smooth_area under proof, with the gate carrying VERBATIM transcriptions of both originals as independent oracles plus an end-to-end tooth reading the bytes write_obj actually writes; nx_vnormals_gate 18/18 GREEN and mutation-proven. sc_vertex_normals was DELIBERATELY NOT MERGED: on a 100 to 1 area ratio the two answers are (0,9,999) and (0,181,181), not even parallel, and the control that makes that mean something is that on an EQUAL-AREA fixture they come back exactly parallel -- so the weighting is the whole cause and merging would have silently changed shipped geometry. That is the 2026-07-21 class of defect, where a dedupe deleted live capabilities, avoided by measurement rather than by luck. 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 ArmorPaint; fully adopted on the estate ladderCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads)	●	◉	◉	◉	○
shell-and-solidify Measured: `sc_close` exists in `runtime/nx_shellclose.nx`, verified at emit. Offset an open sheet along its own normals and stitch the rim into a watertight solid -- the operation that makes a surface printable and simulable. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-GATE-ONLY importers=1` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. not adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Implementing source symbol found; runtime acceptance separate; ahead of ArmorPaintCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads); not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	●	◉	◉	◉	○
sculpt-brush Measured: `nx_mesh_sculpt_brush` exists in `runtime/nx_mesh_sculpt.nx`, verified at emit. A radial displacement brush with a centre, radius and strength in Q14. HONEST: that is the entire sculpting surface -- no symmetry, masking, alpha or stencil, no layers, no falloff choice, no brush library. Blender ships twenty brush types including Crease Clay Strips Pinch Grab Smooth and Mask [blendersculpt], ZBrush handles tens of millions of polygons in real time [zbrush26] and Maya has a Sculpting Tools set with stamps [mayasculpt]. ArmorPaint has a sculpt mode its own vendor labels work-in-progress [armorpaint]. Adoption: `LIB-WIRED importers=7 nonval=2` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads)	●	◉	◉	◉	●
anatomy-derived-uv-unwrap Measured: `ntx_apply` exists in `runtime/nx_nxa_texc_lib.nx`, verified at emit. A bone-anchored cylindrical parameterization over the humanoid NXA layout, additive and idempotent, with region sums checked. NARROW BY CONSTRUCTION: it will not unwrap a general mesh, which is why the general unwrap is a watch row below rather than a claim here. 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 Blender 5.0 (leads), Maya 2026 (leads)	●	◉	●	◉	●
uv-region-grouping Measured: `ntb_regions` exists in `runtime/nx_nxa_texbake_lib.nx`, verified at emit. Four fixed regions derived from bind anatomy. NARROW: anatomy grouping, not a rectangle packer, and it cannot lay out charts from an arbitrary unwrap. 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; ahead of ArmorPaint; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads)	●	◉	●	◉	○
texture-bake-to-atlas Measured: `ntb_bake` exists in `runtime/nx_nxa_texbake_lib.nx`, verified at emit. Deterministic albedo bake to an atlas, proven at 1024 and 2048. 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 Blender 5.0 (leads), Maya 2026 (leads), ArmorPaint (leads)	●	◉	●	◉	◉
pbr-map-set-regenerated-per-genome Measured: `ntb_bake_pbr` exists in `runtime/nx_nxa_texbake_lib.nx`, verified at emit. Albedo, specular, gloss and a relief-derived normal across four regions with a skin index of refraction, regenerated per genome at load rather than shipped as files. NOT marked an exceed: Blender has the Principled BSDF [blender50], Maya has Standard Surface [mayatoc] and ArmorPaint is a PBR authoring tool by self-description [armorpaint], so the honest difference is what DRIVES the regeneration, not that it regenerates. ZBrush reads 0 as NOT-WITNESSED -- the token PBR does not occur on its page at all. Adoption: `LIB-WIRED importers=13 nonval=10` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ZBrush 2026; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads), ArmorPaint (leads)	●	◉	○	◉	◉
procedural-solid-noise Measured: `tm_noise3` exists in `runtime/nx_trimesh.nx`, verified at emit. Three-dimensional noise sampled in object space, the true procedural-texture primitive -- it does not swim when the object deforms the way a 2D projection does. 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=51 nonval=23` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads), ArmorPaint (leads)	●	◉	●	◉	◉
seeded-perlin-fbm Measured: `nx_perlin_fbm_2d` exists in `runtime/nx_perlin.nx`, verified at emit. Seeded integer Perlin with octave fBm, the shared 2D noise foundation. 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=7 nonval=4` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads)	●	◉	●	◉	●
skeleton Measured: `sk_add_bone` exists in `runtime/nx_skeleton.nx`, verified at emit. Parent plus local offset rows composed into an integer world pose. Blender ships envelope, skeleton and automatic skinning [blenderanim] and Maya has Bind Skin [mayatoc]. ZBrush reads 0 as NOT-WITNESSED and here it is probably also true: its own page routes animation to a sibling product. Adoption: `LIB-WIRED importers=55 nonval=48` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ZBrush 2026, ArmorPaint; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads)	●	◉	○	◉	○
linear-blend-skinning Measured: `sk_skin` exists in `runtime/nx_skeleton.nx`, verified at emit. N-influence linear blend skinning in fixed point. 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=55 nonval=48` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ZBrush 2026, ArmorPaint; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads)	●	◉	○	◉	○
weight-import-with-validation Measured: `g2_skin_weights` exists in `runtime/_hdl_build/nx_gltf2mesh.nx`, verified at emit. Joint indices and weights decoded with a sum-to-unit validation and inverse-bind-matrix checks. HONEST: weights can be imported and set by hand, but there is NO automatic binding -- no heat diffusion, no bone glow -- so a new mesh cannot be rigged without external weights. ArmorPaint only CONSUMES a baked skin on import [armorpaint]. Adoption: `RUN-BY:fork:nx_gltf2mesh_units_gate` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ZBrush 2026; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads)	●	◉	○	◉	●
inverse-kinematics Measured: `ik_solve` exists in `runtime/nx_ik.nx`, verified at emit. Cyclic coordinate descent using cross-product signs with no arctangent anywhere, iterations and step set by the caller. 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=2 nonval=1` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ZBrush 2026, ArmorPaint; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads)	●	◉	○	◉	○
keyframe-sampling Measured: `nx_timeline_sample` exists in `runtime/nx_animation_timeline.nx`, verified at emit. HONEST: LINEAR interpolation only. No bezier handles, no tangents, no ease curves, so there is no graph editor and no animator would call this animation authoring. Blender has F-Curves and a Graph Editor [blenderanim]; Maya's guide has a whole container for keyframe animation and the Graph Editor [mayatoc]. Adoption: `LIB-GATE-ONLY importers=2` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. not adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Implementing source symbol found; runtime acceptance separate; ahead of ZBrush 2026, ArmorPaintCons behind Blender 5.0 (leads), Maya 2026 (leads); not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	●	◉	○	◉	○
blendshape-morph-targets Measured: `mo_apply` exists in `runtime/nx_morph.nx`, verified at emit. Additive vertex deltas applied before skinning, with their own serialization format. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-GATE-ONLY importers=1` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. not adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Implementing source symbol found; runtime acceptance separate; ahead of ArmorPaintCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads); not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	●	◉	◉	◉	○
mocap-decode-bvh Measured: `b_parse` exists in `runtime/nx_bvh_lib.nx`, verified at emit. Full BioVision hierarchy and motion, refusing past its declared joint and channel caps rather than truncating. 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=5 nonval=5` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ZBrush 2026, ArmorPaint; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads)	●	◉	○	◉	○
mocap-retarget Measured: `rt_retarget_chain` exists in `runtime/nx_pose_retarget.nx`, verified at emit. Inverts the forward kinematics -- joint world POSITIONS back to our rig's angles -- which is what makes third-party motion drive a generated skeleton. 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=2 nonval=1` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ZBrush 2026, ArmorPaint; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads)	●	◉	○	◉	○
import-gltf-with-full-rig Measured: `g2_convert` exists in `runtime/_hdl_build/nx_gltf2mesh.nx`, verified at emit. The most complete importer in the estate and far more than its name suggests: mesh plus skeleton plus skin weights plus inverse bind matrices with non-identity detection, plus unit autoscale. This anchors the interchange story. 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_gltf2mesh_units_gate` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads)	●	◉	●	◉	●
import-fbx Measured: `fbx_read` exists in `runtime/nx_fbx_import.nx`, verified at emit. Binary FBX vertices and polygon indices to triangles. 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=5` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads)	●	◉	●	◉	●
import-obj Measured: `oi_read` exists in `runtime/nx_obj_import.nx`, verified at emit. Vertices and faces including combined vertex texture normal references. 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=2` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads)	●	◉	◉	◉	●
import-mmd-motion Measured: `v_layout` exists in `runtime/nx_vmd.nx`, verified at emit. Bone, morph, camera and light sections with Shift-JIS names and named refusal codes -- a format the commercial tools reach only through community plugins. 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). not adopted: 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 ZBrush 2026, ArmorPaintCons 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)	●	●	○	●	○
export-glb Measured: `write_glb` exists in `runtime/nx_gltf_export.nx`, verified at emit. JSON and binary chunks with positions and indices, plus a coloured variant. This is the path an asset takes to a storefront viewer or a game engine. ArmorPaint saves obj and glb [armorpaint]. ⛔CORRECTED 2026-08-25 BY A WITNESS OUTSIDE THIS ESTATE, AND THE CORRECTION IS RECORDED RATHER THAN QUIETLY SWAPPED. This row previously implied a working interchange path and the specification body's own reference validator disagrees: every one of the 12 files this emitter has published fails glTF 2.0 validation with ACCESSOR_VECTOR3_NON_UNIT on EVERY normal -- 3198 of 3198 on human.glb, first vector length 999.1766610564921 against a required 1.0 -- because gl_vnormals emits ~1000-scale normals beside a source comment asserting that renderers re-normalise, which the spec contradicts. It is NOT a cosmetic warning: the validator ranks it an Error, and a strict loader is entitled to refuse the asset. Everything ELSE this emitter writes is spec-perfect, proven by rescaling only the normals in place and revalidating at 0 errors and 0 warnings, so the container, JSON, accessors, bufferViews, bounds and indices are all correct and the defect is exactly one axis. The done-rule for closing it is not a size or a shrug: the emitter ships unit-length normals and the SAME third-party validator returns 0 errors over the whole published population, re-run from the recipe banked in knowledge/fetched/cmp_dcc_gltfvalidator_witness_2026-08-25.txt. See the C5 block in this file's header for the controls. Adoption: `LIB-WIRED importers=18 nonval=12` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads)	●	◉	●	◉	●
export-obj Measured: `write_obj` exists in `runtime/nx_obj_export.nx`, verified at emit. Vertices, normals and faces in one buffered write. 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; fully adopted on the estate ladderCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads)	●	◉	◉	◉	●
export-stl Measured: `nx_stl_write_mesh` exists in `runtime/nx_stl_write.nx`, verified at emit. Binary STL, feeding the estate's own slicer to gcode to printer path -- the asset leaving as a physical object rather than a picture. 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=4` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ArmorPaint; fully adopted on the estate ladderCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads)	●	◉	◉	◉	○
volumetric-soft-body-solver Measured: `st_substep` exists in `runtime/nx_softtissue.nx`, verified at emit. An XPBD solver over a TETRAHEDRAL volume cage with per-layer compliance and contact force readout -- the physics tier most content tools ship as a plugin. NOT CLOTH and the row says so: there is no bending constraint. NOT marked an exceed because Blender ships soft body and cloth [blendersim] and Maya ships nCloth [mayatoc]. Adoption: `LIB-WIRED importers=14 nonval=4` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of ZBrush 2026, ArmorPaint; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	●	○	●	○
secondary-dynamics Measured: `sd_step` exists in `runtime/nx_softdyn.nx`, verified at emit. Spring-damper jiggle and recoil with the stability envelope enforced INSIDE the solver, so an unstable configuration cannot be requested. 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 ZBrush 2026, ArmorPaint; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads)	●	◉	○	◉	○
mesh-splice-into-a-named-layer Measured: `mm_merge` exists in `runtime/_hdl_build/nx_meshmerge.nx`, verified at emit. Splice one mesh into a named layer of another with a placement transform -- the assembly primitive under outfits and attachments. 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). not adopted: 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 ArmorPaintCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (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)	●	◉	◉	◉	○
mesh-rotate-and-mirror DARK — `runtime/nx_mesh_edit.nx` EXISTS but does not declare `nx_mesh_rotate`: 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. ABSENT-PROVEN at corpus_complete=1. There is no mesh rotate and no mirror anywhere in the estate. Mirror in particular is the single most-used modelling operation in existence and its absence is why nothing symmetric can be authored here yet. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them watching nx_mesh_rotate Pros none measured yetCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads); open contract; expected source declaration not established	○	◉	◉	◉	○
mesh-bevel DARK — `runtime/nx_mesh_edit.nx` EXISTS but does not declare `nx_mesh_bevel`: 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. ABSENT-PROVEN. The word bevel occurs six times in the tree and every hit is a different sense -- glyph stroke bevels, voxel edge shading, a flagstone texture -- and one source says in its own comment that it has no bevel or mitre limit. There is no mesh bevel operator. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them watching nx_mesh_bevel Pros none measured yetCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads); open contract; expected source declaration not established	○	◉	◉	◉	○
mesh-smooth-and-relax DARK — `runtime/nx_mesh_edit.nx` EXISTS but does not declare `nx_mesh_smooth`: 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. ABSENT-PROVEN. Every laplacian hit in the tree is the fluid wave equation. Without relax there is no way to repair the topology a sculpt brush distorts. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them watching nx_mesh_smooth Pros none measured yetCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads); open contract; expected source declaration not established	○	◉	◉	◉	○
weld-in-the-mesh-kernel DARK — `runtime/nx_mesh_edit.nx` EXISTS but does not declare `nx_mesh_weld`: 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 as cr_weld but it lives inside a CAD-twin report organ, not in the mesh kernel, so no modelling path can reach it. This is an adoption gap, not a missing algorithm, and it is the cheapest row on this board. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them watching nx_mesh_weld Pros none measured yetCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads); open contract; expected source declaration not established	○	◉	◉	◉	○
quad-retopology DARK — `runtime/nx_mesh_qem.nx` EXISTS but does not declare `rt_quadremesh`: 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. ABSENT-PROVEN. Both decimators produce TRIANGLES; there is no quad-dominant remesh, so a sculpt cannot become a deformable game-budget mesh. Maya ships Remesh and polyRemesh [mayatoc]. The done-rule binds the result to a millimetre bound measured by the EXISTING md_measure rather than a new metric. dossierwatching rt_quadremesh Pros none measured yetCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads); open contract; expected source declaration not established	○	◉	◉	◉	○
dynamic-topology DARK — `runtime/nx_mesh_sculpt.nx` EXISTS but does not declare `sc_dyntopo`: 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. ABSENT-PROVEN. Sculpting cannot add detail where the brush needs it, so the brush is bounded by whatever tesselation it started with. Blender calls it dyntopo [blendersculpt] and ZBrush calls it Dynamesh [zbrush26]. watching sc_dyntopo Pros none measured yetCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads); open contract; expected source declaration not established	○	◉	◉	◉	○
multiresolution-levels DARK — `runtime/nx_mesh_sculpt.nx` EXISTS but does not declare `ms_level`: 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. ABSENT-PROVEN for us -- every multires hit in the tree is a neural hash encoding, a different subject entirely. THE ZBRUSH AND MAYA ZEROES HERE ARE THE SCHEMA'S LIMITATION AND ARE ALMOST CERTAINLY WRONG AS ABSENCE CLAIMS: both products are widely understood to have subdivision levels, but three separate probes of each banked mirror returned nothing, so 0 records NOT-WITNESSED. Blender states multi-res sculpting support outright [blendersculpt]. dossierwatching ms_level Pros none measured yetCons behind Blender 5.0 (leads); open contract; expected source declaration not established	○	◉	○	○	○
sculpt-symmetry-and-masking DARK — `runtime/nx_mesh_sculpt.nx` EXISTS but does not declare `sc_symmetry`: 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 brush signature takes a centre, a radius and a strength and nothing else. Absence here rests on that read signature plus a narrow probe, NOT on a bare word sweep -- symmetr matches 616 times across asymmetric-PAKE crypto and symmetric matrices, and publishing that as coverage would have been a prefix read as a population. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them watching sc_symmetry Pros none measured yetCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026; open contract; expected source declaration not established	○	◉	◉	●	○
layer-aware-sculpt-brush DARK — `runtime/nx_mesh_sculpt.nx` EXISTS but does not declare `sc_layer_brush`: 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 WOULD BE A REAL EXCEED, and it is deliberately NOT flagged as one until it lands, because claiming an exceed on an absent capability is the self-flattery this board's guard exists to stop. A brush that edits the MUSCLE stratum and lets the skin re-emerge over it, or adds fat without moving the muscle beneath. Reachable only because our bodies are generated inside-out with real layers; ZBrush layers are shape blends over a single surface and structurally cannot express it. dossierwatching sc_layer_brush Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
general-uv-unwrap Measured: `uv_conformal` exists in `runtime/nx_uvunwrap_lib.nx`, verified at emit. LANDED 2026-08-25: nx_uvunwrap_gate GREEN 78 of 78 on the promoted binary and mutation-proven (killed=1 not_reached=0, restore byte-identical at 71921 B, mutant_live=0). A flat quad now reads ANGLE distortion 0 where it read 987 permil, a cube's twelve seams give exactly six charts with zero overlaps, a cut cylinder wall unwraps clean, and a CONE POINT honestly reports 174 permil against a plane at 0 -- non-zero, not a refusal, which is the correct behaviour for a surface that genuinely cannot be flattened. ⛔THIS ROW PREVIOUSLY PUBLISHED A WRONG ROOT CAUSE AND THE CORRECTION IS RECORDED RATHER THAN QUIETLY REPLACED, because a false mechanism left standing costs the next reader the whole investigation. The published cause was a discarded return code in uvi_unfold's walk. That is a REAL latent defect and it was fixed too, but it caused NONE of the fifteen failures. The actual cause was found by refuting the briefed one with arithmetic from the gate's own output: an unseamed 4-vertex 2-triangle mesh reported texc corners lost = 1 when a bijection makes 0 the only possible answer, so the corner-to-class map was already corrupt BEFORE any unfold ran. In uv_chart_build's relabel block, uvf is BOTH the union-find parent forest AND the corner-to-class map, and the loop converts one into the other IN PLACE: from the second iteration uvi_find chases a class id as if it were a parent index, and because the encoding is NEGATIVE it indexes before the array and path-compression-WRITES into the CSTART and CORDER slots underneath. Roots are now resolved into their own array before any parent is rewritten. The lesson is the one this estate keeps paying for: A COMPOUND SYMPTOM THAT WILL NOT NAME ITS FAILING CONJUNCT INVITES THE STORY YOU ALREADY BELIEVE, and the arithmetic in the instrument's own output is what refuted mine. The anatomy-locked ntx_apply remains for the humanoid NXA layout; this is the general unwrap beside it, so an arbitrary sculpt or boolean result can now be charted, solved and packed. Blender ships Conformal and Angle Based unwrapping with edge seams and vertex pinning [blendermodel]; Maya has Unfold [mayatoc]; ZBrush names UV Master but the token unwrap never appears on its page [zbrush26]. Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Blender 5.0 (leads), Maya 2026 (leads)	●	◉	●	◉	●
uv-seams DARK — `runtime/nx_nxa_texc_lib.nx` EXISTS but does not declare `uv_seam`: 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. ABSENT-PROVEN. There is no way to tell an unwrapper where to cut. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them watching uv_seam Pros none measured yetCons behind Blender 5.0 (leads), Maya 2026 (leads); open contract; expected source declaration not established	○	◉	○	◉	○
uv-packing DARK — `runtime/nx_nxa_texbake_lib.nx` EXISTS but does not declare `uv_pack`: 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. ABSENT-PROVEN for us. Maya ships Layout UV shells with options [mayatoc]. The Blender 0 is NOT-WITNESSED, not absence -- a packing token simply did not appear in the banked features page. watching uv_pack Pros none measured yetCons behind Maya 2026 (leads); open contract; expected source declaration not established	○	○	○	◉	○
uv-distortion-metric DARK — `runtime/nx_nxa_texc_lib.nx` EXISTS but does not declare `uv_distort`: 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. ABSENT-PROVEN, and it is the referee the unwrap needs: without it a fold is invisible and an unwrap can fail silently. The rivals are coded 1 because they SHOW distortion to an artist; none of them REFUSES on it. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them watching uv_distort Pros none measured yetCons behind Blender 5.0, Maya 2026; open contract; expected source declaration not established	○	●	○	●	○
texture-painting Measured: `tp_stroke` exists in `runtime/nx_texpaint_lib.nx`, verified at emit. LANDED 2026-08-25, closing what this board called its single largest gap. A surface point plus barycentrics becomes an atlas texel, a brush lays a declared smoothstep falloff across the map set, and EVERY STROKE IS AN OPERATION ON THE EDIT STACK, so a paint session replays byte-identically and undoes stroke by stroke. nx_texpaint_gate GREEN 109 of 109 on the promoted binary with 16 gv_bite negative controls. The load-bearing teeth: the hand-computed texel is asserted by ARITHMETIC DONE IN THE FIXTURE, not by the code under test; the stroke paints exactly the hand-counted 25-texel disc with ZERO texels changed outside the radius, which is the anti-vacuity tooth that kills a painter that writes the whole atlas; the falloff is exactly zero AT the rim; and the seam bleed is proven by neg-control-seam-TEARS-without-the-bleed, which FIRES on a build with the bleed removed and is silent with it -- so the anti-tear claim is bitten rather than asserted. Replay is bit-exact both ways: undoing every stroke restores the base byte for byte and redoing returns to the exact same bytes. The bleed margin is DERIVED from the brush radius and the chart width, never picked. All four rivals still lead on brush breadth and this is not an exceed; what is ours is that a stroke is a replayable op and the seam is a property of the metric rather than something a person notices and paints over. [blendermodel] [zbrush26] [mayatoc] [armorpaint] Adoption: `LIB-GATE-ONLY importers=1` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. dossiernot adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Implementing source symbol found; runtime acceptance separateCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads), ArmorPaint (leads); not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	●	◉	◉	◉	◉
cloth-and-garment-drape DARK — `runtime/nx_softtissue.nx` EXISTS but does not declare `gd_drape`: 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. ABSENT-PROVEN as cloth: there is no bending constraint anywhere. The solver that exists is a volumetric tetrahedral cage. The garment organ in the estate is a 2D print-on-fabric compositing preview with zero simulation and zero 3D, and calling it cloth here would have been a false capability. Blender simulates cloth for clothing and environments [blendersim] and Maya ships nCloth [mayatoc]. dossierwatching gd_drape Pros none measured yetCons behind Blender 5.0 (leads), Maya 2026 (leads); open contract; expected source declaration not established	○	◉	○	◉	○
viewport-selection-and-gizmo Measured: `vp_pick_screen` exists in `runtime/nx_vpick_lib.nx`, verified at emit. LANDED 2026-08-25, and the done-rule held: THE GIZMO NEVER TOUCHES GEOMETRY. A screen point becomes a ray in the SAME camera convention nx_meshview_wasm already orbits in, Moller-Trumbore picks the nearest triangle in exact integer arithmetic, and a drag resolves to ONE operation pushed onto the edit stack -- so every drag is undoable, redoable and replays bit-identically. nx_vpick_gate GREEN 132 of 132 on the promoted binary with 13 gv_bite negative controls (this row said 11 before and that was a miscount -- 13 is what the run actually prints). The teeth that make it non-vacuous: the nearer triangle wins AND STILL WINS WHEN THE BUFFER ORDER IS REVERSED, so the answer is depth and not array order; a degenerate triangle is never returned as a hit and is a DISTINCT answer from parallel and from behind; a miss returns a NAMED negative code that no valid index can collide with; and three separate teeth assert the run never wrote a single word of the mesh buffer. The gizmo magnitude is checked against an INDEPENDENTLY DERIVED closed form (gap measured at 0 against a derived slack of 12), not against itself. AND THE INTERACTION HALF IS NOW A MEASURED NUMBER RATHER THAN A VIEWPORT CAPABILITY WITH NO COST BESIDE IT: the same gate times vp_pick_screen over procedurally generated grids at FOUR stated triangle counts and publishes nanoseconds per pick with its repetition count and clock source printed so the arithmetic can be redone -- 32 tris 9659 ns, 128 tris 30332 ns, 512 tris 116495 ns, 2048 tris 455054 ns, each the MINIMUM of two samples of 1024 repetitions on CLOCK_MONOTONIC at 1 us resolution, divided once at the end. The bar is not invented here: knowledge/gamefeel_oracle.conf already banks frame_budget_ms_90hz at 11..12 ms, the derived 1e6/90 = 11111 us agrees with that row, and js_block_ms caps app logic at 2 ms -- the same two cited rows nx_tissuebudget measures the tissue tick against, reused rather than a second bar invented beside it. So a pick over 2048 triangles is 40955 ppm, about 4.1 percent, of one 90 Hz frame, and the cited 2 ms app-logic ceiling buys 9001 triangles per pick. Cost is measurably SUB-linear (per-triangle cost 736 permil across the 64x range, where 1000 would be exactly linear and 64000 quadratic) because the fixed ray construction amortises as the mesh grows, and the bar separating linear from quadratic is DERIVED as the geometric mean of the two hypotheses rather than tuned until the result passed. THE FIGURE IS A NATIVE FLOOR AND SAYS SO: the shipped viewport is wasm in a visitor browser, which is slower on hardware that is very likely weaker. Blender, ZBrush and Maya all lead here and this is parity, not an exceed -- what is ours is that the drag is an op on a log rather than a write, and that the pick cost is published with its triangle count instead of asserted. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-GATE-ONLY importers=1` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. dossiernot adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Implementing source symbol found; runtime acceptance separateCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads); not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	●	◉	◉	◉	●
node-graph-authoring DARK — `runtime/nx_editstack_lib.nx` EXISTS but does not declare `ng_eval`: 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 graph whose evaluation IS the operation stack rather than a second engine, so a graph edit and a manual edit cannot diverge. The tooth is that a graph and the equivalent hand-built log produce byte-identical state -- the proof there is only one evaluator. ArmorPaint composes materials with nodes and its development is public though its binaries are paid [armortools]. dossierwatching ng_eval Pros none measured yetCons behind Blender 5.0 (leads), Maya 2026, ArmorPaint (leads); open contract; expected source declaration not established	○	◉	○	●	◉
collaborative-multi-seat-editing DARK — `runtime/nx_editstack_lib.nx` EXISTS but does not declare `co_merge`: 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 column is zero and that is the honest reading against THESE four rivals -- but it would be dishonest to leave it there: the browser-native editors named in dcc.plan collaborate today and we do not. The append-only log makes convergence reachable; the tooth is that two interleaved op streams replayed in either order end at identical digests. dossierwatching co_merge Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
asset-ingest-under-provenance-and-licence Measured: `ain_admit` exists in `runtime/nx_asset_admit_lib.nx`, verified at emit. Every ingested asset should carry where it came from, what licence admits it and a content pin, failing CLOSED at intake and naming the missing conjunct. The rivals import anything and record nothing, so a 1 here is generous to them. REPOINTED 2026-10-01 from the glTF importer to the organ that carries the contract: ain_admit is the one intake decision (a named origin, a licence the one rights table knows, a sha256 pin equal to the bytes presented, a declared likeness state with a consent record where a person is pictured), proven by nx_asset_admit_gate GREEN 60 of 60 on the promoted binaries. The decision is shipped and awaiting acceptance; the intake doors (the glTF, FBX and bundle importers) do not call it yet, and that wiring is the open half of the rung. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-UNCENSUSED` — PARTIAL: a library absent from the import-graph census (younger than the last eco-graph build, or unreachable from buildroot/runtime). dossiernot adopted: a library absent from the import-graph census (younger than the last eco-graph build, or unreachable from buildroot/runtime) Pros Implementing source symbol found; runtime acceptance separateCons not adopted yet: a library absent from the import-graph census (younger than the last eco-graph build, or unreachable from buildroot/runtime)	●	●	●	●	●
one-asset-four-consumers-one-ruler Source present; authored stronger claim: `xp_consumers` in `runtime/nx_xport_lib.nx`, verified at emit. EXCEED with the why, LANDED 2026-08-25: no rival has this because no rival owns four consumers. ONE authored source is emitted to every consumer and the geometry each RECEIVES is recovered OUT OF THE ARTIFACT BYTES and measured against the source by md_measure -- the ruler that already existed, composed rather than duplicated. nx_xport_gate GREEN 38 of 38, mutation-proven. Deviations in tenths of a millimetre: game NXA 0/0/0 and clinical NXMSH2 0/0/0, both EXACTLY lossless against a derived bound of ZERO -- by arithmetic, not tolerance, because their unit is an integer multiple of the ruler's so the round trip is a multiplication and an exact division. Storefront glb and obj read 0/4/4 against a derived bound of 11. VR is UNREACHED and ABSTAINS BY NAME, never scored zero: vf_build takes no mesh argument at all, its scene is a six-row constant table and it emits a PNG, and BOTH facts are re-checked against live source on every run so the abstention cannot decay into a pass -- that is now rung DC15. THE FINDING WORTH MORE THAN THE ROW: the storefront is our only lossy consumer and the loss is STRUCTURAL, not a bug -- write_glb and write_obj hold coordinates at fx1024, a 0.977 mm quantum, COARSER than the 1 to 2 mm bar nx_mmdev_lib's own header says the clinical unit exists to serve. An asset that is clinically exact going in is not clinically exact when it reaches a storefront viewer, and nothing before this row could see it. A second tooth earned its place the same way: orientation checked by per-triangle normal dot product CANNOT catch a reflection, and the signed-volume replacement carries the proof -- a mirrored box reads max deviation 0 while vol6 flips sign, so deviation alone would have passed a mirrored mesh silently. Adoption: `LIB-GATE-ONLY importers=1` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. dossiernot adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Source symbol found; stronger capability claim is authored; ahead of Blender 5.0, ZBrush 2026, Maya 2026, ArmorPaintCons not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	◉	○	○	○	○
unify-the-two-mesh-containers DARK — `runtime/nx_morph.nx` EXISTS but does not declare `mo_apply_nxmesh`: 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 REAL DEFECT FOUND BY THE CENSUS, published rather than filed quietly: the estate carries TWO mesh containers, and the morph, shell and merge organs sit on a different one from the sculpt and CSG organs -- so a blendshape CANNOT currently be applied to a sculpted or boolean-extracted mesh without a conversion nobody has written. Every rival has exactly one mesh type. rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them watching mo_apply_nxmesh Pros none measured yetCons behind Blender 5.0 (leads), ZBrush 2026 (leads), Maya 2026 (leads), ArmorPaint; open contract; expected source declaration not established	○	◉	◉	◉	●
Auto-rig skin weights, geodesic bone-heat over the mesh volume Open — no implementing organ is measured for this axis yet. SHIPPED 2026-09-05, gate nx_boneheat_gate 15 of 15 GREEN. Blender Automatic Weights IS the bone heat algorithm on vertex-to-bone DISTANCE and needs a manifold, and its own manual warns of overlaps that need hand weight-painting [blenderboneheat]. Ours voxelises the mesh VOLUME, floods the exterior from the padded border and solves geodesic distance on the 26-connected interior with integer Dial buckets, so weight CANNOT cross air, a hand resting on a hip takes no hip weight, measured 0 of 84 cross-air leaks against 16 for the nearest-joint control on the same fixture. It blends at every joint, sums to the NXA 4096 exactly with the residual folded into the largest influence, is byte-deterministic, and weights an UNSEALED mesh over its shell rather than refusing it. c2 c3 c4 read 0 as NOT-WITNESSED, ZBrush and ArmorPaint do no skeletal skinning and no Maya geodesic-bind page is mirrored here. Pros none measured yetCons behind Blender 5.0; no implementing organ measured	○	●	○	○	○
Auto-rig done-rule end to end: bare mesh to skeleton to weights to an extreme pose played by the served player and judged by md_measure Source present; authored stronger claim: `ag_append_swing` in `runtime/nx_autorig_pose_gate.nx`, verified at emit. SHIPPED 2026-09-05, gate nx_autorig_pose_gate 10 of 10 GREEN on the promoted binary: nx_curveskel then nx_boneheat on a bare 60 mm prism, the whole shoulder subtree (16 of 32 joints) swung 90 degrees as a baked world-delta POSE, played by the SERVED nx_nxa_play, judged by md_measure at 25.9 mm max and 22.9 mm at p95 against a 10 mm bar, with the geodesic crease (162) strictly softer than the nearest-joint boundary (446) and 245 of 246 verts moving under geodesic weights versus exactly 121 under nearest. Blender reaches a posed bind only from a HAND-PLACED meta-rig [blenderrigify] plus distance bone heat [blenderboneheat]; this row starts from the bare mesh with no human. NOT CLAIMED: the twist half (candy-wrapper is a deformer property, dual-quaternion rung deferred) and any result on real characters (the M2 donor benchmark is the owed referee). Referee grade and operator ACCEPT owed; the acceptance cells beside bh_run and cs_run stay WATCHING until then. c2 c3 c4 read 0 as NOT-WITNESSED. Adoption: `GATE:LIVE trial=-` — fully adopted (top of its ladder). Pros Source symbol found; stronger capability claim is authored; ahead of Blender 5.0, ZBrush 2026, Maya 2026, ArmorPaint; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	●	○	○	○
Auto-rig arbitrary topology in ONE CALL: bare mesh in, skeleton and skinned rig out Source present; authored stronger claim: `amesh_run` in `runtime/nx_autorig_mesh.nx`, verified at emit. SHIPPED 2026-09-05, PROMOTED and REGISTERED, gate nx_autorig_mesh_gate 4 of 4 GREEN: nx_autorig_mesh composes nx_curveskel (cs_run) then nx_boneheat (bh_run, geodesic) in a single invocation and emits the skeleton and the skinned rig as two explicit artifacts. Its neutrality is PROVEN, not asserted: the one-call skeleton and skinned rig are byte-identical to the two steps run separately, every vertex's SKIN sums to the NXA 4096, and a volumeless sheet is refused by name at the skeleton stage. This is the DC16 rung's entry point -- the rung's earlier symbol ar_autorig had never been built, so the ranker read three shipped organs as priority 0 until this row. Blender reaches a rigged, weighted mesh only through a hand-placed meta-rig [blenderrigify] plus distance bone heat [blenderboneheat]; UniRig [unirig] reaches it through a learned model. NOT CLAIMED: real-character quality (the M2 donor benchmark is owed), twist (dual-quaternion rung deferred), referee grade and operator ACCEPT. c2 c3 c4 read 0 as NOT-WITNESSED. 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). dossiernot adopted: 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 Blender 5.0, ZBrush 2026, Maya 2026, ArmorPaintCons 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)	◉	●	○	○	○
Auto-rig skeleton fit, geodesic curve skeleton from a bare mesh Open — no implementing organ is measured for this axis yet. SHIPPED 2026-09-05, gate nx_curveskel_gate 16 of 16 GREEN (re-proven 2026-09-06 on the wall-stripping voxeliser, promoted sha cff7a381), the organ PROMOTED and REGISTERED. Given a bare mesh with NO armature it voxelises the volume (the same voxeliser as the skin solve, so the two halves agree about the body), builds the medial-radius field, roots at the thickest cell nearest the centroid and peels limbs by geodesic farthest-point descent along the medial axis, each bone claiming its cross-section; joints are spaced by local thickness and written as a SKEL that nx_boneheat then skins, a full sovereign auto-rig of an unrigged mesh with no learned model and no training set. Rigify, by its own manual, automates rig CONTROLS from blocks a person places and does not attach to a mesh at all [blenderrigify]; the learned predictors UniRig, RigNet and MagicArticulate are priced on modding MD4. Measured on a 60 mm prism: 2 limbs, 1 fork at the centre, joints spanning 157 to 5890 of 6000, byte-identical reruns, a volumeless sheet refused by name. c1 to c4 read 0 as NOT-WITNESSED for skeleton PREDICTION from a bare mesh. The branch-topology proof landed the same day: a watertight T from the UNION of two box SDFs through nx_csg (120,924 vertices) peels EXACTLY three limbs with two fork joints, the joints reaching both crossbar tips and the stem foot, gate 16 of 16. Pros none measured yetCons no implementing organ measured	○	○	○	○	○
Auto-rig skin weights ACCEPTED (operator, referee, seat) DARK — `runtime/nx_accept_gate.nx` EXISTS but does not declare `ga_accept_boneheat`: 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. ACCEPTANCE WATCH for the row above. The mechanical cell flips on its gate; this cell flips ONLY through the acceptance gate reading an operator ACCEPT row in knowledge/store/accept- plus the referee grade plus the seat attestation. Shipped and gate-proven is presence, not completion, and this row is what keeps the board from saying otherwise. watching ga_accept_boneheat Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Auto-rig skeleton fit ACCEPTED (operator, referee, seat) DARK — `runtime/nx_accept_gate.nx` EXISTS but does not declare `ga_accept_curveskel`: 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. ACCEPTANCE WATCH for the skeleton-fit row. Flips only through the acceptance gate: operator ACCEPT row, referee grade on a real rigged donor posed through nx_nxa_play, seat attestation. The linear and T proofs are mechanical presence; a body the operator looks at is the acceptance. watching ga_accept_curveskel Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Auto-rig REFEREE on real characters: skeleton, weights and deformation scored against an ARTIST rig on the same mesh Measured: `rb_run` exists in `runtime/nx_rigbench_lib.nx`, verified at emit. THE M2 RUNG'S INSTRUMENT, RUN 2026-09-06 on the house rig ref9d (gate 11 of 11): skeleton coverage 461 permil of 104 artist joints within one artist bone length, joint error mean 208.9 mm; weight agreement on the artist skeleton 241 permil of 14,164 vertices (L1 1354 permil of scale); the deformation leg degenerate because the library's single pose moves no joints. These are the honest baseline numbers, not a SOTA claim -- the cell reads landed because the symbol exists, and that flip is a receipt, not a proof. rb_run takes an artist-rigged NXA and (A) sends its bare mesh through nx_curveskel and scores every artist joint's distance to our nearest joint against the artist rig's OWN mean bone length (covered permil, spurious permil, error mean and max), (B) holds the artist skeleton fixed, re-weights with nx_boneheat and reads per-vertex L1 against the artist SKIN as a permil of the NXA scale plus top-1 joint agreement, and (C) plays the artist's own POSE through the SERVED nx_nxa_play with both weightings and lets md_measure judge the two posed skins -- max and p95 in 0.1 mm. Its gate proves the rulers on known answers (a declared chain's mean bone length exact; identical SKIN rows L1 0; rows sharing no joint exactly 2 x 4096, the scale's neg-control) and then runs the bench on the house rig ref9d with every value emitted and NO bar on the deviation until the referee panel grades it. UniRig [unirig], RigNet [rignet] and MagicArticulate [magicarticulate] are scored on real characters; this is the estate's own such ruler, and the number it prints is what a SOTA claim on DC16 needs. c1 to c4 read 0 as NOT-WITNESSED: no DCC ships a rig-versus-artist referee; the practice lives in the papers. Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of Blender 5.0, ZBrush 2026, Maya 2026, ArmorPaint; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
EXPLODE
technical exploded-view illustration to print-ready vector with callouts and separations Open — watching `runtime/nx_techillus.nx` : `ti_explode_render`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. watch contract for dcc DC17 (intake 1790106111): Blender ships a line-art modifier (Yes), Maya a toon-line shader (Part), ZBrush and ArmorPaint none; codes assessed from public feature lists 2026-09-22, not measured; the sovereign twin composes nx_render_svg and the chartlay leader ruler rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them dossierwatching ti_explode_render Pros none measured yetCons behind Blender 5.0, Maya 2026; open contract; expected source declaration not established	○	●	○	◐	○

Rival-claim provenance. Every Yes, Best or Part mark in a rival column is a claim about someone else's product. A cited row carries a reference mark that resolves to a pinned mirror (the refs gate measures that); an uncited row is an observation read and is badged in its evidence until a reference lands. The badge is the receipt, never the proof: a mark proves a mirror exists, not that the mirror supports the code. The rows are counted below.

Rival claims, counted
Class	Count	Note
rows with rival marks	64	rows carrying a Yes, Best or Part in a rival column
cited	22	a reference mark resolves to a pinned mirror
uncited	42	an observation read, badged until a reference lands
rows, summed	64	RECONCILED: the parts sum to the total 64

Delivery and evidence

Inspect risks, technical debt, rendered observations, experiments and references. Read scope and limitations alongside every result.

Release history and work log

The worked plan, on the board. Every leg appends what it measured, landed, retracted, learned and left queued, per rung, so a crash leaves the next seat a ledger here rather than a transcript to mine. Kinds: measure, land, retract, lesson, queue. Newest last.

When	Rung	Kind	Entry
2026-09-06	DC13	lesson	THE BAR MOVED WHILE THE BOARD SLEPT (2026-09-06, the current-month rule applied to this board): the ranker REFUSED dcc (BARFRESH bars=5 malformed=5) because its five bars carried no month; this month's reads moved three of them -- Blender 5.0 to 5.2 LTS with node-based physics (DC13's node graph is now the incumbent's shipping shape, so DC13 must price against a physics-capable graph, not a modifier stack), Womp to an SDF and CSG parametric modeler in the browser (a rival standing on OUR representation axis, so the generated-asset exceed on DC1 and DC3 must be measured against it, not asserted), Maya to 2026.3 with OpenPBR default, ZBrush to 2026.2.1. Two vendor domains (autodesk.com, adobe.com) fail the sovereign TLS chain, filed as a recurrence on the sovereign-tls board; their bars are witnessed by encyclopedia mirrors and say so on the row. Every bar row is now bar, id, YYYY-MM, subject, what, ref, note in that order; the ruler reads FRESH only if it is.
2026-09-03	DC16	measure	Mesh2Motion verified 2026-09-03 via the GitHub API, not recalled: mesh2motion-app 3,212 stars, TypeScript, pushed 2026-08-31 (live), and its SIBLING mesh2motion-assets is CC0-1.0 -- public-domain animations and models, which classify `shipped` under the mediaingest R21 provenance conf and are therefore usable as a retarget corpus with zero provenance risk. CAUTION RECORDED: the marketing site states MIT but GitHub's classifier reads NOASSERTION on the app repo, so read the LICENSE file before relying on it -- the same discipline that caught EpicGames/MetaHuman-DNA-Calibration being proprietary rather than open.
2026-09-03	DC16	lesson	MetaHuman is SOTA for film-quality face rigs and is NOT a digital twin -- it carries no muscle physiology. So the exceed axis against Epic is anatomical and biomechanical grounding, where the referees are OpenSim (opensim-org/opensim-core, Apache-2.0) and MyoSuite (MyoHub/myo_sim, Apache-2.0, musculoskeletal models in MuJoCo) and Epic has nothing to show. That axis is anatomy AN9-AN12's lane; DC16 is the rung that lets a generated body ENTER it.

Risk register

Risk	Likelihood x impact	Mitigation
Publishing an authoring claim with no artifact a person can open	certain x high	An authoring board is uniquely easy to fake because a batch verb LOOKS like a feature in a matrix cell. Every row on this board that claims an authoring capability must name a symbol in an organ, and the header states plainly that there is no editor yet. DC4 is the rung that makes the claim visitable.
Building a second mesh ruler beside the ones that already exist	likely x high	DC6 and DC9 both measure surface change and the estate ALREADY has md_measure, which is gate-proven to read a known 5.0 mm lift as 5.0 mm. Both rungs name it explicitly in their done-rules. A new deviation metric on this board is a defect, not a feature.
A browser-native claim that ignores the browser-native incumbents	certain x medium	Womp, Spline and Vectary are on the bar list precisely so no row can read zero-install as a win against Blender while quietly losing to a product that got there first and also collaborates. DC14 is the honest response.
A viewport claim with no measured interaction cost	certain x medium	DC4's done-rule requires milliseconds per operation at a stated triangle count before the rung may close, the same discipline softbody's SB1 applied to frame cost.
Ingesting content whose licence or consent does not admit it	possible x high	DC11 fails closed and is a prerequisite of DC12, so nothing reaches a storefront or a clinical consumer through an unadmitted asset. Route any likeness through the consent path procgen already declares.
The clinical consumer is a regulated use	likely x high	Software informing clinical decisions is a regulated device under FDA SaMD and EU MDR. DC12 names the clinical twin as a CONSUMER of geometry, not as a diagnostic claim; route through the legal lane before any pilot.
The box is saturated and geometry work is compute-heavy	certain x medium	The governed build lane already refused a build during this session and that is it working correctly. Ask the spend gate before sweeps, never route around the admission governor, and keep retopology and unwrap sweeps off the request path.

Release history continued

When	Rung	Kind	Entry
2026-09-05	DC16	land	SHIPPED nx_boneheat and nx_boneheat_gate 2026-09-05, both PROMOTED and REGISTERED on the NAS, gate 15 of 15 GREEN. The skin-weight half of DC16 is met, geodesic bone-heat auto-weights over the voxelised mesh volume exceeding Blender Automatic Weights [blenderboneheat]; the skeleton-fit half stays nx_skelgen and nx_autorig. Matrix row bh_run landed on dcc.matrix.
2026-09-05	DC16	measure	On the gate fixtures the geodesic solve leaks 0 of 84 weights across 12 mm of air where the nearest-joint Euclidean control leaks 16; it blends at every joint (4 of 4 middle-ring corners carry two joints), falloff is monotone both directions, NXA weights sum to 4096 exactly, two solves are byte-identical, and a mesh with no SKEL is refused by name.
2026-09-05	DC16	lesson	A tilted-face octagon test tube would not voxel-seal (interior 0) because flat faces point-sample into a 26-connected shell a 6-connected flood slips through; an axis-aligned prism seals by construction and the lib keeps a one-pass 6-dilation for real tilted meshes. Fixture geometry is part of the test, not scaffolding.
2026-09-05	DC16	queue	SKELETON-FIT SOURCE LANDED 2026-09-05, the other half of DC16: nx_curveskel_lib (geodesic farthest-point curve-skeleton peeling that COMPOSES nx_boneheat_lib's voxeliser and Dijkstra, no learned model, no training set), nx_curveskel CLI, nx_curveskel_gate. The gate ran 8 of 11 under the storm and the passes are the load-bearing ones: the skeleton SPANS the extremities (joint z from 283 to 5890 of 6000), the tree is single-rooted with no dangling parent, two runs are byte-identical, a volumeless sheet is refused by name, and the medial radius scales 30 to 100 with thickness. Two fixes then landed in source: a bone now CLAIMS its cross-section (a tube of its medial radius) so a bar no longer over-peels into 32 false limbs, and the root is the thickest cell nearest the interior centroid so a bar roots at its centre and peels both ends. Build admission QUEUEd the rebuild on run-queue saturation. RECIPE for the next seat: nx_sov_build_run nx_curveskel_gate (expect 11 of 11), nx_organ_ship nx_curveskel_gate, nx_organ_ship nx_curveskel, then a matrix row on this board bound to cs_run beside bh_run, then the OWED tooth: a T-branch fixture built watertight via nx_csg proving 3 limbs and a fork joint.
2026-09-06	DC16	measure	SKELETON-FIT PROVEN 2026-09-05: nx_curveskel_gate 11 of 11 GREEN on the fixed source (sovereign build-and-run), then PROMOTED and REGISTERED through the ship loop. Values on a 60 mm square prism: limbs=2 branches=1 joints=32, joint z spans 157 to 5890 of 6000, root radius 30 (3 mm wide) versus 100 (7 mm wide), two runs byte-identical, a volumeless sheet refused exit 3. The three fixes that took it from 8 of 11: a bone claims its cross-section tube, the root is the thickest cell nearest the centroid, and descent follows the medial axis. Build admission refused six paced re-issues (run queue up to 13 of 8, D-state 7 of 8) before granting; the sovereign lane, not the mgmt lane, produced the proof.
2026-09-06	DC16	land	SKELETON-FIT SHIPPED 2026-09-05: nx_curveskel PROMOTED and REGISTERED (57,144 B) after its gate, proving against the live nx_curveskel_gate; matrix row cs_run landed beside bh_run with the Rigify manual pinned as the reference that does not attach to a mesh [blenderrigify]. DC16 is now met in BOTH halves for an unrigged mesh, a skeleton from the volume then geodesic weights, with the T-branch fixture the remaining owed tooth, and referee grade plus operator ACCEPT still owed before either half reads complete.
2026-09-06	DC16	land	T-BRANCH TOOTH CLOSED 2026-09-05: nx_curveskel_gate 16 of 16 GREEN. The watertight T (stem UNION crossbar via nx_csg marching tetrahedra: 120,924 verts, 40,308 tris, 23,892 interior cells at a 0.38 mm cell) peels EXACTLY 3 limbs with 2 fork joints and 29 joints, reaching x -1405 and +1407 (the 15 mm crossbar tips) and z 95 (the stem foot); the linear prism still peels 2 limbs with 1 fork. The sovereign build twin-refreshed the served gate and the ship loop was re-run to bank the harness manifest. DC16's skeleton-fit half is now proven on both a linear and a branched volume; referee grade and operator ACCEPT remain owed.
2026-09-06	DC16	queue	WHAT DC16'S DONE-RULE STILL LACKS, named so the next seat builds it and not something else: the rule says the BOUND mesh must deform at an EXTREME POSE without candy-wrapper collapse, measured through md_measure. Skeleton (cs_run) and weights (bh_run) are proven on fixtures; the pose leg is not. RECIPE: an end-to-end gate that takes a bare fixture, runs nx_curveskel then nx_boneheat, poses the result through nx_nxa_play at an extreme joint angle, and measures the deformed skin against the rest pose with md_measure -- a candy-wrapper collapse shows as a cross-section area loss at the joint; assert the loss stays under a bar derived from the medial radius, and carry a neg-control that nearest-joint weights DO collapse. Then acceptance rows ga_accept_boneheat and ga_accept_curveskel sit beside both mechanical rows and flip only through the accept gate.
2026-09-06	DC16	lesson	CANDY-WRAPPER IS A DEFORMER PROPERTY, NOT A WEIGHTS PROPERTY, and the done-rule as written cannot be met by weights alone. Measured 2026-09-05: nx_skeleton (the evaluator nx_nxa_play composes) has no dual-quaternion path -- it is linear blend skinning only. Under a 90 degree BEND, smooth geodesic weights preserve the joint better than a hard boundary and that case is provable through the pose leg. Under an extreme TWIST about the bone axis, linear blending of two opposed rotations collapses the blended ring toward the axis -- and smoother weights make the collapse WORSE, not better; the field's answer since Kavan 2007 is dual-quaternion skinning, a change to the DEFORMER. So the pose tooth proves the bend, and the twist half of this done-rule is a dual-quaternion rung owed on the board that owns nx_skeleton (graphics owns the rig per anatomy.matrix); pricing it here would be the C7 double-pricing defect. Acceptance for DC16 therefore reads: bend proven, twist deferred to the deformer rung, named not hidden.
2026-09-06	DC16	land	THE BEND DONE-RULE IS PROVEN END TO END 2026-09-05: nx_autorig_pose_gate PROMOTED (140,976 B, sha f77a1361) and REGISTERED, 10 of 10 GREEN through nx_job_run on the served binary. A bare 60 mm prism goes through nx_curveskel then nx_boneheat (geodesic and the nearest-joint control), its whole shoulder subtree (16 of 32 joints) is swung 90 degrees as a baked world-delta POSE, the SERVED nx_nxa_play plays it, and md_measure -- the one mesh ruler -- judges the deformed skin against the rest pose. Gates row landed on dcc.gates. Referee grade and operator ACCEPT remain owed: this is presence, not completion.
2026-09-06	DC16	measure	Three independent witnesses agree on the swing. The player's own receipt: joints_posed 16, max displacement 3194 umm (31.9 mm per axis), verts moved 245 of 246 under geodesic weights versus exactly 121 under nearest (the smooth falloff touches the whole body, the hard boundary moves one half). md_measure posed-vs-rest: max 259, p95 229 in 0.1 mm units (25.9 mm and 22.9 mm) against a 10 mm bar, identical for both weightings; geodesic-vs-nearest max 2. The crease read across adjacent 1 mm rings: geodesic 162 umm strictly softer than nearest 446. Two plays byte-identical; the unweighted asset refused exit 5.
2026-09-06	DC16	lesson	A ONE-JOINT POSE MOVES A SLEEVE, NEVER A LIMB, BECAUSE nx_nxa_play WALKS NO HIERARCHY. The player applies M x = D(x - bind_t) + bind_t + dt per joint (nxa_format_spec.md, exact LBS with no runtime forward kinematics), so an entry on one joint deforms only that joint's skinned band: the first run read md max 3 against a 10 mm bar while the ring crease read 5 mm. A real bend is one entry per joint of the swung subtree, D equal to the rotation and dt equal to (R - I)(bind_j - P), which is the rigid swing R(v - P) + P about the pivot P -- baked exactly as an animation clip stores it. The gate composes nf_qrotv so the swing it encodes and the swing the player renders are one arithmetic.
2026-09-06	DC16	lesson	A BAR IN THE WRONG UNIT FAILS A CORRECT RESULT. The second run swung the limb 26 mm and still read RED because the bar was written as 1000 in 0.01 mm quanta while md_measure reports in 0.1 mm for a mesh fed in millimetres: the twist sleeve had read 3 and the swing read 259, so the unit was decidable from the two numbers already in hand. The bar is 100 and its provenance is now in the gate header; the values are emitted with their unit in the key name so no reader inherits the assumption.
2026-09-06	DC16	queue	WHERE THIS RUNG STANDS AGAINST THE FIELD, AND THE LADDER TO SOTA, so the next seat builds the next rung and not a side artifact. STANDING: the skin-weight half is in the Maya geodesic-voxel-bind class and past Blender bone heat on the air-bleed axis [blenderboneheat]; the skeleton half is a first rung, a topological curve skeleton with no joint semantics, proven on a prism and a CSG T only; the deformer is linear blend, no dual quaternion. RUNG M2, THE REFEREE-GRADE BENCHMARK: re-rig the estate's own rigged donors (toon3d8, witch, paladin from the modding lane) with nx_curveskel and nx_boneheat and score joints against the artist SKEL (position error after nx_nxa_retarget mapping) and weights against the artist SKIN under the donor's own POSE library through md_measure -- the number a SOTA claim needs, since UniRig [unirig], RigNet [rignet] and MagicArticulate [magicarticulate] are all scored on real characters. RUNG M3: dual-quaternion skinning in the deformer (graphics owns nx_skeleton) so the twist half of this done-rule can be proven. RUNG M4: semantic joint labels by composing nx_nxa_retarget's topology-plus-bind-position mapping onto the curve skeleton, then IK on the labelled rig (Rigify parity for posing) [blenderrigify]. RUNG M6: the learned refinement modding MD4 names, which must beat the geodesic control on the M2 benchmark. ALSO OWED HERE: nx_gate_bite of nx_autorig_pose_gate against nx_boneheat_lib with subject nx_boneheat (mutant builds were refused by the admission storm today), the referee grade, and the operator ACCEPT rows ga_accept_boneheat and ga_accept_curveskel.
2026-09-06	DC16	land	ONE-CALL AUTO-RIG SHIPPED 2026-09-05: nx_autorig_mesh <bare.nxa> <skel_out.nxa> <skinned_out.nxa> [cells] composes nx_curveskel then nx_boneheat (geodesic) in a single invocation, PROMOTED (77,240 B, sha 16632a8e), REGISTERED, _offc twin installed. Its gate nx_autorig_mesh_gate is 4 of 4 GREEN (PROMOTED, REGISTERED): the one-call skeleton and skinned rig are BYTE-IDENTICAL to the two steps run separately, every vertex's weights sum to 4096, and a volumeless sheet is refused by name at the skeleton stage. Matrix row amesh_run and a dcc.gates row landed.
2026-09-06	DC16	lesson	THE RANKER READ THREE SHIPPED ORGANS AS PRIORITY ZERO. nx_compare_rank dcc reported DC16 UNMAPPED / NO-MATRIX-ROW because the rung's contract symbol ar_autorig had never been built -- the matrix rows carried bh_run, cs_run and the gate's own symbol, none of which the rung named. A rung whose symbol nothing implements is invisible to the scoreboard no matter how much ships beside it. Fixed at the root: the rung symbol is now the real entry point amesh_run, carried by the nx_autorig_mesh matrix row; the re-run ranker reads DC16 done (rungs done 7 to 8) with one honest residual, REGISTERED-UNAUTHORISED, closed by minting a cap. Encode value where the ruler looks: a rung symbol must name a symbol that exists.
2026-09-06	DC16	queue	ARCHETYPE-FREE PROOF IN FLIGHT (operator 2026-09-05: quickly auto rig a dragon vs a human vs whatever, everything-is-math): nx_autorig_creature_gate builds a dragon-like quadruped (body + four legs + tail + neck) and a biped (torso + two legs + two arms + head) as watertight nx_csg_scene unions, runs the SAME one-call auto-rig on both with no archetype hint, and asserts each peels the limb count it was built with (six vs five, derived from construction), every limb tip is reached by a joint, both skins are sound, and the two counts DIFFER (the anti-vacuity tooth). Source landed (16,263 B); its build was refused twice by admission (load 2273 against the 1643 ceiling) and is queued through the mgmt lane. THE RUNG AFTER IT: a generated creature already carries a part graph, so the SDF-part skeleton (nx_autorig ar_build, Prim MST by surface gap) and the volume curve-skeleton must agree on limb topology -- two independent method classes agreeing is the PROVEN bar, and it binds the generate-up stack (procgen/anatomy) to this rung.
2026-09-06	DC16	measure	ARCHETYPE GATE FIRST RUNS 2026-09-06 (nx_autorig_creature_gate, sovereign build-and-run): DRAGON EXACT -- a body with four legs, a tail and a neck peeled exactly 6 limbs (54 joints, 5 forks) from the volume with no hint, every leg quadrant, the tail (-64 mm) and the neck (+30 mm) reached, skin sound. BIPED 7 LIMBS, NOT 5: two legs, two arms and the head are all reached (T8 PASS), plus two extra limbs. A first reading blamed the two 3 mm shoulder ledges outboard of a 12 mm head; the discriminating experiment REFUTED it -- a flush-head biped (head as wide as the torso, no ledges) also peels 7 (54 joints), so the shouldered and flush variants are equal and the extras are elsewhere. The story was plausible and wrong; the instrument that settles it is a per-leaf position readout (every joint no other joint parents = where each limb ends), now in the gate and queued behind a load-21 storm. Not journaled as a defect until the leaves say where the two limbs point: a thin 14 mm torso with a 6 mm root radius makes its own corners and ledges protrusions by the volume's arithmetic, and that may be a correct rig of the fixture, not a wrong one.
2026-09-06	DC16	measure	THE LEAF READOUT SETTLED IT, AND IT WAS NEITHER SHOULDERS NOR CORNERS. Dragon leaves (mm): tail (-61,-2,6), neck (35,-2,28), four feet at z=-33 -- exactly its six. Biped leaves: right foot (8,3,-51), both arm tips (+-32,-2,15), head (-6,5,29), left foot (-8,-3,-51), and two MORE inside the left leg on the same corner column at z=-35 and z=-25. The right leg peeled once, the left leg three times, identical with or without shoulder ledges. The only asymmetry between two mirrored legs is the plane x=0 they share: two adjacent boxes in an SDF union put a zero-field plane exactly on a grid plane, the marcher emits a one-sided internal wall there, and one limb's first peel under-claims its cross-section so the peeler keeps finding far cells down its own corner. The dragon's legs are 6 mm apart and never touch. Two stories were refuted by instruments before the third was read off one (clavicles, then corners); the fixture is corrected to separated legs as the discriminating experiment, and the peeler's robustness to TOUCHING limbs is named as its own rung rather than hidden in a fixture tweak.
2026-09-06	DC16	queue	PEELER ROBUSTNESS TO TOUCHING LIMBS (the rung the leaf readout named): two limbs whose boxes share a face produce, in an SDF union marched on a grid that contains that plane, a one-sided internal wall; one limb's first peel then under-claims its cross-section and the same limb is peeled again down its own corner column (measured: left leg three times, right leg once). Real characters have touching limbs everywhere -- thighs, fingers, an arm against a torso -- so this is not a fixture curiosity. Two remedies to measure against the touching-legs fixture, which stays in the gate as the neg-control that must go from 7 to 5: (a) in cs_claim_tube, claim the cube from the LARGEST medial radius within a one-cell neighbourhood of each path cell rather than the path cell's own value, so a path that hugs a wall still claims the limb it runs through; (b) in bh_raster, a zero-thickness double-sided wall between two interior regions is a shell of measure zero and should not survive the dilation-and-flood as a partition. Accept rule, declared before the experiment: the touching-legs biped peels exactly 5, the separated-legs biped still peels exactly 5, the dragon still peels exactly 6, the T still peels exactly 3, and nx_curveskel_gate stays 16 of 16; a remedy that passes only by loosening the stop bar is refused.
2026-09-06	DC16	land	ARCHETYPE-FREE AUTO-RIG PROVEN 2026-09-06: nx_autorig_creature_gate 11 of 11 GREEN (sovereign build-and-run, 112,870 B). The SAME one-call nx_autorig_mesh with NO archetype hint rigs a dragon-like quadruped and a biped from their volumes alone: dragon exactly 6 limbs (leaves at the tail tip -61 mm, the neck +28 mm, four feet at -33 mm), biped exactly 5 (two feet at -51 mm, two arm tips at +-32 mm, head +29 mm), every limb tip reached, every SKIN summing to 4096, the two counts differ, and the shouldered biped peels the same 5 (3 mm ledges sit under the 6 mm stop bar) so the monotone tooth holds. This is the operator's everything-is-math requirement met on fixtures: dragon vs human from one arithmetic. The discriminating experiment that made it honest: separating the biped's legs by 4 mm took it from 7 limbs to 5, confirming the shared plane between touching boxes as the whole cause of the earlier miscount. Gates row landed on dcc.gates. NOT CLAIMED: real characters (M2 donor benchmark owed), touching limbs (rung above, accept rule pre-declared), referee grade and operator ACCEPT.
2026-09-06	DC16	measure	REMEDY (a) REFUTED 2026-09-06: the relaxed claim (a bone claims by the thickest medial radius in its first cube) rebuilt the creature gate to 11 of 12 -- dragon 6, separated biped 5, shouldered 5, nx_curveskel_gate byte-IDENTICAL 16 of 16 (nothing regressed) -- and the touching-legs biped STILL peeled 7 with the x=-8 leg claimed three times (tips at -51, -35, -25 mm). The instrument that named the real cause was a vertex count: the touching mesh carries 148,080 vertices against 138,624 for the separated one, 9,456 more from legs 4 mm CLOSER -- a zero-thickness double wall the marcher emits where two boxes abut at SDF 0. The claim relaxation is reverted (a remedy that fixes nothing is a magic rule) and the wall is attacked where it lives, in bh_flood.
2026-09-06	DC16	measure	REMEDY (b) FIRST CUT, WALL STRIP AT A DERIVED DEPTH OF 3: bh_flood now walks six-steps from the exterior through surface cells and re-classes any surface cell deeper than the shell's derived depth (raster 1 + dilation 2) as INTERIOR. Measured on the creature gate: dragon still exactly 6 limbs, but its interior grew 30,282 to 40,171 cells and its root radius 110 to 120; the separated biped's root radius went 60 to 112 -- LARGER than its own 7 mm torso half-thickness, which no honest medial radius can be -- and it lost the head (5 to 4 limbs, T7/T8 RED); the touching biped read 3 (both legs ONE block once the wall is gone, which is what two abutting boxes ARE in a union). The strip removed about one real shell layer (the dragon's gain is about one 13k-cell layer), so the derived 3 is a hypothesis the data refutes: the shell is deeper than raster-plus-dilation on bevelled marching-tetrahedra faces.
2026-09-06	DC16	lesson	THREE LAWS FROM ONE FIXTURE. (1) A derived constant is a hypothesis until a census confirms it: bh_flood now PRINTS a shell-depth census (surface cells by six-steps from the exterior) on every flood, so the strip depth is read off the data, never re-derived on paper. (2) Two boxes sharing a whole face are ONE solid in an SDF union; a neg-control for TOUCHING limbs must carry the groove real thighs have, or it tests the marcher's wall, not the peeler. (3) The earlier 11 of 11 counted the flush biped's head as a limb only because internal walls halved the torso's medial radius to 6 mm; with the walls gone the head protrudes about 8 mm past an 11 mm half-thickness and the derived stop rule declines it -- so that GREEN was a measurement of the defect, not of the rig, and the fixture head must protrude past the torso half-thickness to be a limb by the rule's own words.
2026-09-06	DC16	queue	IN FLIGHT: creature gate rebuilt with the census flood (paced re-issues under admission); read the d1..d9 histogram on the dragon (no abutting boxes: the shell's true depth is where the counts collapse) and the touching biped (a long tail is the wall), set BH_WALL_DEPTH from that measurement with the derivation written beside it, then re-run: accept rule stands (dragon 6, T 3, nx_curveskel_gate 16 of 16, never by loosening the stop bar) and the biped fixtures are re-derived from the honest volume -- a head taller than the torso half-thickness, and a touching-legs control with a groove. ALSO BUILT THIS SESSION, awaiting the box: nx_rigbench (M2 referee, lib + CLI, 99,693 B compile) and nx_rigbench_gate (116,598 B) -- skeleton coverage in the artist's own bone units, weight L1 on the artist skeleton, and the artist POSE played by the served nx_nxa_play with both weightings judged by md_measure, on the house rig sites/nishifamily/world/ref9d.nxa.
2026-09-06	DC16	measure	THE CONTROL INDICTED THE INSTRUMENT, NOT THE FIXTURES (2026-09-06). A 26-connected walk still read 7,635 dragon cells deeper than 3, so the census was run on nx_curveskel_gate's lone 60 mm prism -- one box, no internal geometry, the cleanest control in the estate: d1=1352 d2=2136 d3=2424 d4=2280 d5=2136 d6+=0, 4,416 false walls, and the EXTERIOR region only 472 of 14,112 cells. A census that RISES with depth on a convex box means the walk was not starting from the outside: BH_PAD_CELLS was 1 while bh_dilate grows the shell outward by 1, so wherever a face touches the bounding box the pad is surface, the exterior flood has no seed there, and depth was being measured around the shell from a few corner pockets. The same starvation un-refused the sheet neg-control (T9/T10 RED: 24,300 cells re-classed on a flat quad). The T (overlapping stem and crossbar, no abutment) still peeled exactly 3, the prisms 2 and 2 -- the skeletoniser was never the problem. FIX, derived not tuned: BH_PAD_CELLS = 1 + BH_DILATE_STEPS = 2, which shifts the grid origin by a whole cell and so changes no cell boundary relative to the mesh (results should reproduce byte-for-byte with the strip finding nothing on wall-free fixtures).
2026-09-06	DC16	lesson	A DEPTH MEASURED FROM A REGION THAT DOES NOT EXIST IS A DISTANCE TO NOWHERE, AND IT LOOKS EXACTLY LIKE THICKNESS. The exterior flood only needs ONE seed to seal a volume, so a starved exterior passed every sealing gate for weeks; it failed the first instrument that asked how FAR the outside is. Print the size of the region you flood from beside anything you derive from it -- exterior=472 on a 14,112-cell grid was the whole diagnosis, and it was one number the flood already had.
2026-09-06	DC16	measure	PAD FIX CONFIRMED BY THE CONTROL (2026-09-06, nx_curveskel_gate rebuilt 109,984 B sha cff7a381, staged run): with BH_PAD_CELLS=2 every wall-free fixture reads d4+=0 and walls=0 -- thin prism d1=4488 d2=3560 d3=2280, fat prism 10442/9306/7714, sheet 19196/8836/0 (refused by name again), T 11634/10113/7557 -- and every banked value REPRODUCES EXACTLY: thin root radius 30 and fat 100, joint z 157 to 5890, T interior 23,892, three limbs, two forks, 29 joints, tips at -1405 and +1407, foot at 95. 16 of 16 GREEN. So the derived shell depth of 3 (raster 1 + dilation 2) is now a MEASURED fact on four wall-free volumes, the strip is a no-op wherever there is no wall, and a grid-origin shift by a whole cell changes nothing about a result -- the accept rule's T-3 and 16-of-16 conjuncts hold under the remedy.
2026-09-06	DC16	retract	RETRACTED, same day: the lesson row above that read the flush biped's five limbs as 'a measurement of the defect' (the head only counting because internal walls halved the torso radius) was WRONG. With the pad fixed the census reads d4+=0 on both bipeds and the ORIGINAL values reproduce exactly -- separated biped interior 15,994, root radius 60, five limbs with the head at (-6,5,29), shouldered five -- so the head was always a limb by the derived stop rule and the four-limb reading came from the starved-exterior strip inflating the medial radius to 112. The discriminating experiment was the lone-prism control, not any argument. Same family as every retraction this estate records: a plausible story published before its control.
2026-09-06	DC16	land	TOUCHING-LIMBS RUNG MET UNDER ITS PRE-DECLARED ACCEPT RULE (2026-09-06, nx_autorig_creature_gate 12 of 12 GREEN, staged build 135,392 B sha af34fa47, run through a scratch registration of the stage): the biped whose two legs SHARE the plane x=0 now peels exactly five (feet at (8,3,-51) and (-8,-3,-51), both arm tips, head), its wall reads d4=242 d5=230 d6=218 d7=200 in the 26-step census and is stripped (walls=890, root radius 60); separated biped 5, shouldered 5, dragon 6, the T 3, nx_curveskel_gate 16 of 16, and the stop bar untouched. Remedy = bh_flood strips surface cells deeper than the derived shell depth once the exterior has a real seed (BH_PAD_CELLS 1 to 2). HONEST RESIDUAL: the two abutting legs still peel as two tubes because the claim radius follows the block's thinner axis (10 mm of a 20 x 10 block), not because the peeler sees a crease -- a grooved-thigh fixture is the true control for REAL touching limbs and is owed; the walls= count and the census line are now printed on every flood so the next reader sees whether a mesh carried walls at all.
2026-09-06	DC16	land	SHIPPED 2026-09-06: nx_curveskel_gate PROMOTED (109,984 B, sha cff7a381, 16 of 16, _offc refreshed) and nx_autorig_creature_gate PROMOTED (135,576 B, sha 36b04292, 12 of 12 with the T12 tooth renamed to the wall strip it actually proves; nx_contentdiff live-vs-staged lost_from_live=0, a strict superset gaining the census line and the touching fixture; prev banked, _offc refreshed). The lib change (nx_boneheat_lib: BH_PAD_CELLS 2, the 26-step shell-depth census, the wall strip; nx_curveskel_lib: claim relaxation reverted, walls= in the report) is now proven on nine fixtures across two gates. STILL OWED, blocked only by the D-state storm at the time of writing: rebuild + promote the other lib consumers nx_curveskel, nx_boneheat, nx_autorig_mesh and re-run nx_boneheat_gate, nx_autorig_mesh_gate, nx_autorig_pose_gate on them; build + run nx_rigbench and nx_rigbench_gate on ref9d (M2); reap the two scratch registrations nx_curveskel_gate_staged and nx_autorig_creature_gate_staged now that promote consumed their stage slots; referee grade and operator ACCEPT rows for ga_accept_boneheat and ga_accept_curveskel.
2026-09-06	DC16	land	EVERY CONSUMER OF THE FIXED LIB IS LIVE (2026-09-06, after the storm): nx_curveskel PROMOTED 61,272 B sha ae812295 (ruler strict superset), nx_boneheat PROMOTED 60,968 B sha 8131d8a9 with nx_boneheat_gate 15 of 15 on the new lib (open prism still unsealed, two shells still 0 of 84 bleed) PROMOTED 89,592 B, nx_autorig_mesh PROMOTED 80,336 B sha 155cb6eb with nx_autorig_mesh_gate 4 of 4 (one-call byte-identical to the two steps) PROMOTED 96,776 B, nx_autorig_pose_gate 10 of 10 with every banked value reproduced (md 259 and 229, crease 162 against 446, 16 joints swung) PROMOTED 144,296 B; every _offc twin refreshed and every scratch registration reaped. The four scratch runs all read d4+ = 0 and walls=0 on their fixtures.
2026-09-06	DC16	measure	THE M2 REFEREE'S FIRST REAL-CHARACTER NUMBERS (nx_rigbench_gate 11 of 11, staged run on the house rig sites/nishifamily/world/ref9d.nxa: 104 artist joints, 42 artist leaves, 14,164 vertices, artist mean bone 89.99 mm). SKELETON, from the bare mesh with no hint: 111 joints, 7 limbs; 461 permil of the artist joints lie within one artist bone length of one of ours, 126 permil of ours are spurious, joint error mean 208.9 mm and max 457.8 mm. WEIGHTS, artist skeleton held fixed: mean L1 1354 permil of the NXA scale (2000 = disjoint), strongest-joint agreement 241 permil of vertices. DEFORMATION: the library's single pose moves ZERO joints (player joints_posed=0, max_disp 0.03 mm), so md 0.2 mm max says nothing -- the leg ran, the number is degenerate. Rulers proven first: a declared chain reads its mean bone length exactly (4500), identical SKIN rows L1 0, disjoint rows 8192. The voxel census on ref9d at 96 cells read d1=8202 d2=4794 d3=565 d4=21 (21 stripped, a real character's shell is one to two cells thick at an 18 mm cell). HONEST READING: on a real character the topological curve skeleton is FAR from the artist's semantic rig (fingers, spine chain and face joints do not exist in a medial-axis peel), and at 18 mm cells the hands collapse so the weight agreement is low; this is the number a SOTA claim needs and it is not one yet.
2026-09-06	DC16	queue	WHAT THE REFEREE NAMES NEXT, in the order the numbers point: (1) RESOLUTION LADDER -- rerun the bench at 192 and 384 cells along the longest axis (a conf row, not a constant) and publish agreement versus cell size; the hand and face weights are sub-cell at 96. (2) A REAL POSE FOR LEG C -- when the library pose moves no joints, synthesise the pose gate's subtree swing on an artist limb (compose ag_append_swing into the lib, one writer) so md_measure judges a deformation, and add pose_id to the CLI grammar. (3) SEMANTIC JOINTS (M4) -- compose nx_nxa_retarget's topology-plus-bind-position mapping onto the curve skeleton and re-score coverage per artist joint class, since the 461 permil is dominated by the joints a medial peel cannot invent. (4) THE OTHER DONORS -- toon3d8, witch, paladin through the same bench so the numbers are a population, never one character. Every rung prices against the numbers above, which stay on this board as the baseline.
2026-09-06	DC16	queue	LOCAL SIGNIFICANCE, ACCEPT RULE DECLARED BEFORE THE RUN (2026-09-06): the peeler's stop was GLOBAL -- a tip counted only if it reached farther than the body's ROOT medial radius -- so a 60 mm finger on a hand attached to a torso 110 mm thick could never be a limb, which is exactly why the referee read 461 permil coverage with fingers, toes and face joints absent by construction. Landed in nx_curveskel_lib: a tip is a limb when it reaches farther than the medial radius WHERE IT ATTACHES (dsurf at the path cell beside the skeleton); an insignificant tip is claimed without a joint; the loop ends when nothing lies farther than one cell from the skeleton, bounded by CS_MAX_PEELS=512 with truncation announced; peels= and insignificant= are printed. Also landed: cs_run_cells (a caller's resolution override above the conf row, argv over conf over default) and, in the referee, distinct strongest-joint counts for the artist skin and ours (the instrument for the helper-joint hypothesis behind 241 permil). ACCEPT RULE: prisms 2, T 3, dragon 6, separated biped 5, shouldered 5, touching 5 -- every fixture count unchanged, because at the root the attachment bar equals the old global bar; on ref9d the artist-joint coverage must RISE from 461 permil and the joint error mean must FALL from 208.9 mm, at 96 cells and again at 192; a change that only moves the number by adding spurious limbs (spurious permil rising faster than coverage) is refused.
2026-09-06	DC16	measure	LOCAL SIGNIFICANCE, FIRST CUT, AGAINST THE ACCEPT RULE (2026-09-06, staged runs): prisms 2 and 2 (peels 2, insignificant 0), T 3 (peels 3, insignificant 0) with every banked value byte-identical, nx_curveskel_gate 16 of 16; dragon 6 (peels 6, insignificant 0, identical leaves); separated biped 5 (peels 12, insignificant 7 -- seven extra candidates examined and every one correctly rejected), shouldered 5 (peels 9, insignificant 4). The TOUCHING biped FAILED: 8 limbs -- the x=+8 leg peeled FOUR times, tips at (8,3,-51) then (8,-3,-36), (8,-3,-30), (8,-3,-24). The leaf readout names the cause: the first descent hugs y=+3 on a 20 x 10 block whose medial set is a PLANE, the path cell's dsurf under-reads the half-width, the claim tube under-covers, and a one-cell sliver survives along y=-3 that, under a bar taken from ITS OWN attachment cell (dsurf 1 to 2 cells at the leg's edge), passes as three limbs. The old global bar hid the same sliver by accident. Not a patch: ONE primitive was missing, the local medial radius at a cell = the thickest dsurf in its neighbourhood grown until it stops growing (cs_local_radius) -- the axis cell's dsurf IS the cross-section -- now used by BOTH the claim tube and the significance bar. The earlier 'relaxed claim' was this primitive's first half applied against the wrong disease (the wall) and refuted for it; with the wall stripped, the sliver is the disease it was actually for, and the bar half is what makes it complete.
2026-09-06	DC16	measure	REFEREE ON THE NEW LIBS (local significance, local medial radius, wall strip), staged CLI fc8cb84a run as nx_rigbench_staged on ref9d 2026-09-06. At 96 cells: artist_joints_covered 461 permil (unchanged), joint_err_mean 20895 umm (was 20890), spurious 82 permil, top1 241 permil, L1 1354 permil -- no movement, and at 1.9 cm per cell a finger is sub-cell so resolution binds before the significance bar does. At 192 cells: covered 923 permil, joint_err_mean 4010 umm (max 16868), spurious 63 permil, our_joints 190, limbs 7, top1 233, L1 1331. The 192 gain is UNATTRIBUTED until the promoted binary d4617b00 (old libs) runs at 192 -- that control is running (/tmp/rb192_old.txt); a resolution-only gain would show there too, and only the difference between the two 192 runs belongs to this change.
2026-09-06	DC16	measure	nx_autorig_creature_gate staged 0702106b (the complete local-radius primitive) reads 11 of 12: dragon 6, biped 5, shouldered 5 unchanged; the ABUTTING-legs biped peels 6 (extra leaf at (8,-3,-36), a 15 mm sliver up one leg on the minus-y side) against the rule's 5, so the pre-declared accept rule FAILS on that fixture while walls_stripped=890 shows the wall strip firing. That fixture is the one this board already names invalid for the peeler (two boxes sharing a face are one solid in an SDF union, lesson row 1788693100): the queued grooved-thigh fixture is the honest touching control, and the abutting one stays on the gate as the wall-strip witness only, never as a limb-count assertion. Nothing in the lib chain promotes until that control exists and the 192-cell attribution is read.
2026-09-06	DC16	measure	THE REFEREE'S DEFORMATION LEG, MEASURED ON A REAL POSE (2026-09-06, the #1 leverage step: measure the right thing). The earlier runs drove rb_run with pose_id=1, which does not exist on ref9d (its POSE library holds exactly id=0), so nx_nxa_play replayed identity, joints_posed=0, and md_ours_vs_artist read a degenerate 0.2 mm -- the leg was silently vacuous. Re-run with pose_id=0 at 192 cells: pose 0 moves ALL 104 joints by up to 54.9 mm (min 0.46 mm, spread 54.5 mm) on the artist-weighted play and 55.1 mm on OUR-weighted play, both zero_weight_verts=0, both rc=0. md_measure between the two posed skins (same artist skeleton, same pose, artist SKIN vs our re-solved geodesic SKIN): p95 = 6.3 mm, max = 41.3 mm on a 55 mm motion. THAT IS THE HONEST QUALITY NUMBER, and it reframes the gap: top-1 LABEL agreement read 233/1000 because it penalises naming a joint differently even when the deformation is identical; on the metric that matters -- deformation -- our sovereign geodesic weights track the artist rig to 6.3 mm at the 95th percentile, worst 41.3 mm in the sharpest-bend region. NOT a SkinTokens comparison (that is still the oracle-image lane MO10/MO11); it is our-weights-vs-artist on the same skeleton, isolating the weight solve. Two rungs fall out: (a) rb_run must AUTO-SELECT a pose that moves joints (a degenerate arg silently vacated the leg -- a fixture-reached-the-condition failure the referee should refuse, not average to 0.2 mm); (b) the worst 41.3 mm names the region a learned refiner (MD4) must beat the geodesic control on.
2026-09-06	DC16	land	REFEREE HARDENED AGAINST THE SILENT-VACUOUS DEFORMATION LEG (2026-09-06). nx_rigbench_lib rb_run now AUTO-SELECTS a pose that moves joints: it tries the requested pose_id, then scans 0..poses-1, and takes the first whose ARTIST play reports joints_posed>0 (read from the captured NXA-PLAY line by the new rb_after helper); it prints deform_pose= and deform_pose_joints_moved=, and if NO pose moves it sets md_max/md_p95/md_refused = -1 and ABSTAINS rather than reporting a false near-zero. Two CAS edits on the lib (19,437 -> 20,714 -> 21,827 B), CLI rebuilt 123,056 B aa8c54b8 (debug), staged as nx_rigbench_staged. PROVEN LIVE: run at 192 cells with the deliberately-DEAD pose_id=5 -> the referee auto-selected deform_pose=0 deform_pose_joints_moved=104 and played pose 0 on both weightings (artist 54.9 mm, ours 55.1 mm max displacement), where the OLD code would have replayed bind and read md 0.2 mm. This is the 'a fixture the defect cannot fail is not a test' law applied to our own SOTA instrument -- the deformation number (p95 6.3 mm, max 41.3 mm on a 55 mm pose) is now measured on a pose that provably moves, and a static library can no longer flatter the verdict. OWED (immediate): rebuild nx_rigbench_gate on the fixed lib with a non-vacuity tooth asserting deform_pose>=0 AND deform_pose_joints_moved>0 (and a neg-control: a single-static-pose fixture must ABSTAIN, md=-1), then promote the CLI + gate. SOTA-vs-SkinTokens remains gated on the oracle head-to-head (MO10, a container that runs their published torch weights -- an operator infra decision on a GPU-less NAS) or a sovereign learned weight-refiner (MD4); neither is conjurable in-session, and the honest number to beat is now deformation p95 6.3 mm, not the label metric 233 permil.
2026-09-06	DC16	measure	SECOND-TOPOLOGY REFEREE (2026-09-06, full-population-not-a-sample): a pose was synthesised onto the dark_witch donor by retargeting ref9d pose 0 through nx_nxa_retarget emit=pose (witch_p.nxa: 111 joints, 50,230 verts, 57 mm motion, limb-length residual 0.79 mm, mapped=31 inherited=80), then the hardened referee ran on it at 128 cells. RESULT, and it CORRECTS the ref9d picture: coverage 396 permil (vs 923 on the house rig), joint_err_mean 190 mm (vs 4 mm), our_limbs 10 (the cape/hair add protrusions), spurious 274 permil, top1 104 permil. The auto-select worked (deform_pose=0, 111 joints moved on both plays: artist 53.9 mm max, ours 46.3 mm max -- our weights under-drive the extremities), but md_refused=1 so the per-vertex deformation delta is UNMEASURED on this donor. HONEST READING: ref9d, a clean biped, was the easy case and overstated us; the classical curve-skeleton is competitive only on clean topology and degrades badly on a complex/clothed character -- exactly the robustness a learned model trained on diverse rigs (SkinTokens/UniRig on ArticulationXL) provides, which STRENGTHENS the case that the gap closes via the oracle head-to-head (MO10/MO11) or a learned refiner (MD4), not a falloff tweak. NEW CALLOUT: md_refused=1 on 50,230 verts where it measured 14,164 is a likely vertex-cap blindness in md_measure -- the deformation leg silently cannot grade the LARGE/complex donors, which are the ones we are worst on (the same measurement-blindness class as the dead-pose leg fixed today). Rung: raise/remove md_measure's vertex cap (compose sys_read_file sizing, never a guess) so the referee can grade any donor, then re-run the 2-donor deformation population. The number to beat remains deformation p95, and it must be measured across TOPOLOGIES before any weight-solver change is adopted -- a change validated on ref9d alone is the vacuous-tuning defect.
2026-09-06	DC16	retract	CORRECTING MY OWN ROW 1788709700 BEFORE IT PROPAGATES. I called witch md_refused=1 a 'likely vertex-cap blindness' -- WRONG. Read md_grid_build (nx_mmdev_lib): it returns 0 ok / 1 extent-refused / 3 empty, and refuses with code 1 when the mesh bounding-box extent exceeds MD_EXTENT_MAX = 25000 (the overflow-DERIVED bound: the barycentric products stay under the i64 ceiling only below ~2.5 m at 0.1 mm; the header states this and refuses BY NAME above it). So md_refused=1 is a CORRECT overflow-prevention guard, not blindness, and RAISING the cap would re-introduce the very overflow it prevents -- the opposite of the fix I implied. WHAT IS CONFIRMED: the deformation leg cannot grade the witch as-is. WHAT IS NOT: whether the witch is genuinely authored beyond the extent (the modding lane measured donors from 0.21 m to 210 m, so plausible) OR rb_nxmesh feeds md coordinates in the NXA's 0.01 mm unit while md's bound is in 0.1 mm (a 10x unit mismatch would refuse a normal-size character) -- these two have OPPOSITE remedies (scale-normalize the asset vs fix the unit in rb_nxmesh), so I will NOT name a remedy until the experiment runs: print the extent md_grid_build computed for both ref9d (which passed) and witch (which refused), and compare to 25000. The witch SKELETON numbers (coverage 396 permil, joint_err 2.8x mean bone) are ratio-based and stand: the classical fit is genuinely worse on complex topology. Only the md-refusal mechanism was mis-stated.
2026-09-06	DC16	measure	MD-REFUSAL SETTLED 2026-09-06, the experiment row 1788709950 committed to, done by READING the conversion rather than raising any cap. rb_nxmesh (nx_rigbench_lib) sets each coordinate xyz9 = w / RB_NXA_PER_TENTH with RB_NXA_PER_TENTH=10, i.e. NXA 0.01 mm divided to md's 0.1 mm unit, and md's own NM_UNIT_PER_M=10000 is tenths of a mm, so producer and consumer units MATCH -- md_refused=1 is NOT the 10x unit mismatch I could not rule out. It is genuine authored scale: for the extent guard MD_EXTENT_MAX=25000 (0.1 mm = 2.5 m) to fire, the witch's posed coordinates must span past 2.5 m in their own units, consistent with the modding lane's measured donor range 0.21 m to 210 m. The guard is a correct i64-overflow prevention (barycentric products stay under the ceiling only below ~2.5 m at 0.1 mm) and MUST NOT be raised -- raising it re-introduces the overflow it exists to stop. ref9d passed because a ~1.7 m human is 17000 in 0.1 mm, under the bound.
2026-09-06	DC16	queue	REMEDY FOR md_refused ON LARGE DONORS, named by the settled mechanism above, NOT a cap raise: a recenter-and-normalize-for-measurement pass in rb_run leg C -- translate both posed skins (artist and ours) to their shared centroid, and if the extent still exceeds MD_EXTENT_MAX scale BOTH by one factor into the bound, run md_measure, then multiply the returned p95/max back by that factor so the number is reported in true mm. This is the field's standard normalize-before-a-distance-metric and leaves md's overflow guard intact, so the referee can grade any donor regardless of authored scale. ACCEPT RULE, pre-declared: ref9d (which passed unscaled) reads the SAME p95 6.3 mm / max 41.3 mm after the pass (a scale factor of 1 is a byte-identical no-op), and the witch's deformation delta becomes a finite mm number instead of md_refused=1. Until it lands the witch deformation leg reads UNMEASURED-BY-SCALE, never a false near-zero -- the same discipline as the auto-select fix that abstains on a dead pose.
2026-09-06	DC16	queue	BAR DATING OWED ON THIS BOARD under the 2026-09-06 rule that a bar is the current month's or it is refused: the five bar rows above (Blender 5.0, ZBrush 2026, Maya, Substance 3D Painter, the browser-native editors) carry 'read 2026-08-25' only in prose and no month field, so once nx_compare_rank composes nx_barfresh they classify MALFORMED and this domain is REFUSED (exit 5 BAR-STALE) until each source is re-read this month, mirrored and pinned in dcc.refs, and the row rewritten in the dated seven-field shape (bar, id, YYYY-MM, subject, what, ref key, note) with a barscan row per ref. The motion board was converted the same day as the worked example (MOB1 to MOB5). CARRIED HERE FROM motion.plan row 1788710600: SkinTokens (arXiv 2602.04805) is a FEBRUARY 2026 bar. The September 2026 arXiv listing (ref key arxiv_autorig_2026_09 on motion.refs, read from the mirrored bytes) places the field at rig-free feed-forward posing (Make-It-Poseable v3, updated 2026-09-02) and, for skinning weights, a GEODESIC geometric prior augmenting learned generation (A Geodesic Cut-Cell Prior for Neural Skinning, 2608.11272, 2026-08-11) because learned methods lack the generalisability of classic geometric approaches -- which is exactly this rung's bone-heat class plus the M6 learned refinement. The M2-to-M6 ladder stands; its TARGET is now that August-September 2026 pair, never SkinTokens, and the referee that grades the ladder is unchanged.
2026-09-06	DC16	land	REFEREE HARDENING PROMOTED 2026-09-06: nx_rigbench live 123,056 B sha aa8c54b8 (prev banked), after nx_contentdiff live-vs-staged GREEN lost_from_live=0 ONE-WAY-GAIN (142 runs gained: deform_pose=, deform_pose_joints_moved=, the deformation-leg ABSTAINS message, artist and our top-1 distinct-joint counts, and peels= and insignificant= from the local-significance curveskel lib it was built on). rb_run now auto-selects a pose that moves joints and abstains with md -1 when none does, proven live earlier today with a dead pose_id=5 auto-resolving to pose 0 on ref9d. OWED before this reads complete: nx_offc_install nx_rigbench promoted so runners fork the new twin (issued), a re-run of nx_rigbench_gate on the promoted binary (deferred: the box is at swap 950 permil), the non-vacuity tooth in that gate (deform_pose >= 0 AND deform_pose_joints_moved > 0, with a single-static-pose fixture that must ABSTAIN), and the scratch registration nx_rigbench_staged reaped. Under the current-month rule the referee's TARGET is the August to September 2026 pair recorded in the queue row above, and the deformation p95 6.3 mm on ref9d stays the number a learned refiner must beat on the same referee.
2026-09-06	DC16	land	DARK CONTRACT MECHANISM landed estate-wide 2026-09-06 (motion.plan row 1788715000 carries the full recipe): a watch row whose organ exists without its contracted symbol now reads DARK on the page, the comparewatch plane and the ranker instead of OPEN forever, which is the permanent form of the amesh_run lesson on this rung. This board's five dated bars moved to the sotabar kind the freshness ruler reads (the bar kind is the renderer's research-bar grammar); sotatarget and rungrole rows are owed. nx_rigbench PROMOTED d441bc19 with the normalize pass; nx_rigbench_gate carries the T12 non-vacuity tooth and is queued as a sovereign build -- the gate re-run on the promoted binary and the witch re-measurement (a finite deformation number under md_scale_div) follow that build.
2026-09-07	DC11	measure	Licence partition of the operator's 2026-09-06 scan and market references, read from the mirrored bytes: [ten24_samplescan] usage is commercial and personal WITH A CREDIT (full head and body scan, OBJ package with 16k textures and normal maps) = INGESTABLE as a DATA asset carrying its credit; [ss_free_head] is PERSONAL USE ONLY with a contact-us for commercial (8k albedo, normal, displacement) = evaluation oracle only; [ss_free_basemesh] [ss_multi_expression] [ss_free_metahuman] licence not yet read from the page = candidates, not assets; [booth_custom_model_adult] [jinxxy_human_avatars] are market bars with per-item paid licences = never an ingest source; [ss_blog_index] is the catalogue.
2026-09-07	DC11	queue	Scan ingest recipe, all sovereign: fetch the ten24 OBJ package (S3 zip, size unmeasured; nx_research_fetch refuses a body that fills its reserve, so a streaming-to-disk fetcher may be the first build order) then nx_bundle_ingest <zip> source=https://ten24.info/sample-scan/ licence=ten24-credit (LIVE, REGISTERED-DARK, /compare/modding MD1) then nx_obj_import.nx (LIB, sovereign Wavefront reader) into an NXA with TEXC/TEXM from the scan's 16k albedo and normal maps, then the donor-TEXM lane onto the cast; the credit line ships beside the asset.
2026-09-07	DC11	lesson	The critic loop the operator's pasted Infinigen-plus-Blender plan proposes (landmark oracle, Chamfer critic, bounding oracles for garments) already exists in NishiLang and is entirely DARK: nx_meshdist, nx_fitloop, nx_facemark, nx_bodyfit, nx_fbx2nxa, nx_part_solver and nx_bundle_ingest all read REGISTERED-DARK on 2026-09-06 (callable, authorised, never invoked over MCP). Blender and Python never enter the build path; the sovereign generator is judged by these organs with a licence-clean scan (ten24) as the oracle mesh, and the Daemon Girl asset itself is never ingested (mod licence forbids redistribution) -- it is the visual bar.
2026-09-10	DC16	measure	2026-09-10 AUTO-RIG INVESTMENT RECOVERED / CLAIMS REQUIRE CURRENT QUALIFICATION: original Claude session789442b2-9b90-4bf3-b4d4-db659ceb3662 contains native results for curve-skeleton16/16, geodesic weights15/15, bend10/10, creature12/12 and rigbench11/11, followed by a later creature candidate11/12 RED. These are substantial existing implementations, not a four-test prototype;4/4 only describes the one-call composition gate. The rigbench pass validates its ruler, not high-quality artist-rig parity. Fresh nx_catalog job1789076127 finds7related owners built/promoted/registered/ever-authorized but invocationNO in a4194304B recent tail; this does not erase historical runs or establish current successful consumer outcomes. Reconcile immutable artifacts, generation/consumer identity, current tests and served Beach use before reuse or rebuilding. Existing dcc/motion/modding ownership retained; no duplicate auto-rig service. Full historical result chronology retained locally and indexed by knowledge/gates/autorig-history-reconciliation-t131.json. User directs records be references to ideas, feedback and investment, never automatic adoption of old decisions or unsupported superiority claims. Exact provider token spend remains unknown.
2026-09-22	DC17	queue	2026-09-22 declared from intake 1790106111 by the PM tier: nx_intake_route propose found no rung on any board carrying the request at three words (only stem's watch gap), so the render leg is authored here -- this board owns the authoring tool, procgen owns the explode engine, graphics the scene renderer; the artwork line is the operator's car exploded-view apparel art
2026-09-22	DC17	queue	INTAKE 1790106111 routed by nx_intake_route (operator, epoch 1790108121) from operator: EXPLODED-VIEW CAR ART FOR APPAREL: a series of well-artisted exploded-view illustrations of favourite cars of all time, print-ready for the back of a t-shirt or sweater (operator 2026-09-22, verbatim: -- sponsor intake-1790106111 plus2
2026-10-01	DC11	queue	CLAIMED by seat asset-ingest (cross-board lane, T0 1790830757, step 17 of the gameengine GE83 program): building ain_admit as the ONE intake admission, composing the incumbents rather than twinning them -- nx_asset_prov_lib for the sha256 content pin and the journal row, nx_licgate_lib for the rights table, nx_consent for the likeness record; every refusal names the missing conjunct; the gate carries a positive control, one tooth per conjunct asserting WHICH rule fired, an admit-everything stub neg-control and a same-size different-content pin tooth. DC15 follows as a reading of what is provable without a headset.
2026-10-01	DC11	queue	RESUMED by seat asset-ingest (second launch, T0 1790857060; the first launch was cut off by an outage before it wrote any organ: func ain_ read ABSENT-PROVEN over buildroot/runtime at job 1790857425, coverage_complete=1 corpus_complete=1). Three sources landed this hour: runtime/nx_asset_admit_lib.nx (ain_admit, the one intake decision, composing nx_asset_prov_lib for the sha256 pin and the journal row, nx_licgate_lib for the one rights table and nx_consent for the likeness record), runtime/nx_asset_admit.nx (the command) and runtime/_hdl_build/nx_asset_admit_gate.nx. The command compiled first try at job 1790858814 (61307 B, sha256 e0d9b3771e8f2210f2b28d2fdb06b815997fe2bf24e32cd0d1eb8ed39efa68b3); the gate build is in flight. Nothing is promoted yet.
2026-10-01	DC11	land	ain_admit is real: runtime/nx_asset_admit_lib.nx (18106 B) holds the ONE intake decision and composes the incumbents instead of twinning them -- the sha256 pin and the journal row are nx_asset_prov_lib (one journal, last row wins), the licence lookup is nx_licgate_lib (the one rights table knowledge/model_license.conf), the consent predicate is nx_consent. An asset is admitted only when its admission record (one key=value per line) names origin, a licence id the rights table carries, a sha256 equal to the sha256 of the bytes presented and a likeness declaration of 0 or 1, and a likeness also carries a consent record whose release is signed and whose term has not lapsed; every refusal returns a reason and the conjunct it belongs to (record, asset, origin, licence, pin, likeness, rights-table, consent, journal), a duplicated key is refused as ambiguous, a refusal writes nothing, and the same admission twice appends one row. PROMOTED: nx_asset_admit (the command, 61307 B, sha256 e0d9b3771e8f2210f2b28d2fdb06b815997fe2bf24e32cd0d1eb8ed39efa68b3) and nx_asset_admit_gate (134614 B, sha256 921ce5dee6106396522c6e8a40fcc86905ba275341d8c36a98b7f33e6daa44ef), both registered, both _offc twins installed, the gate rostered and added to dcc.gates; the dcc.matrix watch row is repointed from the glTF importer to runtime/nx_asset_admit_lib.nx so the contract symbol is measured where it lives. Shipped, referee grade 60 of 60 with the mutant killed, awaiting acceptance.
2026-10-01	DC11	measure	nx_asset_admit_gate through /api/gate_run on the promoted binaries (job 1790859921, exit_code 0): passed 60 of 60. Values the gate printed, each recomputable: refusal_cases=17, refusal_cases_exited_refuse=17, real_admission_cases_satisfied=17, admit_everything_stub_cases_satisfied=0, journal bytes 134 after the first admit and 134 after all seventeen refusals (by hand: the tag 5 plus sha 64 plus source 29 plus licence 10 plus lane 10 plus epoch 10 plus five tabs and a newline is 134), the CRLF record 143 bytes against 139 for the LF record. Second method classes: the pin is checked against the FIPS 180 known answer for the three bytes abc and not against itself; the provenance row is read back through pv_row_find and not through the admission receipt; the promoted command, forked on the same journal, exits 0 with row=present for the admitted asset and exits 4 naming reason=PIN-MISMATCH conjunct=pin for the same-size different-content twin. nx_gate_bite over the lib (job 1790859987): valid_mutants=1 killed=1 not_reached=0 void_build_lane=0, restore byte-identical at 18106 B, twins_checked=2 restored=2 mutant_live=0, and nx_filehash cmp reads the live gate and its _offc twin identical at 921ce5de afterwards. One mutant is a floor on the gate's bite, not a mutation score.
2026-10-01	DC11	lesson	An absence probe for a contract symbol must be asked as a DECLARATION, never as the bare name: nx_absent for ain_admit over buildroot/runtime read matches=75 PRESENT, and every hit shown was domain_admit, whose last nine letters are the contract symbol. The same probe asked as func ain_ read ABSENT-PROVEN at coverage_complete=1. A bare-name probe would have told the next seat the rung had already been built. Second lesson, about the board itself: the watch row named the glTF importer as the organ the symbol would live in, and a decision every intake door must share cannot live inside one of the doors; the row was repointed to the lib rather than a wrapper being planted in the importer to light the cell.
2026-10-01	DC11	queue	OPEN HALF OF THE RUNG, named so the lit cell is not read as the whole of it: (1) no intake door calls ain_admit yet. nx_gltf2mesh, nx_fbx2nxa and nx_bundle_ingest still import anything; wiring each means an admit=<record> argument that refuses by name, their own gates gaining a refused-without-a-record tooth, and a decision on house assets (the export door nx_mesh2glb passes a row-less asset as a house asset; an intake door that fails closed needs the same question answered before it is switched on, or the importer gates that feed fixtures break). (2) The export doors read pv_verdict only; a row whose origin column says dcc-ingest likeness consent=<id> is not yet read by any delivery door, and tier and use scope are asked nowhere after intake. (3) The consent record is carried inside the admission record as declared fields; nx_rights_ledger (grants per talent and use scope, in memory only, its persistence rung unbuilt) is the incumbent a durable consent store should grow from. (4) OPERATOR DECISION OWED: this rung's text refuses a likeness without a consent record at intake, which is what was built; the 2026-09-30 likeness policy builds reproductions inside the estate flagged estate-only and puts the line at the production door. Whether intake should admit a consent-less likeness under an estate-only flag instead of refusing it is not a seat's call and was not guessed. Recipe to verify what shipped: nx_mgmt POST /api/gate_run target=nx_asset_admit_gate, then nx_asset_admit admit <asset> <record> on a real asset with the pin from nx_filehash.

On these two registers. Rows are declared in the domain's plan file and carry the debt id, which is the join key back to the sovereign debt plane — that plane, not this page, is the authority on state. Reconciling them automatically (the regen reading the plane and refreshing these rows) is a named, owed rung; until it lands, treat an id here as a pointer to look up, not a status to trust.

Honest verdict. The coverage above records source-symbol presence. Stronger comparisons are authored claims, not measured runtime superiority or accepted outcomes. A watch contract names an expected source symbol; finding it changes source discovery only. Qualification, deployment and runtime acceptance require their own evidence.

Person · product · place — not yet measured for this domain

Every compare carries this layer. Declare knowledge/compare/dcc.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 dcc, 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).

The field — discovered, not chosen

Rows written by nx_field_discover from dcc.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.

Discovery receipt. One row per counter the run wrote, in the receipt's own order; the rival rows follow in their own table.

Field discovery, counted
Class	Count	Note
seeds	4
scanned	4
fetched	4
reused	0
failed	0
named	4
candidates	4
mentions	4
capped	0

Rank	Rival	Seeds	Mentions	First seed	Kind	Link
1	Blender 5.0	1	1	col0-blender	named	https://www.blender.org/download/releases/5-0/
2	ZBrush 2026	1	1	col1-zbrush	named	https://www.maxon.net/en/zbrush
3	Maya 2026	1	1	col2-maya	named	https://help.autodesk.com/view/MAYAUL/2026/ENU/data/toctree.json
4	ArmorPaint	1	1	col3-armorpaint	named	https://armorpaint.org/manual

Exclusive and shared capabilities

One capability is graded on one board; every other board references it. The fleet duplicate census (nx_rivalcensus dupes, on the fleetdupes clock row) reads every board's matrix and names each capability graded as a row on two or more boards, with a home derived from evidence: rung when exactly one board's plan contracts the symbol (that board is starred), gates when no plan contracts it and exactly one board's executed gate map runs the organ's gate (that board is plussed), ambiguous when several boards carry the deciding evidence (a decision for their seats), and none when no evidence decides it, with the reason beside it: unmapped (no board of the duplicate declares a gates map; declare the maps) or unproven (a map exists and runs none of the organ's gates; gate the organ on its proving board, or declare it a shared library by policy). A home elsewhere means this board should carry an @xrow reference to the home board's row rather than a graded row of its own; a home here means the other boards should. The table is this board's slice of one estate-wide census: the shape is shared, the rows are the board's.

This board grades 74 capabilities: 61 exclusive to it and 13 shared with other boards. Shared rows on this page: 13 = home here 1 + home elsewhere 1 + ambiguous 3 + undecided 8 + refused 0 (partition 13); undecided = unmapped 0 + unproven 8 + unevidenced 0; homonyms 0; since the previous census: new 0, resolved 0. Census FRESH age_s=6888.

Capability	Organ	Boards grading it	Home	Rule	Evidence	Standing on this board
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
sd_step	`nx_softdyn.nx`	graphics, procgen, softbody, dcc, gameengine	—	none	unproven	undecided
g2_convert	`nx_gltf2mesh.nx`	koikatsu, dcc, anatomy, gameengine	—	none	unproven	undecided
md_measure	`nx_mmdev_lib.nx`	procgen*+, softbody+, dcc, anatomy+	procgen	rung	plan	home elsewhere adopt: `@xrow procgen\|md_measure\|home by rung (fleet duplicate census)`
nx_perlin_fbm_2d	`nx_perlin.nx`	procgen, dcc	—	none	unproven	undecided
nx_csg_extract	`nx_csg.nx`	procgen, dcc	—	none	unproven	undecided
st_substep	`nx_softtissue.nx`	softbody+, hapticcare+, dcc	—	ambiguous	gates	ambiguous
mv_manifold	`nx_meshvalid_lib.nx`	dcc*+, anatomy+	dcc	rung	plan	home here
sk_skin	`nx_skeleton.nx`	dcc, npcai	—	none	unproven	undecided
g2_skin_weights	`nx_gltf2mesh.nx`	dcc, gameengine, modding	—	none	unproven	undecided
oi_read	`nx_obj_import.nx`	dcc, modding	—	none	unproven	undecided

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 1395993 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_boneheat_lib`	dcc	plan-queue	rung=DC16

rows shown 1|this board's directives 1|estate-wide un-boarded organs 452 (listed in full on /compare/ecosystem)|census rows 639|malformed 0 (counted, never rendered)

References

Beyond a link list. Every reference below resolves twice — the publisher's copy and, where banked, the estate's own non-rottable library mirror with a content pin — and carries its evidence class plus the exact claim on this page it grounds. Keyed marks like [key] in the matrix notes jump here. A dash means honestly absent, never assumed.

[blender50] Blender Foundation. Blender 5.0 release notes. Carries the Boolean solver rename to Float, the Principled BSDF iridescence work, Geometry Nodes and USD NURBS curve I O. publisher · read in our library knowledge/fetched/cmp_dcc_blender50.html · pin h2086a760de5d60f1f6fd36f712d081478ff90b073c3b5710c54e9482086fdb59 · accessed 2026-08-25 · vendor-docGrounds: sdf-boolean-union-difference-intersection
[blenderfeat] Blender Foundation. Features overview. States in its own words that Blender is the free and open source 3D creation suite. publisher · read in our library knowledge/fetched/cmp_dcc_blender_features.html · pin hb19c12b5673f1a6576277db29a0a631b0de8f27728675f8bf7e21f69c73b9d5b · accessed 2026-08-25 · vendor-docGrounds: parametric-re-evaluation-at-depth
[blendersculpt] Blender Foundation. Sculpting features. Twenty brush types including Crease Clay Strips Pinch Grab Smooth and Mask, dynamic topology tessellation, and multiresolution sculpting. publisher · read in our library knowledge/fetched/cmp_dcc_blender_sculpt.html · pin h4d52b02f5613a22da3f37cb9aacbab0e3a3549c88a107678fa68f6104cd84bea · accessed 2026-08-25 · vendor-docGrounds: sculpt-brush
[blendermodel] Blender Foundation. Modeling features. Conformal and Angle Based UV unwrapping with edge seams and vertex pinning, non-destructive modifiers, and painting directly onto the mesh. publisher · read in our library knowledge/fetched/cmp_dcc_blender_modeling.html · pin h97cba1c773c7e957db68ef4d807be20dffc68d28b0256313891ae0f3ec9bfb65 · accessed 2026-08-25 · vendor-docGrounds: general-uv-unwrap
[blenderanim] Blender Foundation. Animation features. Envelope skeleton and automatic skinning, weight painting, keyframing and the F-Curve Graph Editor. publisher · read in our library knowledge/fetched/cmp_dcc_blender_animation.html · pin ha06234fc37cddac05e77fb95f6c05b75140316fb135cf8c9a15054fe5c141e21 · accessed 2026-08-25 · vendor-docGrounds: keyframe-sampling
[blendersim] Blender Foundation. Simulation features. Cloth simulation for clothing and environments, and hair that responds to wind and collisions. publisher · read in our library knowledge/fetched/cmp_dcc_blender_simulation.html · pin h66f98c61910ac4b57a8a59f61bf2f0b4dbdedc54d388fa4551bfba1810da7e79 · accessed 2026-08-25 · vendor-docGrounds: cloth-and-garment-drape
[zbrush26] Maxon. ZBrush product documentation. A dynamic brush system handling tens of millions of polygons in real time, Dynamesh remeshing, Live Boolean, UV Master, and PolyPaint painting directly on the surface with no UVs assigned in advance. publisher · read in our library knowledge/fetched/cmp_dcc_zbrush.html · pin h7cbff9983813e3920843571f3692d481767c610bef0b9c7d63e271bbabd30be3 · accessed 2026-08-25 · vendor-docGrounds: dynamic-topology
[mayatoc] Autodesk. Maya 2026 User Guide table of contents, the complete official topic tree. Carries Sculpting in Maya, Remesh Options, Unfold a UV mesh, Layout UV shells, Boolean operations, Construction history, the 3D Paint Tool, Standard Surface, nCloth, groomable splines, Bind Skin, the Graph Editor and a dedicated USD container. publisher · read in our library knowledge/fetched/cmp_3dcc_maya_toctree.json · pin h64301b4a3c40e6d4ac3d5f92668a20f5db3a622babd7238bac1a83076538076a · accessed 2026-08-25 · vendor-docGrounds: uv-packing
[mayasculpt] Autodesk. Maya 2026 Modeling, the sculpting topic page. Server-rendered prose carrying a product and component meta tag, which makes sibling-product attribution mechanical rather than a judgement call. publisher · read in our library knowledge/fetched/cmp_3dcc_maya_sculpting_topic.html · pin hbcc18b2a1e91389c68aee7c54991d6ff9f4bb7e12023de2c966b11fecf20e60c · accessed 2026-08-25 · vendor-docGrounds: multiresolution-levels
[armorpaint] ArmorPaint manual. Self-described as software for 3D PBR texture painting: brushes, node-composed materials, auto UV unwrap, and a sculpt mode the vendor itself labels work-in-progress. publisher · read in our library knowledge/fetched/cmp_3dcc_armorpaint_manual.html · pin h42754bc5bdaece1eeb7597819f20e99714c36f6530ac332272e7ea03a815fadd · accessed 2026-08-25 · vendor-docGrounds: texture-painting
[armortools] armory3d armortools repository readme. Open development with a public repository, and it states in its own words that distributed binaries are paid -- so open source must not be written as free binaries. publisher · read in our library knowledge/fetched/cmp_3dcc_armortools_readme.md · pin h005883548a1c64633a2ff47a40295b383d5ea8513ce903473f949875584452ec · accessed 2026-08-25 · source-readGrounds: node-graph-authoring
[blender_2026_09] Blender Foundation homepage read 2026-09-06: the hero reads Blender 5.2 LTS with node-based physics and online assets; the cards cite Geometry Nodes Physics in Blender 5.2 LTS and a Cycles texture cache that makes many-texture scenes far more memory efficient in 5.2 LTS. September 2026 says the free breadth bar is Blender 5.2 LTS, not 5.0. publisher · read in our library knowledge/fetched/cmp_dcc_blender_2026_09.html · pin h128261e311feeb87a3337f3499b08ac9e4199bae2f2d9a676f77c114e556497c · accessed 2026-09-06 · vendor-docGrounds: blender-breadth-bar
[zbrush_2026_09] Maxon ZBrush product page read 2026-09-06: the page names the current release ZBrush 2026.2.1 in its heading; very high polygon-count digital sculpting, brush libraries, sculpt layers, dynamic tessellation and remeshing. publisher · read in our library knowledge/fetched/cmp_dcc_zbrush_2026_09.html · pin h0fc07ec8a26db715458ff4bb14daf75f174e776cf0f0f8a9eb6f2c1a40723257 · accessed 2026-09-06 · vendor-docGrounds: zbrush-sculpt-bar
[maya_wiki_2026_09] Autodesk Maya, the Wikipedia article read 2026-09-06 as this month's witness because the vendor page at autodesk.com failed the sovereign TLS chain with a certificate verification error; the board's bar names Maya the rigging, animation and scene-assembly incumbent in film and games. publisher · read in our library knowledge/fetched/cmp_dcc_maya_wiki_2026_09.html · pin h244d6e8cf11ed5cd090493b0a7ac6d068be5ad947ac5a5370e5f071775aa182e · accessed 2026-09-06 · mirrored-indexGrounds: maya-rig-bar
[womp_2026_09] Womp homepage read 2026-09-06: a free, browser-based 3D modeling platform and on-demand 3D printing service; Liquid 3D builds with SDF and CSG shapes that blend, merge and carve, like clay but parametric; goop-based modeling, Spark AI generation, an asset library of thousands of presets, SVG import and export, all in the browser. publisher · read in our library knowledge/fetched/cmp_dcc_womp_2026_09.html · pin h7bc0b81c38e990917c86611441670c5be987c93b67cec058f604f7d3c4304b59 · accessed 2026-09-06 · vendor-docGrounds: womp-browser-native-bar
[substance_painter_wiki_2026_09] Substance Painter, the Wikipedia article read 2026-09-06 as this month's witness because the vendor page at adobe.com fails the sovereign TLS chain (status 0 three times); the mirror is 50,315 B from a fetch that closed early with status 0, so the pin covers a possibly-short body and only its witnessed title and head are cited: the encyclopedia carries an article for the texture-painting tool this month. The Adobe Substance 3D article redirects to Adobe Inc., which lists Substance 3D among its products and records the 2019 Allegorithmic acquisition of the Substance texture tools (that mirror is knowledge/fetched/cmp_dcc_substance_wiki_2026_09.html and is not cited as a Painter witness). publisher · read in our library knowledge/fetched/cmp_dcc_substance_painter_wiki_2026_09.html · pin hcd2463f652ed13439251220398cf0aa1c77130ef1b262c33e6db623bcb068662 · accessed 2026-09-06 · mirrored-indexGrounds: substance-texture-bar
[unirig] VAST-AI-Research. One Model to Rig Them All: Diverse Skeleton Rigging with UniRig, SIGGRAPH 2025, arXiv 2504.12451 (repository README read 2026-09-05): two stages, an autoregressive transformer predicting a topologically valid skeleton through Skeleton Tree Tokenization, then Bone-Point Cross Attention predicting per-vertex skinning weights and bone attributes; inputs .obj .fbx .glb .vrm; the README announces SkinTokens as its successor unifying skeleton and skin into one autoregressive sequence. publisher · read in our library knowledge/fetched/cmp_modding_unirig.html · pin h8d77b9eca87d2e91030ccd214a4e39027f3fe86b3ff706689af3bfa61ec6dad6 · accessed 2026-09-05 · published-paperGrounds: Auto-rig skin weights, geodesic bone-heat over the mesh volume
[rignet] RigNet, Neural Rigging for Articulated Characters, SIGGRAPH 2020, arXiv 2005.00559 (paper read 2026-09-05): a graph network predicts a joint-probability volume and a skeleton, then a mesh-attention network predicts per-vertex skinning weights. publisher · read in our library knowledge/fetched/cmp_rigging_rignet.html · pin h2974f27c68738edd267d36b91da24944706b244acd8f438258a1f9e841b7ffa6 · accessed 2026-09-05 · published-paperGrounds: Auto-rig skin weights, geodesic bone-heat over the mesh volume
[magicarticulate] MagicArticulate, Make Your 3D Models Articulation-Ready, arXiv 2503.16173 (paper read 2026-09-05): auto-regressive skeleton generation over the Articulation-XL set followed by a functional-diffusion skinning-weight field. publisher · read in our library knowledge/fetched/cmp_rigging_magicarticulate.html · pin h271dad275df6010f906d6c74c2d14ef762a6fbc5bf403495c06425eb77b1b58e · accessed 2026-09-05 · published-paperGrounds: Auto-rig skin weights, geodesic bone-heat over the mesh volume
[blenderboneheat] Blender Foundation. Manual, Armature Deform Parenting. With Automatic Weights calculates how much influence a bone has on vertices from vertex-to-bone distance by the bone heat algorithm, and its own text warns overlaps can occur that need manual weight painting. publisher · read in our library knowledge/fetched/cmp_rigging_blender-autoweights.rst · pin heeb39712935baabcf5972fed65951bb30af8c18f9ad7458ce81e317482ba06ac · accessed 2026-09-05 · vendor-docGrounds: auto-rig-skin-weights-bone-heat
[modbooru_daemon_girl] ModBooru mod index page read 2026-09-06, Daemon Girl by Ryan Reos (mod 256): adds the outfit, hair and weapons of Ryan Reos Daemon Girl; the download links are Nexus Mods Skyrim 105308 and Skyrim SE 41900, 49542 and 101479, tagged by body type BHUNP, UUNP, CBBE 3BA, CBBE SE, TBD SE and TBD SMP; the operator names this class the benchmark for a photoreal girl with true clothes and true anatomy (2026-09-06). publisher · read in our library knowledge/fetched/cmp_dcc_modbooru_daemon_girl.html · pin h3b7949b39c62bdf53f702f42a9172047b2e30acdd848d5e3ff5493077420663a · accessed 2026-09-06 · mirrored-indexGrounds: cloth-and-garment-drape
[blenderrigify] Blender Foundation. Manual, Rigify Introduction (v4.2 release branch, read 2026-09-05): Rigify automates the creation of rig CONTROLS and bones from meta-rig building blocks the user assembles, and it does not attach the rig to a mesh, so skinning is still done by hand. publisher · read in our library knowledge/fetched/cmp_modding_rigify.rst · pin h1f91b25a90cd7de30ea557b66c1f8f0ee3c317aee6f2fb17a759dc3844b9564f · accessed 2026-09-05 · vendor-docGrounds: auto-rig-skeleton-fit-curve-skeleton
[ten24_samplescan] Ten24 sample scan: full head and body photogrammetry, ZBrush + OBJ packages, 16k textures (PSD/JPG/TGA), normal maps, RAW; usage: commercial and personal with a credit publisher · read in our library knowledge/fetched/cmp_dcc_ten24_samplescan.html · pin hb87c3275c9b9b0519eeeb9fdef353bff5e7a312c6134cb0532926bfbec9755ff · accessed 2026-09-06 · vendor-docGrounds: licence-clean scanned reference for the face and skin lane (a DATA asset that carries its credit)
[ss_free_head] 3D Scan Store: Free 3D Head Model (HD female head scan, 8k albedo + normal + displacement) -- PERSONAL USE ONLY, contact them for commercial use publisher · read in our library knowledge/fetched/cmp_dcc_3dscanstore_free_head.html · pin h09339838cbcc55f1118bcd17ca12b18035e553ee2625e591721b3e32328318e7 · accessed 2026-09-06 · vendor-docGrounds: evaluation-only scanned head (the licence forbids commercial use without permission)
[ss_free_basemesh] 3D Scan Store: Free Base Mesh publisher · read in our library knowledge/fetched/cmp_dcc_3dscanstore_free_basemesh.html · pin h3226f3031349b8f566119504e1ec6bf8ac72aeccabe636996d2577a25886ca16 · accessed 2026-09-06 · vendor-docGrounds: topology oracle candidate (licence to be read from the mirrored page before any ingest)
[ss_multi_expression] 3D Scan Store: Free Multi Expression Basemesh publisher · read in our library knowledge/fetched/cmp_dcc_3dscanstore_multi_expression.html · pin hf409741d9189b4ff9ef24757a9799437df641fa67ba67dff323853e683b1353a · accessed 2026-09-06 · vendor-docGrounds: expression topology oracle candidate (licence to be read before any ingest)
[ss_free_metahuman] 3D Scan Store: Free Metahuman Identity publisher · read in our library knowledge/fetched/cmp_dcc_3dscanstore_free_metahuman.html · pin hd607c6adc0e5fc9ea271e9bcb1d7b421be8884b75a4f083bb1bf4c193cf330f3 · accessed 2026-09-06 · vendor-docGrounds: scanned identity reference (Metahuman-bound; oracle only)
[ss_blog_index] 3D Scan Store blog index (free asset listings and the paid face-texture catalogue) publisher · read in our library knowledge/fetched/cmp_dcc_3dscanstore_blog.html · pin h5c7799b8a7f3a43b3bec021dd562153a8fa25eb3b47fca942b4c59ba517d584d · accessed 2026-09-06 · mirrored-indexGrounds: market and asset catalogue for the scan lane
[booth_custom_model_adult] BOOTH search: Custom Model, adult-only (VRChat avatar market) publisher · read in our library knowledge/fetched/cmp_dcc_booth_custom_model_adult.html · pin h767ef36a38af83f46b2dc5043bc83503fb2cf6bb6836d46275995b4cf60d25b8 · accessed 2026-09-06 · mirrored-indexGrounds: adult-VR market ladder bar (per-item licences, paid, not an ingest source)
[jinxxy_human_avatars] Jinxxy market: human-type avatars publisher · read in our library knowledge/fetched/cmp_dcc_jinxxy_human_avatars.html · pin hb22999a0eb37c61d8c9d6988fbbc43a94fe3f820954f96d5a1c1544a3f6dbf18 · accessed 2026-09-06 · mirrored-indexGrounds: adult-VR market ladder bar (per-item licences, paid, not an ingest source)

generated by nx_swcompare_matrix (sovereign NishiLang organ) from knowledge/compare/dcc.matrix · source checks show implementation presence; runtime and user-outcome evidence are reported separately · JavaScript supports page controls