Nishi Family › Compare › Upload, Auto-Rig, Animate, Export

Nishi Compare · capabilities, evidence and delivery

Upload, Auto-Rig, Animate, Export

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

Nishi vs Mesh2Motion and Mixamo and Tripo3D and AccuRIG and Cascadeur and SkinTokens -- a visitor uploads a mesh, the sovereign auto-rig peels a skeleton from its volume with no archetype hint and solves geodesic weights, a library clip is retargeted onto it, and the result plays and exports, with the referee grade and the operator's verdict on the same page. Measured 2026-09-06: the dcc referee reads 923 permil artist-joint coverage and 40.1 mm mean joint error at 192 cells on the house rig but only 233 permil top-1 weight agreement; every organ the surface composes is live, and the surface itself did not exist -- the world's viewer pages were hand-written (emitter ABSENT-PROVEN) and the /mesh route fronts the gallery, not a mesh tool.

Overview

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

Where we are. MEASURED 2026-09-06 through the estate's own instruments, not recalled. The AUTO-RIG exists and is live: nx_autorig_mesh peels a curve skeleton from the mesh volume with no archetype hint and solves geodesic bone-heat weights in one call (dcc DC16, gate 4 of 4), the same code rigging a seven-box dragon to six limbs and a biped to five; the dcc referee nx_rigbench scores it against the artist rig on the house character ref9d: at 96 cells 461 permil of 104 artist joints within one bone length and 208.9 mm mean joint error, at 192 cells 923 permil and 40.1 mm, with top-1 weight agreement 233 permil and L1 1331 permil of scale on the artist skeleton -- coverage moved with resolution, weight topology did not, and the 192 gain is unattributed between resolution and the local-significance change until the global-bar arm runs. EVERY OTHER ORGAN THE SURFACE NEEDS IS LIVE: nx_gltf2mesh carries a rigged glTF into NXA (charsim R22), nx_fbx2nxa carries FBX geometry, nx_nxa_retarget drives a rig from a donor's POSE library preserving its own bone lengths by construction (37 of 37 on the NAS), nx_nxa_play bakes a pose over the whole mesh, nx_nxa_texbake view renders a three-view turntable with the sovereign rasteriser, nx_mesh2glb exports the rig as a GLB with skin, nx_bvhmotion reads BVH exactly, nx_content_put is a chunked sha-receipted upload door, and nx_container is the oracle-image incumbent. WHAT DOES NOT EXIST: a door a visitor can drop a mesh into, a job that composes the organs, a page that shows the rig and lets a clip be applied, a clip library as data, an acceptance row per upload, the rival riggers as oracle containers, and a shared benchmark. The world's own viewer pages (rigview, refmesh, walk3d) were hand-written: nx_absent over every organ source found no emitter (ABSENT-PROVEN, corpus_complete=1), and the /mesh route fronts the NSFW gallery gateway with no daemon behind it, so there is no mesh surface to grow. The rivals were read from their own pages, mirrored and pinned in motion.refs: Mesh2Motion imports four formats, ships six archetype rigs the user positions and over 150 CC0 clips, exports multi-clip GLBs and runs in the browser under MIT and CC0; Mixamo auto-rigs humanoids and retargets a motion-capture library; Tripo auto-rigs humanoids in seconds from a generated mesh and auto-animates; AccuRIG rigs humanoids and quadrupeds on the desktop; Cascadeur authors physics-based motion; SkinTokens unifies skeleton and skin into one learned autoregressive sequence and claims 98 to 133 percent better skinning and 17 to 22 percent better bone prediction than prior SOTA.

Where we need to go. A visitor drops ANY mesh -- GLB, FBX, NXA, later OBJ and the game-packed formats modding reads -- on nishifamily.com/motion, the sovereign auto-rig peels a skeleton from its volume with no archetype hint and solves its weights at the resolution the visitor chooses, the page shows the rig from three sides with the referee's own numbers beside it, a library clip retargets onto the fresh rig and plays, the result exports as a GLB any engine loads, and the operator's ACCEPT or REJECT lands on the acceptance plane from the same page -- all of it emitted from data by the estate, none of it hand-written, no third-party model or runtime in the run path. Exceed the field where it structurally cannot follow: rig any topology with no hint and no checkpoint, deterministically, on a NAS CPU; publish the referee grade with every result; run the rival riggers as outside oracles on the same upload so consensus replaces marketing; and grade everything on a shared licence-clean corpus run whole. Meet SkinTokens on its own axis before claiming to pass it.

The unit. 1 u = one measured session-leg. Calibration inherited from the sibling boards: mangagen R1 went from substrate to shipped, registered and live-verified in one leg on 2026-08-13, charsim R22 closed in two, and the DC16 one-call auto-rig closed in one on 2026-09-05. Estimates recalibrate as each rung lands.
Cost to see an upload rigged on the page: 3 u. Milestone M0: the daemon and its route row, the upload door with a named cap, the rig job at the ladder resolution, and the page emitted from data with a static fallback. After M0 the operator can drop a Kenney-class GLB and see its skeleton and weights from three sides with the composed organs' own numbers.
Cost to animate the fresh rig and accept it: 5.5 u. Through M1: the clip library as rows, the anim verb through nx_nxa_retarget and nx_nxa_play with a posed render and export, the referee card on every demo rig, and the ACCEPT or REJECT row to the acceptance plane.
Cost to ingest the world's clips and run the rivals as oracles: 10.5 u. Through M2: glTF animation channels and BVH into POSE libraries so the CC0 Mesh2Motion and Quaternius clips drive our rigs under a provenance row, the chunked door for models past the edge budget, and SkinTokens plus UniRig as oracle containers graded by the same referee on the same upload.
Cost to a shared benchmark number: 12.5 u. Milestone M3: a licence-clean artist-rigged corpus the referee runs over WHOLE, published on this board beside the rivals' own claims measured the same way. This is where meet or exceed becomes a number instead of a sentence.

Latest recorded release

2026-09-23 · MO3

DETERMINISM TEST REPAIR ACCEPTED 2026-09-23: the prior RED 43/44 did not prove nondeterminism because the repeated fixtures were not byte-identical (run A 592 B SHA 23336a627f1ff562a3d954e538f213070891790d69ef0262207d5fa7dc60a119; run B 238 B SHA f8c979d6ab879ea59e4cc97249c7d07b3dd8e9c79aeb7bbd52f1067e3c485b50). The canonical source now pins the fixture length and gates T16h byte identity. Staged job 1790174357, promoted live job 1790174780 and compatibility alias job 1790174894 each passed 45/45 with T16h and T17 GREEN. Source SHA 3001731f29587b8626b08d819c62e12fb243863c9da03aabd50f399afa3d4140; ELF SHA 9bfbd8fcbe8525f340c1247360627019c238e9fcb196db0f602065b300325131; retained receipt buildroot/knowledge/gates/motion-determinism-20260923.json; PM correction triaged as 1790175588. Test-gate and release-record work only; no visual or production-motion change.

Release history and work log

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

9 of 18 capability source symbols found|5 of them authored stronger-claims|9 open|coverage 500/1000|adoption 8 full / 1 partial

Evidence profile — what the gaps on this board actually are

Unstamped. No evidence profile is published for this domain yet: run nx_swcompare_evidence motion and this section fills itself on the next beat. No numbers are shown, deliberately — a zero-filled profile would read as a board with no gaps, which is the one wrong answer nobody would question.

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

RungCloses withDefinition of done (pre-declared)ExecutorEst.
The /motion door: a loopback daemon behind one route row (MO1)mo_servenx_motion_serve binds loopback, is registered in daemons.reg and deploy_targets.conf, sits behind one proxy_routes.conf row, answers GET with the emitted page and POST with a job, and never runs a script or a third-party runtime; done-rule: nx_page_verify reads the live page with the upload form present and no script tag, the daemon refuses an unknown path by name with a status, and the static fallback file is byte-identical to the empty-form responseOrgan1 u
Upload door with a named cap and content typing by magic (MO2)
after MO1		mo_upload_acceptA multipart POST is parsed without a library, the model part is typed by its first bytes (GLB magic, FBX magic, NXANIM01, else refused by name), the byte cap is a NAMED constant derived from the edge proxy's measured budget and announced on the page, and an accepted upload becomes a digit-named job directory the visitor is redirected to; done-rule: a GLB, an FBX and an NXA fixture each land as a job, an over-cap body and an OBJ are refused by name with 413 and 415, and a malformed boundary is refused rather than half-parsedOrgan0.5 u
The rig job at the ladder resolution, with its digest (MO3)
after MO2		mr_rignx_motion_rig rig forks nx_gltf2mesh or nx_fbx2nxa to reach a bare NXA, then nx_autorig_mesh at the visitor's cells, then nx_nxa_texbake view for the three-view render and nx_mesh2glb for the export, writing every composed report line and the sha256 of the rigged NXA into the job's status rows and a done marker last; done-rule: the creature gate's dragon fixture rigs to six limbs through this job, two runs of one input print one digest, a sheet is refused at the skeleton stage by name, and the job never holds the daemon (spawned detached, status readable while it runs)Organ1 u
Apply a library clip to the fresh rig (MO4)
after MO3,MO5		mr_animnx_motion_rig anim retargets the chosen donor's POSE entry onto the job's rig through nx_nxa_retarget (emit=pose), bakes it with nx_nxa_play at full weight, renders the posed rig and exports it beside the bind pose; done-rule: the salsa clip on the house rig plays with joints_posed equal to the mapped count, the limb-length tooth from nx_nxa_retarget_gate holds on the fresh rig, and a clip key absent from the rows is refused by nameOrgan1 u
Clip library as data rows (MO5)
after MO1		mo_clip_listknowledge/motion_clips.rows carries key, donor path, pose id, title, licence class and provenance per clip, read by the page and the anim verb; done-rule: adding a row adds a clip to the page and the verb with no rebuild, a row whose donor is absent renders as ABSENT and refuses to apply, and every row carries a licence class the page printsOrgan0.5 u
The page emitted from data with a byte-identical static fallback (MO6)
after MO1		mo_body_indexOne renderer reads knowledge/motion_page.tmpl, the clip rows and the demo rows, serves the page live and emits the same bytes to sites/nishifamily/motion/index.html; done-rule: the two are byte-identical for the empty form, the page passes nx_page_verify, and no HTML for this surface exists that the renderer did not writeOrgan0.5 u
The referee grade on every demo rig (MO7)
after MO6		mo_demo_cardsEach demo card on the page carries the dcc referee's numbers for that rig (coverage, joint error, top-1 agreement, cells, method) read from a rows file the referee writes, and each uploaded job carries the composed organs' counts or the word UNGRADED; done-rule: the house rig's card prints the values in dcc.plan, and a demo whose referee row is absent prints UNGRADED rather than a blankOrgan0.5 u
Operator verdict per upload to the acceptance plane (MO8)
after MO3		mo_verdict_rowAn ACCEPT or REJECT form on the job page writes id, subject, verdict, date, actor, words and targets to knowledge/store/accept- through the store organ, and the job page shows the latest verdict; done-rule: a REJECT then an ACCEPT on one job yield two rows with the later one shown, and a verdict on an absent job is refused by nameOrgan0.5 u
Ingest the world's clip libraries into POSE (MO9)
after MO5		ci_gltf_animglTF animation channels (rotation and translation samplers per node) and BVH frames read into NXA POSE library entries on the donor's own skeleton, so the CC0 Mesh2Motion and Quaternius clips and any BVH become donors for MO4 under a provenance row; done-rule: one Mesh2Motion clip round-trips node count and key count exactly, plays through nx_nxa_play with joints_posed equal to the animated node count, retargets onto the house rig with the limb-length tooth holding, and carries its CC0 row from the mirrored licenceOrgan2 u
The rival rigger as an oracle container (MO10)
after MO7		mo_oracle_skintokensSkinTokens and UniRig run inside an oracle image the estate starts and stops through nx_container, fed the same upload, their skeleton and weights read back into NXA and graded by the dcc referee on the same page; nothing of theirs enters the build or run path; done-rule: one upload yields two rigs on one job page with two referee rows, and the oracle's absence prints UNAVAILABLE rather than a blankOrgan2 u
The shared benchmark, run whole (MO11)
after MO10		mo_bench_corpusA licence-clean artist-rigged corpus (CC0 sets and the operator's own rigged donors) the referee runs over every model, never a sample, publishing coverage, joint error and weight agreement for ours and for each oracle on this board; done-rule: the partition of the corpus into graded, refused and unavailable sums to its size, the numbers land as rows the generator renders, and the meet-or-exceed sentence on this page is derived from themOrgan2 u
A global-bar arm for the referee ladder (MO12)mo_bar_modenx_curveskel_lib exposes the significance bar mode as a conf row so the 192-cell gain on the house rig can be attributed between resolution and the local-significance change with the same binary; done-rule: the two arms print two numbers on dcc.plan and only the difference is credited to the changeOrgan0.5 u
Chunked door for models past the edge budget (MO13)
after MO2		mo_upload_chunkedUploads larger than the named cap go through nx_content_put's chunked sha-receipted door into the same job directory, the page saying which door applies at which size; done-rule: toon3d8 (61,939,208 B) lands as a job through the chunked door with its sha matching the client receiptOrgan1 u

Milestones

MilestoneRungsCumulative
M0 · An upload rigged on the pageMO1,MO2,MO3,MO63 u
M1 · Motion on the rig and the operator's verdictMO4,MO5,MO7,MO85.5 u
M2 · The world's clips and the rivals as oraclesMO9,MO10,MO12,MO1310.5 u
M3 · The shared benchmarkMO1112.5 u
RungCloses withDefinition of done (pre-declared)ExecutorEst.
A bind-smoke tooth so the daemon's listen is gated, not gambled (MO14)
after MO1		mo_bind_smokenx_motion_gate forks the promoted nx_motion_serve.elf on a port nx_port_audit proves free, waits for the REFUSED->ACCEPT edge, GETs /motion for a 200 carrying the upload form, then kills it and asserts the port is free again; done-rule: the tooth is RED against a deliberately-occupied port and against a daemon that exits before accept, GREEN on the real binary, and its captured stderr names LISTEN-FAILED vs listening so a NEVER-LISTENED deploy is diagnosable from the gate, never from a lost spawn logOrgan1 u
Ladder verdict. NOT DECLARED. This board names no dated best-in-class or frontier target and no rung roles (sotatarget and rungrole rows on its plan); the ranker labels it NO-LADDER until it does, and until then its rungs climb toward a target nobody has written down. Bars read STALE against the current month 2026-10.
Bar freshness, counted
ClassCountNote
bars7sotabar rows on the plan
fresh0the bar month is the current month
attested0an older bar with a this-month barscan reporting no newer entry
stale0a this-month scan saw a newer month the bar ignores
unattested7nobody looked this month
malformed0a sotabar row missing a field
duplicate0bar ids an earlier row already carries
bars, summed7RECONCILED: the parts sum to the total 7
current month2026-10the 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 14. 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 49d375e0a05624ec39f78cee46f558e1cd1bc63de6f87db766c629b39211ac82. Target rows 0; role rows 0. Existing risks and release worklog retain their own scope; no node completion is inferred.

Runtime observation coverage: 0 of 14 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.

  1. MO1 — The /motion door: a loopback daemon behind one route row

    Prerequisites: None declared; this does not establish execution eligibility.

    Contract: mo_serve

    Acceptance: nx_motion_serve binds loopback, is registered in daemons.reg and deploy_targets.conf, sits behind one proxy_routes.conf row, answers GET with the emitted page and POST with a job, and never runs a script or a third-party runtime; done-rule: nx_page_verify reads the live page with the upload form present and no script tag, the daemon refuses an unknown path by name with a status, and the static fallback file is byte-identical to the empty-form response

    Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

    Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.

  2. MO2 — Upload door with a named cap and content typing by magic

    Prerequisites: MO1 (acceptance unverified)

    Contract: mo_upload_accept

    Acceptance: A multipart POST is parsed without a library, the model part is typed by its first bytes (GLB magic, FBX magic, NXANIM01, else refused by name), the byte cap is a NAMED constant derived from the edge proxy's measured budget and announced on the page, and an accepted upload becomes a digit-named job directory the visitor is redirected to; done-rule: a GLB, an FBX and an NXA fixture each land as a job, an over-cap body and an OBJ are refused by name with 413 and 415, and a malformed boundary is refused rather than half-parsed

    Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

    Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.

  3. MO3 — The rig job at the ladder resolution, with its digest

    Prerequisites: MO2 (acceptance unverified)

    Contract: mr_rig

    Acceptance: nx_motion_rig rig forks nx_gltf2mesh or nx_fbx2nxa to reach a bare NXA, then nx_autorig_mesh at the visitor's cells, then nx_nxa_texbake view for the three-view render and nx_mesh2glb for the export, writing every composed report line and the sha256 of the rigged NXA into the job's status rows and a done marker last; done-rule: the creature gate's dragon fixture rigs to six limbs through this job, two runs of one input print one digest, a sheet is refused at the skeleton stage by name, and the job never holds the daemon (spawned detached, status readable while it runs)

    Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

    Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.

  4. MO4 — Apply a library clip to the fresh rig

    Prerequisites: MO3 (acceptance unverified), MO5 (acceptance unverified)

    Contract: mr_anim

    Acceptance: nx_motion_rig anim retargets the chosen donor's POSE entry onto the job's rig through nx_nxa_retarget (emit=pose), bakes it with nx_nxa_play at full weight, renders the posed rig and exports it beside the bind pose; done-rule: the salsa clip on the house rig plays with joints_posed equal to the mapped count, the limb-length tooth from nx_nxa_retarget_gate holds on the fresh rig, and a clip key absent from the rows is refused by name

    Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

    Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.

  5. MO5 — Clip library as data rows

    Prerequisites: MO1 (acceptance unverified)

    Contract: mo_clip_list

    Acceptance: knowledge/motion_clips.rows carries key, donor path, pose id, title, licence class and provenance per clip, read by the page and the anim verb; done-rule: adding a row adds a clip to the page and the verb with no rebuild, a row whose donor is absent renders as ABSENT and refuses to apply, and every row carries a licence class the page prints

    Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

    Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.

  6. MO6 — The page emitted from data with a byte-identical static fallback

    Prerequisites: MO1 (acceptance unverified)

    Contract: mo_body_index

    Acceptance: One renderer reads knowledge/motion_page.tmpl, the clip rows and the demo rows, serves the page live and emits the same bytes to sites/nishifamily/motion/index.html; done-rule: the two are byte-identical for the empty form, the page passes nx_page_verify, and no HTML for this surface exists that the renderer did not write

    Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

    Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.

  7. MO7 — The referee grade on every demo rig

    Prerequisites: MO6 (acceptance unverified)

    Contract: mo_demo_cards

    Acceptance: Each demo card on the page carries the dcc referee's numbers for that rig (coverage, joint error, top-1 agreement, cells, method) read from a rows file the referee writes, and each uploaded job carries the composed organs' counts or the word UNGRADED; done-rule: the house rig's card prints the values in dcc.plan, and a demo whose referee row is absent prints UNGRADED rather than a blank

    Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

    Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.

  8. MO8 — Operator verdict per upload to the acceptance plane

    Prerequisites: MO3 (acceptance unverified)

    Contract: mo_verdict_row

    Acceptance: An ACCEPT or REJECT form on the job page writes id, subject, verdict, date, actor, words and targets to knowledge/store/accept- through the store organ, and the job page shows the latest verdict; done-rule: a REJECT then an ACCEPT on one job yield two rows with the later one shown, and a verdict on an absent job is refused by name

    Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

    Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.

  9. MO9 — Ingest the world's clip libraries into POSE

    Prerequisites: MO5 (acceptance unverified)

    Contract: ci_gltf_anim

    Acceptance: glTF animation channels (rotation and translation samplers per node) and BVH frames read into NXA POSE library entries on the donor's own skeleton, so the CC0 Mesh2Motion and Quaternius clips and any BVH become donors for MO4 under a provenance row; done-rule: one Mesh2Motion clip round-trips node count and key count exactly, plays through nx_nxa_play with joints_posed equal to the animated node count, retargets onto the house rig with the limb-length tooth holding, and carries its CC0 row from the mirrored licence

    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.

  10. MO10 — The rival rigger as an oracle container

    Prerequisites: MO7 (acceptance unverified)

    Contract: mo_oracle_skintokens

    Acceptance: SkinTokens and UniRig run inside an oracle image the estate starts and stops through nx_container, fed the same upload, their skeleton and weights read back into NXA and graded by the dcc referee on the same page; nothing of theirs enters the build or run path; done-rule: one upload yields two rigs on one job page with two referee rows, and the oracle's absence prints UNAVAILABLE rather than a blank

    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.

  11. MO11 — The shared benchmark, run whole

    Prerequisites: MO10 (acceptance unverified)

    Contract: mo_bench_corpus

    Acceptance: A licence-clean artist-rigged corpus (CC0 sets and the operator's own rigged donors) the referee runs over every model, never a sample, publishing coverage, joint error and weight agreement for ours and for each oracle on this board; done-rule: the partition of the corpus into graded, refused and unavailable sums to its size, the numbers land as rows the generator renders, and the meet-or-exceed sentence on this page is derived from them

    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.

  12. MO12 — A global-bar arm for the referee ladder

    Prerequisites: None declared; this does not establish execution eligibility.

    Contract: mo_bar_mode

    Acceptance: nx_curveskel_lib exposes the significance bar mode as a conf row so the 192-cell gain on the house rig can be attributed between resolution and the local-significance change with the same binary; done-rule: the two arms print two numbers on dcc.plan and only the difference is credited to the change

    Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

    Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.

  13. MO13 — Chunked door for models past the edge budget

    Prerequisites: MO2 (acceptance unverified)

    Contract: mo_upload_chunked

    Acceptance: Uploads larger than the named cap go through nx_content_put's chunked sha-receipted door into the same job directory, the page saying which door applies at which size; done-rule: toon3d8 (61,939,208 B) lands as a job through the chunked door with its sha matching the client receipt

    Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

    Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.

  14. MO14 — A bind-smoke tooth so the daemon's listen is gated, not gambled

    Prerequisites: MO1 (acceptance unverified)

    Contract: mo_bind_smoke

    Acceptance: nx_motion_gate forks the promoted nx_motion_serve.elf on a port nx_port_audit proves free, waits for the REFUSED->ACCEPT edge, GETs /motion for a 200 carrying the upload form, then kills it and asserts the port is free again; done-rule: the tooth is RED against a deliberately-occupied port and against a daemon that exits before accept, GREEN on the real binary, and its captured stderr names LISTEN-FAILED vs listening so a NEVER-LISTENED deploy is diagnosable from the gate, never from a lost spawn log

    Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

    Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.

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 motion 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.
MO1 The /motion door: a loopback daemon behind one route row · cell WATCHING · gate NO-GATE-DECLARED
Contractmo_serve · executor Organ · 1 u · stands on nothing
Definition of donenx_motion_serve binds loopback, is registered in daemons.reg and deploy_targets.conf, sits behind one proxy_routes.conf row, answers GET with the emitted page and POST with a job, and never runs a script or a third-party runtime; done-rule: nx_page_verify reads the live page with the upload form present and no script tag, the daemon refuses an unknown path by name with a status, and the static fallback file is byte-identical to the empty-form response
Matrix cellSURFACE: the /motion door, a loopback daemon behind one route row · runtime/nx_motion_serve.nx · WATCHING decl
GateNO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_motion_serve; declare nx_motion_serve TAB <gate> and the ship loop proves it and the verdict lands here.
Work log2026-09-06 measure DAEMON NOT SERVED, witnessed 2026-09-06: nx_services lists 16 services UP (sites 8443, wiki_gw, gallery gateway and serve, media auth, redirect, torrent_gw, clock, mgmt_api 18098, docportal 18456, email 18465, siteedit 18466, inventory 18122, vroom 8446, tools_api 18096, opaque_login 9091) and nx_motion_serve is NOT among them, so the signed daemons.reg revive row did not bring it up and MO1 stays open by measurement, not by caution. The binary is PROMOTED at the serving root (92,397 B) and nx_motion_smoke proved it binds; the two /api/deploy attempts rolled back NEVER-LISTENED (the level-probe defect this estate already records). NEXT EXPERIMENT, not a cause: read the supervisor's own stdout for HELD-TAMPER (a daemons.reg whose signature does not verify makes the supervisor hold its last-good roster, which would not carry the new row) versus a roster read only at startup; whichever it names, the remedy is one re-sign or one supervisor restart, never a hand-launched process on a shared port. A cap was minted for nx_motion_rig the same round (30 days), closing the REGISTERED-UNAUTHORISED residual the ranker named on MO3 and MO4.
2026-09-06 measure THE SUPERVISOR-LOG EXPERIMENT RAN 2026-09-06: supervisor.log (2,862,652 B at the nishihost root, scanned whole) carries NO line naming nx_motion_serve and NO TAMPER line, so the HELD-TAMPER arm is refuted (a rejected signature would have been logged) and the remaining arm stands: the supervisor read daemons.reg at its own start and has not re-read it since, so a row appended afterwards is invisible to it until it restarts or is told to reload. The box reads DEGRADED (mem-pressure, swap 952 permil), so neither a supervisor restart nor a third /api/deploy attempt is taken now: a slow start under memory pressure would read NEVER-LISTENED again and teach nothing. RECIPE when health reads OK: /api/deploy target=nx_motion_serve (never-brick, auto-rollback) and, if it rolls back NEVER-LISTENED against a binary nx_motion_smoke proves binds, the supervisor reload is the remedy, never a hand-launched process on a shared port.
2026-09-06 measure BOARD OPENED 2026-09-06 from the operator's ask. Measured that day: the dcc referee on the house rig reads 461 permil coverage and 208.9 mm at 96 cells and 923 permil and 40.1 mm at 192 cells on the new libs (unattributed between resolution and local significance until MO12), top-1 weight agreement 233 permil; the world's viewer pages have no emitter (nx_absent ABSENT-PROVEN over every organ source); the /mesh route has no daemon behind it and its source is the gallery gateway; nx_autorig_mesh, nx_gltf2mesh, nx_fbx2nxa, nx_nxa_retarget, nx_nxa_play, nx_nxa_texbake and nx_mesh2glb are all LIVE in the catalogue; SkinTokens' own README states 98 to 133 percent skinning and 17 to 22 percent bone gains over prior SOTA; Tripo's page states auto-rig for humanoids in seconds; Mesh2Motion's page states four import formats, six archetype rigs and over 150 CC0 clips.
2026-09-06 land BOARD ADMITTED 2026-09-06 (nx_domain_admit motion: VERDICT=ADMIT, CLAIM-ONLY, rows=14 grounded=3 watch-new-organ=10, bars declared unpinned) and the surface's sources WRITTEN, NOT YET BUILT OR DEPLOYED: buildroot/runtime/nx_motion_lib.nx 41,587 B (typing by magic, multipart, form decoding, status rows, row files, the fork runner, both page renderers), nx_motion_rig.nx 14,931 B (rig and anim verbs over the promoted organs, status rows at every stage, done marker last), nx_motion_serve.nx 21,464 B (loopback 18132, plain forms, detached spawn, art and demo serving, the verdict row to the journal and the acceptance plane), nx_motion_serve_gate.nx 17,221 B (rulers on planted inputs, the box fixture rigged twice for a determinism tooth, an OBJ refusal control); data knowledge/motion_page.tmpl, knowledge/motion_clips.rows (house_pose0: ref9d.nxa pose 0, 104 joints posed, max displacement 549 mm -- the only real pose on the house assets, ref9d_salsa6 carries the identical frame), knowledge/motion_demos.rows (ref9d_192 with the referee's own numbers), demo assets knowledge/motion/demos/ref9d_192.nxa (10,758,232 B, the 192-cell rig banked from the referee run) and ref9d_192.png (TEXVIEW-OK nv=14164). First compile found two let-reassignments in the lib, fixed by CAS edit; the recompile is in flight. Still owed before the door opens: green gate, promote and register the three organs, deploy_targets.conf and daemons.reg rows, the proxy_routes.conf row, nx_page_verify on the live page.
2026-09-06 measure nx_motion_gate v2 (95,011 B, sha256 005cfbdf10ee5adbbd800d9f72c7a083dc3256d4ebda7962e89db3a021ee7d19) staged and run as nx_motion_gate_staged: 46 of 46 GREEN. The two fixture-reached-the-condition teeth added after the first run (the box read back with the bytes the writer reported; the second run rigged with exit 0) both pass, and T17b now proves determinism honestly: the same box rigged twice through the promoted job organ prints one 64-hex digest. Emitted values: box_limbs 2, box_joints 12, box_mixed_verts 8, rig.glb 3,124 B, rig_view.png 331,814 B, 1,738 ms, upload cap 8,388,608 B. The static fallback was emitted by the staged daemon's own emit verb (MOTION-EMIT OK bytes=5602 sites/nishifamily/motion/index.html, sha 4a684627): the same renderer the live page will serve. Promote of the gate and the health-checked deploy of the daemon are in flight; the supervisor row, the route row and nx_page_verify on the live page remain.
2026-09-06 land SHIPPED AND GATE-PROVEN, PUBLIC DOOR BLOCKED, LIVE UNTOUCHED. nx_motion_gate PROMOTED+REGISTERED+_offc (95,011 B 005cfbdf, 46/46 GREEN incl. the real-mesh determinism tooth), declared oracle; nx_motion_rig v2 PROMOTED+REGISTERED+_offc (54,168 B bf4d6fd7), declared oneshot; the static fallback emitted by the daemon's own emit verb (sites/nishifamily/motion/index.html 5,602 B 4a684627); deploy_targets.conf and daemons.reg rows added and daemons.reg re-signed. BLOCKED: /api/deploy nx_motion_serve ROLLED-BACK NEVER-LISTENED twice, so the serving root has NO nx_motion_serve.elf (catalogue BUILT-UNPROMOTED) -- the never-brick rollback fired correctly and LIVE is untouched. The startup cause is UNCAPTURED: the sanctioned launcher for a daemon is /api/deploy, no deploy spawn-log is exposed under knowledge/status or /tmp, and a daemon cannot be nx_job_run (jlb_denied) to capture its own stderr, so I cannot read the LISTEN-FAILED/listening line. Two hypotheses, neither hammered further under the active I/O storm (procs_blocked 5-6): (a) another listener already holds loopback 18132 -> EADDRINUSE at bind (nx_http_probe returned 301 on BOTH 18132 and 18133, which means it is answered by the host redirector and is NOT a valid bind test either way); (b) a startup crash. THE PERMANENT FIX, rung MO14: a bind-smoke TOOTH in nx_motion_gate that forks the daemon binary on a throwaway port proven free by nx_port_audit, connects for a 200 on /motion, and kills it -- turning 'does the daemon listen' from a deploy gamble into a gated repeatable check, and capturing exactly the stderr the deploy hides. Until MO14 or a calm window: the daemon source and staged .elf.new (92,397 B 18205859) stand ready; one clean /api/deploy off-storm, or a port off tier_ladder.conf proven free, opens the door. Everything the surface needs is built and proven; only the listen remains.
MO2 Upload door with a named cap and content typing by magic · cell WATCHING · gate NO-GATE-DECLARED
Contractmo_upload_accept · executor Organ · 0.5 u · stands on MO1
Definition of doneA multipart POST is parsed without a library, the model part is typed by its first bytes (GLB magic, FBX magic, NXANIM01, else refused by name), the byte cap is a NAMED constant derived from the edge proxy's measured budget and announced on the page, and an accepted upload becomes a digit-named job directory the visitor is redirected to; done-rule: a GLB, an FBX and an NXA fixture each land as a job, an over-cap body and an OBJ are refused by name with 413 and 415, and a malformed boundary is refused rather than half-parsed
Matrix cellSURFACE: upload a model in the browser and get it rigged · runtime/nx_motion_serve.nx · WATCHING decl
GateNO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_motion_serve; declare nx_motion_serve TAB <gate> and the ship loop proves it and the verdict lands here.
Work logno log row names this rung yet
MO3 The rig job at the ladder resolution, with its digest · cell PRESENT · gate NO-GATE-DECLARED
Contractmr_rig · executor Organ · 1 u · stands on MO2
Definition of donenx_motion_rig rig forks nx_gltf2mesh or nx_fbx2nxa to reach a bare NXA, then nx_autorig_mesh at the visitor's cells, then nx_nxa_texbake view for the three-view render and nx_mesh2glb for the export, writing every composed report line and the sha256 of the rigged NXA into the job's status rows and a done marker last; done-rule: the creature gate's dragon fixture rigs to six limbs through this job, two runs of one input print one digest, a sheet is refused at the skeleton stage by name, and the job never holds the daemon (spawned detached, status readable while it runs)
Matrix cellRIG: skeleton and weights from the volume with no archetype hint · runtime/nx_motion_rig.nx · PRESENT decl
GateNO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_motion_rig; declare nx_motion_rig TAB <gate> and the ship loop proves it and the verdict lands here.
Work log2026-09-06 lesson THE CURRENT BAR VALIDATES THE CLASS AND NAMES THE GAP. The August 2026 skinning paper argues in its own abstract that data-driven weight generation lacks the generalisability of classic geometric approaches and proposes a robustly computable geometric prior to bridge it [arxiv_autorig_2026_09]; our geodesic bone-heat weights are that prior class, computed sovereignly with no training set, proven at p95 6.3 mm deformation against the artist rig on the house mesh (dcc.plan). The September paper names inaccurate skinning weights on fused AI-generated geometry as the open failure, which is exactly the witch result (coverage 396 permil, extremities under-driven). So the ladder to the current bar is prior PLUS learned refinement (modding MD4) graded by the same referee, and a seat that tunes the falloff instead is climbing the wrong rung.
2026-09-06 land SCOREBOARD TRUTH 2026-09-06: nx_compare_rank read motion 0 of 14 done while nx_motion_rig (PROMOTED bf4d6fd7), nx_motion_lib and nx_motion_gate (46 of 46, PROMOTED) had shipped, because six matrix watch rows and five rung contracts named symbols that were never built under those names -- the DC16 lesson again, a rung symbol must name a symbol that exists. Repointed BOTH the rung and its matrix row to the real entry point read from the organ: MO3 mr_rig, MO4 mr_anim, MO5 mo_clip_list, MO6 mo_body_index, MO7 mo_demo_cards, plus the digest row mr_sha_from_log; the three renderer rows now point at runtime/nx_motion_lib.nx where the symbols live. DELIBERATELY NOT REPOINTED: mo_serve, mo_upload_accept and mo_verdict_row, because their done-rules need the daemon LIVE behind a route row and nx_motion_serve, though PROMOTED at the serving root (92,397 B), is not yet witnessed serving 18132 (deploy rolled back NEVER-LISTENED, supervisor row signed, route row and nx_page_verify owed) -- a flip is a receipt and there is no live surface to receipt yet. Every edit landed under a size CAS; a thirteen-edit batch proved the async lane runs a batch CONCURRENTLY, so only the rung whose expect matched the original size landed and the rest were refused by name -- one write per file per round from then on.
2026-09-06 land THE PERMANENT FIX FOR DARK CAPABILITY, source-landed 2026-09-06 (operator: this needs a permanent fix, we do not want to build and lose capabilities to darkness). The one shared ruler already told organ-unreadable (-1) from organ-read-and-symbol-absent (0); every consumer folded both into OPEN. Now: nx_swcompare_matrix records rdark per watch row (0 under a rule that can judge, the JS abstention excluded through the applied rule), renders the cell as data-s dk with its own evidence prose, a Dark filter chip, a method sentence and dark_contracts= in both its receipts (BUILT by the governor 455,816 B sha 708d15d7, promoted with the one superseded prose run named); nx_compare_regen writes state DARK to the comparewatch plane with the partition OPEN plus LANDED plus DARK equal to stored (PROMOTED 187,616 B sha 42fabc16, contentdiff ONE-WAY-GAIN, _offc refreshed); nx_compare_rank carries rdark_r per rung and prints DARK-CONTRACT on the rung line (build submitted); nx_swcompare_watch_gate T30, T30a, T30b and T30c pin the predicate on the ruler with a missing-organ and a JS-abstain neg-control (build queued). ALSO: every dated bar on this board and on dcc was moved to the sotabar kind the sibling seat's freshness contract reads (bar is the renderer's research-bar grammar and would have rendered these rows garbled), so the served ranker reads NO-BAR until the rebuilt one reads FRESH; the sotatarget and rungrole ladder rows are owed on both boards. The hardened referee nx_rigbench is PROMOTED at 125,528 B sha d441bc19 with the normalize pass (md_scale_div= printed) and its _offc twin refreshed; nx_rigbench_gate carries T12 and is queued as a sovereign build.
2026-09-06 land nx_motion_rig PROMOTED 2026-09-06 (45,559 B, sha256 7b8b6a30caf0f66165cd9eea8f107d7e8c4d8aad20efdfda3424726d65adf92e, prev=1), REGISTERED on the MCP surface and installed to _offc (OFFC-INSTALL-PROMOTED installed=1). This is the job behind the door: rig <jobdir> [cells] converts through nx_gltf2mesh or nx_fbx2nxa, rigs through the promoted nx_autorig_mesh, renders through nx_nxa_texbake view, exports through nx_mesh2glb, hashes through nx_filehash, and writes status rows at every stage with done last; anim <jobdir> <donor> <pose> <key> retargets and bakes a clip. NOT YET PROVEN BY ITS GATE: nx_motion_serve_gate is written and its build is queued behind admission (run queue 14 of 8); the daemon build was refused the same minute and is paced. The presence of the binary is a receipt, not a proof.
2026-09-06 measure FIRST REAL JOB through the promoted nx_motion_rig (house mesh, 1,014,256 B bare NXA, 96 cells, 2026-09-06): rig OK in 16,928 ms -- limbs 7, joints 111, branches 6, verts 14164, tris 28092, GLB exported -- and two stages refused BY NAME, each a composition fact the first run measured: nx_nxa_texbake view exit 4 (TEXBAKE-REFUSE no TEXC section, run nx_nxa_texc first) and nx_filehash exit 127 (the elf lives at the serving root, not in _offc). Fixed at the roots in the job organ (an elf resolver that tries _offc then the root; a UV-unwrap stage before the render) -- rebuilt 54,168 B sha bf4d6fd7, staged, promote in flight. The gate was renamed nx_motion_gate (it proves the door AND the job, and the async launcher's deny-by-class bound refuses any name segment starting with 'serve' -- debt 1787686197, a THIRD measured recurrence after nx_deployjrnl; the correction row 1787707354 already records that the guard is correct and the declaration file is knowledge/status/organ_kind.conf, where all three organs are now declared). nx_motion_serve_gate.nx could not be retired -- nx_retire_path protects any name carrying 'serve' (same class) -- so it stands superseded beside nx_motion_gate.nx; the rebuild under the new name is owed.
2026-09-06 land nx_motion_rig v2 PROMOTED 2026-09-06 (54,168 B, sha256 bf4d6fd72809f31a72f30a0b76bf1e5564b2641b46acefbf6974a70797566f2c) over v1 7b8b6a30. The promote guard first refused CAPLOSS 76 permil (10 of 131 live tokens absent); nx_contentdiff enumerated every lost run and all ten were the eight _offc/nx_<organ>.elf path literals the new elf resolver replaced (lost_srcpath=0 lost_symbolname=0 lost_other=10, gains rig_tex.nxa, log_texc.txt, _tex.nxa), so the loss was intended and named and the promote was re-sent with allow_capability_loss=yes. No matched-vintage control was needed: the cause was read from the diff, not inferred. _offc twin re-installed from the promoted root.
2026-09-06 measure nx_motion_gate (staged 94,756 B decb5cb9, run as nx_motion_gate_staged through nx_job_run, 2026-09-06): 43 of 44. Every ruler tooth passed on planted inputs -- typing by magic incl. the PNG negative control, the multipart parser incl. the cut-body refusal, form and query decoding incl. the truncated-escape refusal, path shape incl. traversal refusals, line-anchored status rows, the report-line fields, both renderers (no script tag, UNGRADED and ABSENT rendered, the cap printed from the one constant) -- and the box fixture rigged end to end through the promoted nx_motion_rig v2 in 1,658 ms: limbs 2, joints 12, mixed_verts 8, rig.glb 3,124 B, rig_view.png 331,814 B (the UV-unwrap-then-render stage now works), a 64-hex digest, and the OBJ control refused at stage convert with 415 and done written. The one FAIL was the determinism tooth, and the cause was the GATE's fixture, not the pipeline: the length pointer was re-read after run a's status read had overwritten it, so run b's box was written truncated and refused at the skeleton stage (its log: no valid VERT section). Fixed by pinning the fixture length once with a fixture-reached-the-condition tooth beside it, and T17 split so a second run that never rigged cannot be read as a digest mismatch. Rebuild under the new name and the re-run are next; the real-mesh determinism check (the house mesh rigged twice into two job directories) is running beside it.
2026-09-06 measure DETERMINISM ON THE REAL MESH, 2026-09-06: the house rig's bare mesh (1,014,256 B, 14,164 verts, 28,092 tris) rigged twice through the promoted nx_motion_rig v2 into two different job directories (1788706300 and 1788706301) at 96 cells printed ONE rig digest, e5fa2f5497c07a351fe8a981f8797e005600897b37c994b03c8f860e153fc601, with identical counts (limbs 7, joints 111, branches 6, mixed_verts 14164); wall time 26,089 ms cold and 4,189 ms warm, so the cost was the box's page cache, not the arithmetic. This is the mo_rig_digest row's claim met by measurement on a real character, beside the gate's box.
2026-09-23 land DETERMINISM TEST REPAIR ACCEPTED 2026-09-23: the prior RED 43/44 did not prove nondeterminism because the repeated fixtures were not byte-identical (run A 592 B SHA 23336a627f1ff562a3d954e538f213070891790d69ef0262207d5fa7dc60a119; run B 238 B SHA f8c979d6ab879ea59e4cc97249c7d07b3dd8e9c79aeb7bbd52f1067e3c485b50). The canonical source now pins the fixture length and gates T16h byte identity. Staged job 1790174357, promoted live job 1790174780 and compatibility alias job 1790174894 each passed 45/45 with T16h and T17 GREEN. Source SHA 3001731f29587b8626b08d819c62e12fb243863c9da03aabd50f399afa3d4140; ELF SHA 9bfbd8fcbe8525f340c1247360627019c238e9fcb196db0f602065b300325131; retained receipt buildroot/knowledge/gates/motion-determinism-20260923.json; PM correction triaged as 1790175588. Test-gate and release-record work only; no visual or production-motion change.
2026-09-23 measure TORRENT GATEWAY REQUEST-LIFETIME REPAIR QUALIFIED 2026-09-23 in the canonical owner (not installed or promoted): source buildroot/runtime/_hdl_build/nx_torrent_gw.nx 14,246 B sha256 25d4f820068ec70812bd70ff27c02b5f56fd84638ac1ddb037ad482e3f5c5184; shared dependency buildroot/runtime/nx_sha512.nx 19,081 B sha256 bc430061bf7f335f6c0d262520fd6586ecc8d1e834da9f3a3bccb0fa31242fd7. Root cause was per-request arena scope plus Ed25519 and SHA-512 schedule mappings that survived request return; canonical owner now closes request scope/arena, wraps Ed25519 scratch and places sha512_compress W[0..79] in a nested scratch frame. Candidate build job 1790177568 produced buildroot/_build/nx_torrent_gw.sov.elf 167,256 B sha256 c595f401a7128ae351cdfa013daf14e2617bacc2047f85387962d192d6c66f70. Isolated gate job 1790177585 passed 15/15 over 30 repeated mixed requests with VmSize/RSS delta 0; extended job 1790177609 passed 15/15 over 100 repeated mixed requests with VmSize 1,808->1,808 KiB and VmRSS 1,500->1,500 KiB, both deltas 0. Gate source 9a639c5859bc1e32e7775815f19fd299620924dc58697bad4dd93e2356dc6a6f and gate ELF a28a32aa276bc5cbb3a87a7205ae7373b6c7233f6e82890d133f86518361c0e6. Prior RED job 1790177354 was 13/15 at 10 rounds and 360 KiB growth; this causal delta is the accepted evidence. Live PID 21376 was not restarted; publisher/registration remains a separate owner action after source review.
2026-09-23 reclassify The preceding 1790177609 torrent lifetime row is retained as historical cross-reference only; torrent gateway/resource evidence belongs canonically in buildroot/knowledge/compare/loadgov.plan. Motion owns only the dependency cross-reference: nx_request_scope lifetime contract is consumed by nx_torrent_gw. No history was deleted.
MO4 Apply a library clip to the fresh rig · cell PRESENT · gate NO-GATE-DECLARED
Contractmr_anim · executor Organ · 1 u · stands on MO3, MO5
Definition of donenx_motion_rig anim retargets the chosen donor's POSE entry onto the job's rig through nx_nxa_retarget (emit=pose), bakes it with nx_nxa_play at full weight, renders the posed rig and exports it beside the bind pose; done-rule: the salsa clip on the house rig plays with joints_posed equal to the mapped count, the limb-length tooth from nx_nxa_retarget_gate holds on the fresh rig, and a clip key absent from the rows is refused by name
Matrix cellMOTION: apply a library clip to the fresh rig · runtime/nx_motion_rig.nx · PRESENT decl
GateNO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_motion_rig; declare nx_motion_rig TAB <gate> and the ship loop proves it and the verdict lands here.
Work logno log row names this rung yet
MO5 Clip library as data rows · cell PRESENT · gate NO-GATE-DECLARED
Contractmo_clip_list · executor Organ · 0.5 u · stands on MO1
Definition of doneknowledge/motion_clips.rows carries key, donor path, pose id, title, licence class and provenance per clip, read by the page and the anim verb; done-rule: adding a row adds a clip to the page and the verb with no rebuild, a row whose donor is absent renders as ABSENT and refuses to apply, and every row carries a licence class the page prints
Matrix cellMOTION: clip library as data rows · runtime/nx_motion_lib.nx · PRESENT decl
GateNO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_motion_lib; declare nx_motion_lib TAB <gate> and the ship loop proves it and the verdict lands here.
Work logno log row names this rung yet
MO6 The page emitted from data with a byte-identical static fallback · cell PRESENT · gate NO-GATE-DECLARED
Contractmo_body_index · executor Organ · 0.5 u · stands on MO1
Definition of doneOne renderer reads knowledge/motion_page.tmpl, the clip rows and the demo rows, serves the page live and emits the same bytes to sites/nishifamily/motion/index.html; done-rule: the two are byte-identical for the empty form, the page passes nx_page_verify, and no HTML for this surface exists that the renderer did not write
Matrix cellPAGE: emitted from data, never hand-written, static fallback byte-identical · runtime/nx_motion_lib.nx · PRESENT decl
GateNO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_motion_lib; declare nx_motion_lib TAB <gate> and the ship loop proves it and the verdict lands here.
Work logno log row names this rung yet
MO7 The referee grade on every demo rig · cell PRESENT · gate NO-GATE-DECLARED
Contractmo_demo_cards · executor Organ · 0.5 u · stands on MO6
Definition of doneEach demo card on the page carries the dcc referee's numbers for that rig (coverage, joint error, top-1 agreement, cells, method) read from a rows file the referee writes, and each uploaded job carries the composed organs' counts or the word UNGRADED; done-rule: the house rig's card prints the values in dcc.plan, and a demo whose referee row is absent prints UNGRADED rather than a blank
Matrix cellREFEREE: the rig grade travels with the surface · runtime/nx_motion_lib.nx · PRESENT decl
GateNO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_motion_lib; declare nx_motion_lib TAB <gate> and the ship loop proves it and the verdict lands here.
Work logno log row names this rung yet
MO8 Operator verdict per upload to the acceptance plane · cell WATCHING · gate NO-GATE-DECLARED
Contractmo_verdict_row · executor Organ · 0.5 u · stands on MO3
Definition of doneAn ACCEPT or REJECT form on the job page writes id, subject, verdict, date, actor, words and targets to knowledge/store/accept- through the store organ, and the job page shows the latest verdict; done-rule: a REJECT then an ACCEPT on one job yield two rows with the later one shown, and a verdict on an absent job is refused by name
Matrix cellACCEPT: operator verdict per upload as data · runtime/nx_motion_serve.nx · WATCHING decl
GateNO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_motion_serve; declare nx_motion_serve TAB <gate> and the ship loop proves it and the verdict lands here.
Work logno log row names this rung yet
MO9 Ingest the world's clip libraries into POSE · cell WATCHING · gate NO-GATE-DECLARED
Contractci_gltf_anim · executor Organ · 2 u · stands on MO5
Definition of doneglTF animation channels (rotation and translation samplers per node) and BVH frames read into NXA POSE library entries on the donor's own skeleton, so the CC0 Mesh2Motion and Quaternius clips and any BVH become donors for MO4 under a provenance row; done-rule: one Mesh2Motion clip round-trips node count and key count exactly, plays through nx_nxa_play with joints_posed equal to the animated node count, retargets onto the house rig with the limb-length tooth holding, and carries its CC0 row from the mirrored licence
Matrix cellINGEST: the world's clip libraries into POSE · runtime/nx_clip_ingest.nx · WATCHING decl
GateNO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_clip_ingest; declare nx_clip_ingest TAB <gate> and the ship loop proves it and the verdict lands here.
Work log2026-09-06 queue REAL THINGS TO INGEST, named from the mirrors so they are rungs and not citations: the Mesh2Motion assets repository (Blender source rigs and CC0 animation GLBs) and the Quaternius human clips are donor libraries for MO9; SkinTokens' Hugging Face weights and UniRig's checkpoints are oracle images for MO10; ModelsResource and VRoid Hub are named training sets whose licence terms decide whether any subset can join the MO11 corpus, and the operator's own rigged donors (witch, paladin, knight, toon3d8, ref9d) already can.
2026-09-12 queue Operator acceptance 2026-09-12: one user action and matching API/MCP workflow must ingest a referenced asset, discover body/outfit/hair dependencies, preserve source units/materials/provenance, recover and qualify reference motion, fit interchangeable catalogue garments with physical tightness controls, qualify hair appearance/dynamics and publish a reviewable Beach candidate without root-agent manual coordination. Current implementation does not satisfy this end-to-end requirement. Dark Elf Beastmaster https://modbooru.com/mods/dark-elf-beastmaster.2182 is observed as outfit/accessory listing with body variants; full pictured character and hair inclusion unknown. Linked Nexus189114/189845 returned content-preference/login restriction; archive contents and permissions unverified.
2026-09-12 measure Reference strut.gif SHA256 2d17175184e088f93cff9f165644c93f0a9bbfb77757ae154d28e5602b747811: browser ImageDecoder36 frames268x402 over1800000us; retained knowledge/gates/strut-gif-reference-20260912.json. Native nx_gif_decode currently first-image-only and skips timing/transparency/disposal; differential repair under review, not delivered motion recovery. Public GIF cannot establish hidden geometry or physical parameters.
MO10 The rival rigger as an oracle container · cell WATCHING · gate NO-GATE-DECLARED
Contractmo_oracle_skintokens · executor Organ · 2 u · stands on MO7
Definition of doneSkinTokens and UniRig run inside an oracle image the estate starts and stops through nx_container, fed the same upload, their skeleton and weights read back into NXA and graded by the dcc referee on the same page; nothing of theirs enters the build or run path; done-rule: one upload yields two rigs on one job page with two referee rows, and the oracle's absence prints UNAVAILABLE rather than a blank
Matrix cellORACLE: the rival rigger runs OUTSIDE as a container the estate starts and stops · runtime/nx_motion_oracle.nx · WATCHING decl
GateNO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_motion_oracle; declare nx_motion_oracle TAB <gate> and the ship loop proves it and the verdict lands here.
Work log2026-09-06 retract SKINTOKENS WAS NEVER THE CURRENT-MONTH BAR, corrected 2026-09-06 under the rule that a bar is the current month's or it is refused. Every meet-or-exceed sentence on this board and on dcc DC16 measured against SkinTokens (arXiv 2602.04805), which the mirrored listing dates 2026-02-04, seven months old, inherited from its README with no month attached. The September 2026 listing was fetched and read [arxiv_autorig_2026_09]: the rigging-relevant newest entries are Make-It-Poseable v3 (rig-free feed-forward posing, updated 2026-09-02) and, for skinning weights, A Geodesic Cut-Cell Prior for Neural Skinning (2026-08-11). Bar row MOB1 now carries the month; the oracle rung MO10 must run the CURRENT bar beside SkinTokens and MO11 scores against it. Nothing shipped is retracted, only the target.
2026-09-06 queue WHAT CLOSES THE WEIGHT-TOPOLOGY GAP (operator asked 2026-09-06 whether a code miner helps). The gap is top-1 weight agreement ~233/1000, a WEIGHTS-quality gap, not a skeleton gap (coverage 923). A rival SOURCE miner is LOW leverage here: SkinTokens' method is already published and stable (Skeleton Tree Tokenization + Bone-Point Cross-Attention + an FSQ skin-token VAE, read from its README) -- a standing source miner earns its keep only where the frontier is opaque and churning across many repos, which is exactly the langintel lane for compilers (nx_langintel_gh_fetch/llvm_ingest/rust_ingest feed /compare/lang from 1,651 banked oracle bugs). Ranked leverage: (1) FIX THE REFEREE'S METRIC FIRST -- top-1 LABEL agreement penalises us for naming a joint differently even when the deformation is identical; the honest metric is posed-DEFORMATION agreement (md_measure on a played pose), which is degenerate on ref9d because its one pose moves zero joints, so get a real posed comparison (a donor with a POSE library or a synthesised swing); cheap, and it may show the true gap is far smaller than 233. (2) THE ORACLE-IMAGE LANE (MO10) -- run SkinTokens/UniRig published WEIGHTS in an nx_container image the estate spins up and down, feed the same upload, score BOTH with our referee on a shared licence-clean corpus (MO11); this turns their marketing number into our measured delta, the consensus the progress law requires, and it is the operator's own 'repos as container oracle images' direction generalised estate-wide. (3) A SOVEREIGN LEARNED WEIGHT-REFINER (modding MD4 autorig_humanoid, a Local-model rung) trained on our OWN rigged corpus plus CC0 sets, no third-party checkpoint in the build path, with the pre-declared accept rule that it must beat the geodesic control on the referee. The investment that actually helps is therefore NOT a code miner but (a) the estate-wide ORACLE-IMAGE contract on /compare so any published rival becomes a measured competitor, and (b) a licence-clean ASSET miner to build the training corpus (Mesh2Motion-assets CC0, Quaternius, MakeHuman, VRoid), each carrying a modding-MD9 provenance row. The oracle and its weights stay OUTSIDE the build and run path by construction.
MO11 The shared benchmark, run whole · cell WATCHING · gate NO-GATE-DECLARED
Contractmo_bench_corpus · executor Organ · 2 u · stands on MO10
Definition of doneA licence-clean artist-rigged corpus (CC0 sets and the operator's own rigged donors) the referee runs over every model, never a sample, publishing coverage, joint error and weight agreement for ours and for each oracle on this board; done-rule: the partition of the corpus into graded, refused and unavailable sums to its size, the numbers land as rows the generator renders, and the meet-or-exceed sentence on this page is derived from them
Matrix cellBENCH: a shared artist-rigged corpus the referee runs over WHOLE · runtime/nx_rigbench.nx · WATCHING decl
GateNO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_rigbench; declare nx_rigbench TAB <gate> and the ship loop proves it and the verdict lands here.
Work log2026-09-12 lesson Frontier qualification requires dated representative comparative results, not newest publication date or symbol existence. Do not interpret historical roadmap training proposals as training authorization. Character workflow success is intermediate; photoreal Beach requires separate terrain/material/light/sky/water/vegetation and contact evidence plus actual served visual review.
MO12 A global-bar arm for the referee ladder · cell WATCHING · gate NO-GATE-DECLARED
Contractmo_bar_mode · executor Organ · 0.5 u · stands on nothing
Definition of donenx_curveskel_lib exposes the significance bar mode as a conf row so the 192-cell gain on the house rig can be attributed between resolution and the local-significance change with the same binary; done-rule: the two arms print two numbers on dcc.plan and only the difference is credited to the change
Matrix cellREFEREE: a global-bar arm for the ladder · runtime/nx_curveskel_lib.nx · WATCHING decl
GateNO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_curveskel_lib; declare nx_curveskel_lib TAB <gate> and the ship loop proves it and the verdict lands here.
Work logno log row names this rung yet
MO13 Chunked door for models past the edge budget · cell WATCHING · gate NO-GATE-DECLARED
Contractmo_upload_chunked · executor Organ · 1 u · stands on MO2
Definition of doneUploads larger than the named cap go through nx_content_put's chunked sha-receipted door into the same job directory, the page saying which door applies at which size; done-rule: toon3d8 (61,939,208 B) lands as a job through the chunked door with its sha matching the client receipt
Matrix cellSURFACE: chunked door for models past the edge budget · runtime/nx_motion_serve.nx · WATCHING decl
GateNO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_motion_serve; declare nx_motion_serve TAB <gate> and the ship loop proves it and the verdict lands here.
Work logno log row names this rung yet
MO14 A bind-smoke tooth so the daemon's listen is gated, not gambled · cell WATCHING · gate NO-GATE-DECLARED
Contractmo_bind_smoke · executor Organ · 1 u · stands on MO1
Definition of donenx_motion_gate forks the promoted nx_motion_serve.elf on a port nx_port_audit proves free, waits for the REFUSED->ACCEPT edge, GETs /motion for a 200 carrying the upload form, then kills it and asserts the port is free again; done-rule: the tooth is RED against a deliberately-occupied port and against a daemon that exits before accept, GREEN on the real binary, and its captured stderr names LISTEN-FAILED vs listening so a NEVER-LISTENED deploy is diagnosable from the gate, never from a lost spawn log
Matrix cellGATE: the daemon's listen is proven, not gambled · runtime/nx_motion_smoke.nx · WATCHING decl
GateNO-GATE-DECLARED — knowledge/organ_gate.conf carries no row for nx_motion_smoke; declare nx_motion_smoke TAB <gate> and the ship loop proves it and the verdict lands here.
Work log2026-09-06 queue MAPPING OWED so the ranker prices what exists: MO9 ci_gltf_anim, MO12 mo_bar_mode, MO13 mo_upload_chunked and MO14 mo_bind_smoke read UNMAPPED (no matrix row carries the symbol) and are priced at zero; each gets a watch row on motion.matrix. MO14 is the odd one: nx_motion_smoke SHIPPED (8731c7fe, exit 0 LISTENED / 3 BIND-FAILED / 4 NO-OUTPUT, proved the daemon binds 18132 three ways) but its only symbols are smw, smn and main, so the contract symbol mo_bind_smoke must be given to its core function in source before the row can flip -- a rename and rebuild, queued behind the storm (swap_used_permil 950 at the time of writing, no builds). THE /motion-LIVE RECIPE, in order: confirm the supervisor holds nx_motion_serve on 18132 (nx_health reads 16 services OK and 1 supervisor; the port itself is unwitnessed), add the proxy_routes.conf row for /motion at 18132, reload the front door, nx_page_verify the live page, then flip mo_serve, mo_upload_accept and mo_verdict_row to their real symbols and run nishi_compare_regen so /compare/motion renders with the dated bars.
2026-09-06 measure DAEMON PROVEN HEALTHY, DEPLOY PROBE IS THE FALSE NEGATIVE (2026-09-06, rung MO14 core shipped as nx_motion_smoke, built 22,906 B 8731c7fe, registered staged). nx_motion_smoke forks the daemon binary on a throwaway port with gk_run_capture_ms and captures its own stdout+stderr: on port 28139, on 18132, and on the installed serving-root ./nx_motion_serve.elf, ALL THREE print '[nx_motion_serve] listening on loopback' and are killed by the 2.5 s watchdog (rc=137). So the binary binds 18132 in under a second and the two /api/deploy NEVER-LISTENED rollbacks are the health-probe missing the REFUSED->ACCEPT edge under load -- the documented deploy-lane fragility, not this daemon. Bring-up: the proven binary is installed at the serving root (nx_motion_serve.elf 92,397 B, mode 0755) and daemons.reg carries a signed revive row, so the fleet supervisor -- which does NOT use the deploy edge-probe -- starts and keeps it. /api/restart has a FIXED service allowlist that does not include nx_motion_serve, and nx_http_probe answers the host redirector's 301 on any loopback port, so neither nudges nor verifies it; the live URL waits on the supervisor cycle. MO14 finishes when nx_motion_smoke also does the HTTP GET /motion round-trip (200 + the upload form) and becomes a tooth in nx_motion_gate, so the daemon's listen is gated repeatably and a NEVER-LISTENED deploy is diagnosable from the gate.

dossiers=14 cells_joined=14 no_matrix_row=0 (sum 14) gates_recorded=0 gates_unrecorded=0 no_gate_declared=14 (sum 14 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. 18 feature rows mapped of 18 matrix rows; 0 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/motion.cdmap (cut|x|y, cube|x|y|z); absent = the HBR defaults
Position map, two-dimensional cutOne bubble per player. Bubble area is breadth, the ring colour is momentum, the dashed lines are the category means, and both axes are fitted to the field of play with their tick numerals shown. Every value is repeated in the table that follows. 100 200 300 400 500 600 700 800 900 0 200 400 600 800 centrality (permil, fitted 100–900) distinctiveness (permil, fitted 0–800) mean 611 mean 369 Unconventional Aspirational Peripheral Mainstream Nishi, centrality 714, distinctiveness 617, breadth 425, depth 851, momentum 0 Mesh2Motion, centrality 785, distinctiveness 357, breadth 277, depth 714, momentum 0 Mixamo, centrality 785, distinctiveness 357, breadth 277, depth 714, momentum 0 Tripo3D, centrality 785, distinctiveness 230, breadth 222, depth 571, momentum 0 AccuRIG, centrality 428, distinctiveness 74, breadth 74, depth 444, momentum 0 Cascadeur, centrality 547, distinctiveness 255, breadth 166, depth 600, momentum 0 SkinTokens, centrality 238, distinctiveness 694, breadth 185, depth 833, momentum 0 Nishi Mesh2Motion Mixamo Tripo3D AccuRIG Cascadeur SkinTokens

bubble area = breadth · ring = momentum (green rising, red falling, grey until day two) · dashed = category means · axes fitted to the field of play: centrality 100–900, distinctiveness 0–800 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 1 (a fact of the data: two players that close are that close) · data spread 730 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)
Position map, three-dimensional cubeThe same players in an isometric cube. Each axis is fitted to its own field of play and carries numerals on the floor grid and the vertical axis; each bubble drops a dotted line to its floor shadow so height reads as height. Values are in the table that follows. 100 200 300 400 500 600 700 800 900 0 200 400 600 800 0 100 200 300 400 500 centrality distinctiveness breadth AccuRIG: centrality 428, distinctiveness 74, breadth 74 Cascadeur: centrality 547, distinctiveness 255, breadth 166 SkinTokens: centrality 238, distinctiveness 694, breadth 185 Tripo3D: centrality 785, distinctiveness 230, breadth 222 Mesh2Motion: centrality 785, distinctiveness 357, breadth 277 Mixamo: centrality 785, distinctiveness 357, breadth 277 Nishi: centrality 714, distinctiveness 617, breadth 425 Nishi Mesh2Motion Mixamo Tripo3D AccuRIG Cascadeur SkinTokens
axes fitted: centrality 100–900 · distinctiveness 0–800 · breadth 0–500 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 1 (a fact of the data: two players that close are that close) · data spread 372 permil of the plot
Position map, small multiplesEvery pair of registered axes as one small cut. Each panel fits both axes to the field of play, the dashed lines are the category means, bubble area is breadth and the ring colour is momentum; the legend below names the colours. 100 900 0 800 distinctiveness vs centrality Nishi: centrality 714, distinctiveness 617 Mesh2Motion: centrality 785, distinctiveness 357 Mixamo: centrality 785, distinctiveness 357 Tripo3D: centrality 785, distinctiveness 230 AccuRIG: centrality 428, distinctiveness 74 Cascadeur: centrality 547, distinctiveness 255 SkinTokens: centrality 238, distinctiveness 694 100 900 0 500 breadth vs centrality Nishi: centrality 714, breadth 425 Mesh2Motion: centrality 785, breadth 277 Mixamo: centrality 785, breadth 277 Tripo3D: centrality 785, breadth 222 AccuRIG: centrality 428, breadth 74 Cascadeur: centrality 547, breadth 166 SkinTokens: centrality 238, breadth 185 100 900 300 900 depth vs centrality Nishi: centrality 714, depth 851 Mesh2Motion: centrality 785, depth 714 Mixamo: centrality 785, depth 714 Tripo3D: centrality 785, depth 571 AccuRIG: centrality 428, depth 444 Cascadeur: centrality 547, depth 600 SkinTokens: centrality 238, depth 833 100 900 480 520 momentum vs centrality Nishi: centrality 714, momentum 500 Mesh2Motion: centrality 785, momentum 500 Mixamo: centrality 785, momentum 500 Tripo3D: centrality 785, momentum 500 AccuRIG: centrality 428, momentum 500 Cascadeur: centrality 547, momentum 500 SkinTokens: centrality 238, momentum 500 0 800 0 500 breadth vs distinctiveness Nishi: distinctiveness 617, breadth 425 Mesh2Motion: distinctiveness 357, breadth 277 Mixamo: distinctiveness 357, breadth 277 Tripo3D: distinctiveness 230, breadth 222 AccuRIG: distinctiveness 74, breadth 74 Cascadeur: distinctiveness 255, breadth 166 SkinTokens: distinctiveness 694, breadth 185 0 800 300 900 depth vs distinctiveness Nishi: distinctiveness 617, depth 851 Mesh2Motion: distinctiveness 357, depth 714 Mixamo: distinctiveness 357, depth 714 Tripo3D: distinctiveness 230, depth 571 AccuRIG: distinctiveness 74, depth 444 Cascadeur: distinctiveness 255, depth 600 SkinTokens: distinctiveness 694, depth 833 0 800 480 520 momentum vs distinctiveness Nishi: distinctiveness 617, momentum 500 Mesh2Motion: distinctiveness 357, momentum 500 Mixamo: distinctiveness 357, momentum 500 Tripo3D: distinctiveness 230, momentum 500 AccuRIG: distinctiveness 74, momentum 500 Cascadeur: distinctiveness 255, momentum 500 SkinTokens: distinctiveness 694, momentum 500 0 500 300 900 depth vs breadth Nishi: breadth 425, depth 851 Mesh2Motion: breadth 277, depth 714 Mixamo: breadth 277, depth 714 Tripo3D: breadth 222, depth 571 AccuRIG: breadth 74, depth 444 Cascadeur: breadth 166, depth 600 SkinTokens: breadth 185, depth 833 0 500 480 520 momentum vs breadth Nishi: breadth 425, momentum 500 Mesh2Motion: breadth 277, momentum 500 Mixamo: breadth 277, momentum 500 Tripo3D: breadth 222, momentum 500 AccuRIG: breadth 74, momentum 500 Cascadeur: breadth 166, momentum 500 SkinTokens: breadth 185, momentum 500 300 900 480 520 momentum vs depth Nishi: depth 851, momentum 500 Mesh2Motion: depth 714, momentum 500 Mixamo: depth 714, momentum 500 Tripo3D: depth 571, momentum 500 AccuRIG: depth 444, momentum 500 Cascadeur: depth 600, momentum 500 SkinTokens: depth 833, momentum 500
Nishi Mesh2Motion Mixamo Tripo3D AccuRIG Cascadeur SkinTokens
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 14 (a fact of the data: two players that close are that close) · data spread 566 permil of the plot
PlayerQuadrantcentralitydistinctivenessbreadthdepthmomentumRows heldDays on spineFirst seen
NishiAspirational7146174258510 since 2026-09-239102026-09-15
Mesh2MotionMainstream7853572777140 since 2026-09-237102026-09-15
MixamoMainstream7853572777140 since 2026-09-237102026-09-15
Tripo3DMainstream7852302225710 since 2026-09-237102026-09-15
AccuRIGPeripheral42874744440 since 2026-09-233102026-09-15
CascadeurPeripheral5472551666000 since 2026-09-235102026-09-15
SkinTokensUnconventional2386941858330 since 2026-09-234102026-09-15
DayMatrix rows reviewedPlayers recorded
2026-09-15187
2026-09-16187
2026-09-17187
2026-09-18187
2026-09-19187
2026-09-20187
2026-09-21187
2026-09-22187
2026-09-23187
2026-10-08187

players 7|matrix rows 18|feature mass 42|centrality mean 611|distinctiveness mean 369|axes 5|spine days 10 (shown 10)|rows written today 0|readability defects 2D 0 cube 0|spine knowledge/status/cdmap/motion.spine

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

CapabilityNishiMesh2MotionMixamoTripo3DAccuRIGCascadeurSkinTokens
SURFACE
the /motion door, a loopback daemon behind one route rowDARK — runtime/nx_motion_serve.nx EXISTS but does not declare mo_serve: 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. MO1 watch. A public NishiLang daemon on loopback behind one proxy_routes.conf row (the /math pattern: plain GET and POST forms, no script, no third-party runtime, every job a shareable URL) composing nx_gltf2mesh, nx_fbx2nxa, nx_autorig_mesh, nx_nxa_texbake view, nx_nxa_retarget, nx_nxa_play and nx_mesh2glb by fork. Mesh2Motion runs entirely in the visitor's browser as TypeScript [mesh2motion]; Mixamo is Adobe's hosted service [mixamo]; Tripo is a hosted product with SDKs [tripo3d] [vastai]; AccuRIG and Cascadeur are desktop applications [accurig] [cascadeur]; SkinTokens is a research repository with weights on Hugging Face [skintokens]. The exceed claimed is sovereignty of the whole path, and it is claimed only when the daemon ships and the page is verified live.
dossierwatching mo_serve
Pros none measured yetCons behind Mesh2Motion, Mixamo, Tripo3D; open contract; expected source declaration not established
○●●●○○○
upload a model in the browser and get it riggedDARK — runtime/nx_motion_serve.nx EXISTS but does not declare mo_upload_accept: 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. MO2 watch. A multipart upload with a NAMED byte cap (the edge proxy's measured body budget, never a guess), content-typed by MAGIC (GLB, FBX, NXA; OBJ refused by name until nx_obj_import is composed), refused by name over the cap or on an unknown type, and handed a job id the visitor can return to. Mesh2Motion imports GLB, GLTF, DAE and FBX [mesh2motion]; Mixamo uploads FBX, OBJ or a zip [mixamo]; Tripo's hero upload takes an IMAGE and generates the mesh, so a mesh upload reads Part [tripo3d]; AccuRIG and Cascadeur open local files in a desktop app [accurig] [cascadeur].
dossierwatching mo_upload_accept
Pros none measured yetCons behind Mesh2Motion, Mixamo, Tripo3D; open contract; expected source declaration not established
○●●◐○○○
RIG
skeleton and weights from the volume with no archetype hintSource present; authored stronger claim: mr_rig in runtime/nx_motion_rig.nx, verified at emit. MO3 watch. The surface's rig job forks the DC16 one-call auto-rig (dcc board, amesh_run: curve-skeleton peeling then geodesic bone heat -- no archetype, no checkpoint, no training data) at the ladder resolution the visitor picks and writes every number the composed organs print into the job's status rows. Measured on the dcc board 2026-09-06 at 192 cells on the house rig: 923 permil of artist joints within one bone length, 40.1 mm mean joint error, top-1 weight agreement 233 permil -- coverage is competitive, weight topology is not, stated plainly. Mesh2Motion offers six archetype rigs the user positions [mesh2motion]; Mixamo and Tripo auto-rig humanoids [mixamo] [tripo3d]; AccuRIG rigs humanoids and quadrupeds [accurig]; Cascadeur animates an existing rig [cascadeur]; SkinTokens predicts skeleton and skin for any category with a learned model and reports 17 to 22 percent better bone prediction and 98 to 133 percent better skinning than prior SOTA [skintokens] [vastai] -- Best on the learned axis, and the bar MO11 must meet on a shared benchmark before any exceed beyond sovereignty is claimed. Adoption: RUN-BY:fork:nx_motion_serve — fully adopted (top of its ladder).
dossier
Pros Source symbol found; stronger capability claim is authored; ahead of Mesh2Motion, Mixamo, Tripo3D, AccuRIG, Cascadeur; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)
◉◐◐◐◐○◉
resolution ladder chosen by the visitor, announced on the jobSource present; authored stronger claim: cs_run_cells in runtime/nx_curveskel_lib.nx, verified at emit. cs_run_cells (2026-09-06) takes a caller's cell count above the conf row above the bootstrap default; the referee ladder measured 96 to 192 cells moving coverage 461 to 923 permil and mean joint error 208.9 to 40.1 mm on the house rig (dcc.plan). No rival exposes a rig resolution; each ships one fixed pipeline, so c1 to c6 read 0. The surface exposes the ladder as a form choice with its measured cost beside it. Adoption: LIB-WIRED importers=6 nonval=4 — fully adopted (top of its ladder).
Pros Source symbol found; stronger capability claim is authored; ahead of Mesh2Motion, Mixamo, Tripo3D, AccuRIG, Cascadeur, SkinTokens; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)
◉○○○○○○
MOTION
apply a library clip to the fresh rigMeasured: mr_anim exists in runtime/nx_motion_rig.nx, verified at emit. MO4 watch. The anim verb retargets a donor's POSE library entry onto the fresh rig through nx_nxa_retarget (modding MD3: topology plus normalised bind position, limb length preserved by construction), bakes it with the served nx_nxa_play, renders the posed rig with the sovereign rasteriser and exports it. Mesh2Motion ships over 150 CC0 clips per archetype [mesh2motion]; Mixamo's motion-capture library is Best on breadth [mixamo]; Tripo auto-animates [tripo3d]; AccuRIG hands the rig to Mixamo or iClone [accurig]; Cascadeur authors motion physically on a rig [cascadeur]; SkinTokens rigs only [skintokens]. Adoption: RUN-BY:fork:nx_motion_serve — fully adopted (top of its ladder).
dossier
Pros Implementing source symbol found; runtime acceptance separate; ahead of AccuRIG, SkinTokens; fully adopted on the estate ladderCons behind Mixamo (leads)
●●◉●◐●○
clip library as data rowsMeasured: mo_clip_list exists in runtime/nx_motion_lib.nx, verified at emit. MO5 watch. knowledge/motion_clips.rows, one row per clip: key, donor NXA carrying the POSE library, pose id, title, licence class, provenance -- read by the page and by the anim verb, never a list in code. Today's donors are the house rig's own salsa library and the walk; the CC0 Mesh2Motion and Quaternius libraries [mesh2motion] are the MO9 ingest rung, not a row yet. Adoption: LIB-WIRED importers=5 nonval=3 — fully adopted (top of its ladder).
dossier
Pros Implementing source symbol found; runtime acceptance separate; ahead of AccuRIG, Cascadeur, SkinTokens; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)
●●●●○◐○
BVH clip ingestMeasured: m_joints exists in runtime/nx_bvhmotion.nx, verified at emit. m_joints walks a BVH joint tree (index, parent, channels, offsets, names) and the reader reconciles values to frames times channels EXACTLY on two real files; it does not yet emit a POSE library entry, so a BVH clip cannot drive the fresh rig until MO9 lands. Cascadeur imports and exports motion files [cascadeur]; the web tools take only their own libraries. [mesh2motion_customclip_20260919] 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 Mesh2Motion, Mixamo, Tripo3D, AccuRIG, SkinTokensCons 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
rigged mesh as GLB with skinMeasured: mg_convert_nxa exists in runtime/_hdl_build/nx_mesh2glb.nx, verified at emit. mg_convert_nxa writes the NXA's VERT, TRIS, SKEL and SKIN as a glTF binary with a skin (charsim R22 round-trips it byte-identical); animation channels in the GLB are modding MD10 and are not re-priced here, so an exported rig plays in Godot, Unity, Unreal, Three.js and Babylon.js but carries no clip yet. Mesh2Motion exports several animations packed into one GLB and is Best [mesh2motion]; Mixamo exports FBX [mixamo]; Tripo exports GLB and FBX [tripo3d]; AccuRIG and Cascadeur export FBX [accurig] [cascadeur]. [mesh2motion_loader_20260919] [mesh2motion_export_20260919] Adoption: RUN-BY:fork:nx_nxa2glb_gate — fully adopted (top of its ladder).
Pros Implementing source symbol found; runtime acceptance separate; ahead of SkinTokens; fully adopted on the estate ladderCons behind Mesh2Motion (leads)
●◉●●●●○
PAGE
emitted from data, never hand-written, static fallback byte-identicalSource present; authored stronger claim: mo_body_index in runtime/nx_motion_lib.nx, verified at emit. MO6 watch. One renderer (the /math pattern) reads knowledge/motion_page.tmpl plus the clip rows and the demo rows, serves it live, and emits the same bytes as the static fallback file; the world's hand-written viewer pages (rigview, refmesh) were measured ABSENT from every emitter on 2026-09-06 and are the defect this row exists to stop. No rival publishes how its page is produced; c1 to c6 read 0 as NOT-WITNESSED. Adoption: LIB-WIRED importers=5 nonval=3 — fully adopted (top of its ladder).
dossier
Pros Source symbol found; stronger capability claim is authored; ahead of Mesh2Motion, Mixamo, Tripo3D, AccuRIG, Cascadeur, SkinTokens; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)
◉○○○○○○
REFEREE
the rig grade travels with the surfaceSource present; authored stronger claim: mo_demo_cards in runtime/nx_motion_lib.nx, verified at emit. MO7 watch. Every demo rig on the page carries the dcc referee's numbers (rb_run: coverage, joint error, top-1 weight agreement, cells, method) beside its render, and every uploaded rig carries the composed organs' own counts; an upload the referee cannot grade says UNGRADED, never nothing. UniRig and SkinTokens publish accuracy against artist rigs on shared benchmarks [unirig] [skintokens]; the product sites publish none. Adoption: LIB-WIRED importers=5 nonval=3 — fully adopted (top of its ladder).
dossier
Pros Source symbol found; stronger capability claim is authored; ahead of Mesh2Motion, Mixamo, Tripo3D, AccuRIG, Cascadeur; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)
◉○○○○○◉
ACCEPT
operator verdict per upload as dataDARK — runtime/nx_motion_serve.nx EXISTS but does not declare mo_verdict_row: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the comparewatch- plane row reads DARK until then. MO8 watch. An ACCEPT or REJECT form on every job writes a row (id, subject, verdict, date, actor, words, targets) to the acceptance plane knowledge/store/accept- so the three-party completion this estate requires is a click on the surface, never a message in a chat, and a later verdict supersedes by a new row. No rival keeps an acceptance record.
dossierwatching mo_verdict_row
Pros none measured yetCons open contract; expected source declaration not established
○○○○○○○
DETERMINISM
same upload, same rig bytesSource present; authored stronger claim: mr_sha_from_log in runtime/nx_motion_rig.nx, verified at emit. MO3 writes the sha256 of the rigged NXA into the job status; the auto-rig is integer and deterministic by construction (dcc), so two uploads of one file must print one digest and the gate asserts it. Learned riggers sample and the product sites do not state determinism, so c1 to c6 read 0 as NOT-WITNESSED. Adoption: RUN-BY:fork:nx_motion_serve — fully adopted (top of its ladder).
Pros Source symbol found; stronger capability claim is authored; ahead of Mesh2Motion, Mixamo, Tripo3D, AccuRIG, Cascadeur, SkinTokens; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)
◉○○○○○○
ORACLE
the rival rigger runs OUTSIDE as a container the estate starts and stopsOpen — watching runtime/nx_motion_oracle.nx : mo_oracle_skintokens, re-measured on every compare beat. Ship that symbol and this mark flips itself; the comparewatch- plane row flips with it. MO10 watch. SkinTokens' published weights [skintokens] run in an oracle container the estate spins up and down (operator direction 2026-09-05; incumbent runtime/nx_container.nx), never in the build or run path, so the same upload can be rigged by the rival and graded by the same referee on the same page -- consensus, not a claim. [meshy_feature_20260919] [meshy_api_20260919] [meshy_mcp_repo_20260919] [meshy_mcp_source_20260919] [mesh2motion_home_20260919] [mesh2motion_repo_20260919] [mesh2motion_desktop_20260919]
dossierwatching mo_oracle_skintokens
Pros none measured yetCons behind SkinTokens; open contract; expected source declaration not established
○○○○○○●
BENCH
a shared artist-rigged corpus the referee runs over WHOLEDARK — runtime/nx_rigbench.nx EXISTS but does not declare mo_bench_corpus: 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. MO11 watch. ModelsResource and VRoid Hub are named as SkinTokens' training sets [skintokens]; a licence-clean corpus (CC0 sets plus the operator's own rigged donors: witch, paladin, knight, toon3d8 and the house rig) is what the dcc referee runs over whole, never sampled, and the number on this board is that population's, not one donor's. Until it runs, the meet-or-exceed question stays open on this page by design. [mesh2motion_assets_20260919]
dossierwatching mo_bench_corpus
Pros none measured yetCons behind SkinTokens; open contract; expected source declaration not established
○○○○○○●
INGEST
the world's clip libraries into POSEOpen — watching runtime/nx_clip_ingest.nx : ci_gltf_anim, re-measured on every compare beat. Ship that symbol and this mark flips itself; the comparewatch- plane row flips with it. MO9 watch. glTF animation channels and BVH frames read into NXA POSE library entries on the donor's own skeleton so the CC0 Mesh2Motion and Quaternius clips [mesh2motion] and any BVH become donors for the anim verb under a provenance row; the intended home is a clip-ingest organ composing nx_gltf2mesh's node walk and nx_bvhmotion's frame reader. Mesh2Motion and Mixamo ship clip libraries (Best); Tripo auto-animates; Cascadeur imports motion files; SkinTokens rigs only.
dossierwatching ci_gltf_anim
Pros none measured yetCons behind Mesh2Motion (leads), Mixamo (leads), Tripo3D, Cascadeur; open contract; expected source declaration not established
○◉◉●○●○
REFEREE
a global-bar arm for the ladderDARK — runtime/nx_curveskel_lib.nx EXISTS but does not declare mo_bar_mode: 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. MO12 watch. The significance bar mode (global root radius versus local attachment radius) exposed as a conf row so the 192-cell gain on the house rig is attributed between resolution and the local-significance change with one binary; no rival exposes any such arm, so c1 to c6 read 0.
dossierwatching mo_bar_mode
Pros none measured yetCons open contract; expected source declaration not established
○○○○○○○
SURFACE
chunked door for models past the edge budgetDARK — runtime/nx_motion_serve.nx EXISTS but does not declare mo_upload_chunked: 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. MO13 watch. Uploads over the named cap go through nx_content_put's chunked sha-receipted door (dataio DI5 to DI13) into the same job directory, the page saying which door applies at which size; done-rule: toon3d8 (61,939,208 B) lands as a job with its sha matching the client receipt. Mesh2Motion runs in-browser with no server cap and the hosted products publish none, so c1 to c6 read 0 as NOT-WITNESSED.
dossierwatching mo_upload_chunked
Pros none measured yetCons open contract; expected source declaration not established
○○○○○○○
GATE
the daemon's listen is proven, not gambledDARK — runtime/nx_motion_smoke.nx EXISTS but does not declare mo_bind_smoke: 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. MO14 watch. nx_motion_smoke (8731c7fe) forks the daemon binary on a throwaway port and exits 0 LISTENED, 3 BIND-FAILED or 4 NO-OUTPUT; it proved nx_motion_serve binds three ways after the health-checked deploy reported NEVER-LISTENED against a live binary. The contract symbol is owed to its core function by rename (today it carries only smw, smn and main), so this row flips on that rebuild and never on the presence of main. No rival publishes a bind gate.
dossierwatching mo_bind_smoke
Pros none measured yetCons 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
ClassCountNote
rows with rival marks11rows carrying a Yes, Best or Part in a rival column
cited11a reference mark resolves to a pinned mirror
uncited0an observation read, badged until a reference lands
rows, summed11RECONCILED: the parts sum to the total 11

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.
WhenRungKindEntry
2026-09-06MO10retractSKINTOKENS WAS NEVER THE CURRENT-MONTH BAR, corrected 2026-09-06 under the rule that a bar is the current month's or it is refused. Every meet-or-exceed sentence on this board and on dcc DC16 measured against SkinTokens (arXiv 2602.04805), which the mirrored listing dates 2026-02-04, seven months old, inherited from its README with no month attached. The September 2026 listing was fetched and read [arxiv_autorig_2026_09]: the rigging-relevant newest entries are Make-It-Poseable v3 (rig-free feed-forward posing, updated 2026-09-02) and, for skinning weights, A Geodesic Cut-Cell Prior for Neural Skinning (2026-08-11). Bar row MOB1 now carries the month; the oracle rung MO10 must run the CURRENT bar beside SkinTokens and MO11 scores against it. Nothing shipped is retracted, only the target.
2026-09-06MO3lessonTHE CURRENT BAR VALIDATES THE CLASS AND NAMES THE GAP. The August 2026 skinning paper argues in its own abstract that data-driven weight generation lacks the generalisability of classic geometric approaches and proposes a robustly computable geometric prior to bridge it [arxiv_autorig_2026_09]; our geodesic bone-heat weights are that prior class, computed sovereignly with no training set, proven at p95 6.3 mm deformation against the artist rig on the house mesh (dcc.plan). The September paper names inaccurate skinning weights on fused AI-generated geometry as the open failure, which is exactly the witch result (coverage 396 permil, extremities under-driven). So the ladder to the current bar is prior PLUS learned refinement (modding MD4) graded by the same referee, and a seat that tunes the falloff instead is climbing the wrong rung.
2026-09-06MO3landSCOREBOARD TRUTH 2026-09-06: nx_compare_rank read motion 0 of 14 done while nx_motion_rig (PROMOTED bf4d6fd7), nx_motion_lib and nx_motion_gate (46 of 46, PROMOTED) had shipped, because six matrix watch rows and five rung contracts named symbols that were never built under those names -- the DC16 lesson again, a rung symbol must name a symbol that exists. Repointed BOTH the rung and its matrix row to the real entry point read from the organ: MO3 mr_rig, MO4 mr_anim, MO5 mo_clip_list, MO6 mo_body_index, MO7 mo_demo_cards, plus the digest row mr_sha_from_log; the three renderer rows now point at runtime/nx_motion_lib.nx where the symbols live. DELIBERATELY NOT REPOINTED: mo_serve, mo_upload_accept and mo_verdict_row, because their done-rules need the daemon LIVE behind a route row and nx_motion_serve, though PROMOTED at the serving root (92,397 B), is not yet witnessed serving 18132 (deploy rolled back NEVER-LISTENED, supervisor row signed, route row and nx_page_verify owed) -- a flip is a receipt and there is no live surface to receipt yet. Every edit landed under a size CAS; a thirteen-edit batch proved the async lane runs a batch CONCURRENTLY, so only the rung whose expect matched the original size landed and the rest were refused by name -- one write per file per round from then on.
2026-09-06MO14queueMAPPING OWED so the ranker prices what exists: MO9 ci_gltf_anim, MO12 mo_bar_mode, MO13 mo_upload_chunked and MO14 mo_bind_smoke read UNMAPPED (no matrix row carries the symbol) and are priced at zero; each gets a watch row on motion.matrix. MO14 is the odd one: nx_motion_smoke SHIPPED (8731c7fe, exit 0 LISTENED / 3 BIND-FAILED / 4 NO-OUTPUT, proved the daemon binds 18132 three ways) but its only symbols are smw, smn and main, so the contract symbol mo_bind_smoke must be given to its core function in source before the row can flip -- a rename and rebuild, queued behind the storm (swap_used_permil 950 at the time of writing, no builds). THE /motion-LIVE RECIPE, in order: confirm the supervisor holds nx_motion_serve on 18132 (nx_health reads 16 services OK and 1 supervisor; the port itself is unwitnessed), add the proxy_routes.conf row for /motion at 18132, reload the front door, nx_page_verify the live page, then flip mo_serve, mo_upload_accept and mo_verdict_row to their real symbols and run nishi_compare_regen so /compare/motion renders with the dated bars.
2026-09-06MO1measureDAEMON NOT SERVED, witnessed 2026-09-06: nx_services lists 16 services UP (sites 8443, wiki_gw, gallery gateway and serve, media auth, redirect, torrent_gw, clock, mgmt_api 18098, docportal 18456, email 18465, siteedit 18466, inventory 18122, vroom 8446, tools_api 18096, opaque_login 9091) and nx_motion_serve is NOT among them, so the signed daemons.reg revive row did not bring it up and MO1 stays open by measurement, not by caution. The binary is PROMOTED at the serving root (92,397 B) and nx_motion_smoke proved it binds; the two /api/deploy attempts rolled back NEVER-LISTENED (the level-probe defect this estate already records). NEXT EXPERIMENT, not a cause: read the supervisor's own stdout for HELD-TAMPER (a daemons.reg whose signature does not verify makes the supervisor hold its last-good roster, which would not carry the new row) versus a roster read only at startup; whichever it names, the remedy is one re-sign or one supervisor restart, never a hand-launched process on a shared port. A cap was minted for nx_motion_rig the same round (30 days), closing the REGISTERED-UNAUTHORISED residual the ranker named on MO3 and MO4.
2026-09-06MO1measureTHE SUPERVISOR-LOG EXPERIMENT RAN 2026-09-06: supervisor.log (2,862,652 B at the nishihost root, scanned whole) carries NO line naming nx_motion_serve and NO TAMPER line, so the HELD-TAMPER arm is refuted (a rejected signature would have been logged) and the remaining arm stands: the supervisor read daemons.reg at its own start and has not re-read it since, so a row appended afterwards is invisible to it until it restarts or is told to reload. The box reads DEGRADED (mem-pressure, swap 952 permil), so neither a supervisor restart nor a third /api/deploy attempt is taken now: a slow start under memory pressure would read NEVER-LISTENED again and teach nothing. RECIPE when health reads OK: /api/deploy target=nx_motion_serve (never-brick, auto-rollback) and, if it rolls back NEVER-LISTENED against a binary nx_motion_smoke proves binds, the supervisor reload is the remedy, never a hand-launched process on a shared port.
2026-09-06MO3landTHE PERMANENT FIX FOR DARK CAPABILITY, source-landed 2026-09-06 (operator: this needs a permanent fix, we do not want to build and lose capabilities to darkness). The one shared ruler already told organ-unreadable (-1) from organ-read-and-symbol-absent (0); every consumer folded both into OPEN. Now: nx_swcompare_matrix records rdark per watch row (0 under a rule that can judge, the JS abstention excluded through the applied rule), renders the cell as data-s dk with its own evidence prose, a Dark filter chip, a method sentence and dark_contracts= in both its receipts (BUILT by the governor 455,816 B sha 708d15d7, promoted with the one superseded prose run named); nx_compare_regen writes state DARK to the comparewatch plane with the partition OPEN plus LANDED plus DARK equal to stored (PROMOTED 187,616 B sha 42fabc16, contentdiff ONE-WAY-GAIN, _offc refreshed); nx_compare_rank carries rdark_r per rung and prints DARK-CONTRACT on the rung line (build submitted); nx_swcompare_watch_gate T30, T30a, T30b and T30c pin the predicate on the ruler with a missing-organ and a JS-abstain neg-control (build queued). ALSO: every dated bar on this board and on dcc was moved to the sotabar kind the sibling seat's freshness contract reads (bar is the renderer's research-bar grammar and would have rendered these rows garbled), so the served ranker reads NO-BAR until the rebuilt one reads FRESH; the sotatarget and rungrole ladder rows are owed on both boards. The hardened referee nx_rigbench is PROMOTED at 125,528 B sha d441bc19 with the normalize pass (md_scale_div= printed) and its _offc twin refreshed; nx_rigbench_gate carries T12 and is queued as a sovereign build.

Risk register

RiskLikelihood x impactMitigation
The edge proxy is crashed by an oversized uploadlikely x highThe edge crashed on oversized proxied POST bodies on 2026-07-20; MO2 caps the body at a NAMED constant derived from that budget and refuses over it by name, and MO13 routes larger models through the chunked door that already carries the estate's own deploys.
An untrusted mesh crashes a parserlikely x mediumEvery reader the job forks is fail-closed by name (nx_gltf2mesh refuses at the offset it cannot parse, nx_fbx2nxa refuses non-FBX, the skeleton stage refuses a volumeless sheet), the daemon binds loopback and only ever forks named organs on digit-named job paths, and a job that dies leaves its status rows and no done marker.
A rival's marketing number is read as a measurementcertain todayEvery rival cell is a documented-capability read and every bar is UNPINNED (C2) until MO11 runs the same referee over a shared corpus; the page says so in its own words.
A clip library's licence is not carriedpossible x highMO5 rows carry a licence class and MO9 attaches the mirrored licence text through the modding provenance row (MD9) before any ingested clip is served; a clip without a class renders as UNLICENSED and does not apply.
The NAS is hammered by concurrent rig jobslikely x mediumOne job runs at a time per daemon, a second upload queues with its position printed, the daemon reads the build-admission envelope before spawning and prints DEFERRED with the box's own numbers when the array is storming.

Release history continued

WhenRungKindEntry
2026-09-06MO1measureBOARD OPENED 2026-09-06 from the operator's ask. Measured that day: the dcc referee on the house rig reads 461 permil coverage and 208.9 mm at 96 cells and 923 permil and 40.1 mm at 192 cells on the new libs (unattributed between resolution and local significance until MO12), top-1 weight agreement 233 permil; the world's viewer pages have no emitter (nx_absent ABSENT-PROVEN over every organ source); the /mesh route has no daemon behind it and its source is the gallery gateway; nx_autorig_mesh, nx_gltf2mesh, nx_fbx2nxa, nx_nxa_retarget, nx_nxa_play, nx_nxa_texbake and nx_mesh2glb are all LIVE in the catalogue; SkinTokens' own README states 98 to 133 percent skinning and 17 to 22 percent bone gains over prior SOTA; Tripo's page states auto-rig for humanoids in seconds; Mesh2Motion's page states four import formats, six archetype rigs and over 150 CC0 clips.
2026-09-06MO9queueREAL THINGS TO INGEST, named from the mirrors so they are rungs and not citations: the Mesh2Motion assets repository (Blender source rigs and CC0 animation GLBs) and the Quaternius human clips are donor libraries for MO9; SkinTokens' Hugging Face weights and UniRig's checkpoints are oracle images for MO10; ModelsResource and VRoid Hub are named training sets whose licence terms decide whether any subset can join the MO11 corpus, and the operator's own rigged donors (witch, paladin, knight, toon3d8, ref9d) already can.
2026-09-06MO1landBOARD ADMITTED 2026-09-06 (nx_domain_admit motion: VERDICT=ADMIT, CLAIM-ONLY, rows=14 grounded=3 watch-new-organ=10, bars declared unpinned) and the surface's sources WRITTEN, NOT YET BUILT OR DEPLOYED: buildroot/runtime/nx_motion_lib.nx 41,587 B (typing by magic, multipart, form decoding, status rows, row files, the fork runner, both page renderers), nx_motion_rig.nx 14,931 B (rig and anim verbs over the promoted organs, status rows at every stage, done marker last), nx_motion_serve.nx 21,464 B (loopback 18132, plain forms, detached spawn, art and demo serving, the verdict row to the journal and the acceptance plane), nx_motion_serve_gate.nx 17,221 B (rulers on planted inputs, the box fixture rigged twice for a determinism tooth, an OBJ refusal control); data knowledge/motion_page.tmpl, knowledge/motion_clips.rows (house_pose0: ref9d.nxa pose 0, 104 joints posed, max displacement 549 mm -- the only real pose on the house assets, ref9d_salsa6 carries the identical frame), knowledge/motion_demos.rows (ref9d_192 with the referee's own numbers), demo assets knowledge/motion/demos/ref9d_192.nxa (10,758,232 B, the 192-cell rig banked from the referee run) and ref9d_192.png (TEXVIEW-OK nv=14164). First compile found two let-reassignments in the lib, fixed by CAS edit; the recompile is in flight. Still owed before the door opens: green gate, promote and register the three organs, deploy_targets.conf and daemons.reg rows, the proxy_routes.conf row, nx_page_verify on the live page.
2026-09-06MO3landnx_motion_rig PROMOTED 2026-09-06 (45,559 B, sha256 7b8b6a30caf0f66165cd9eea8f107d7e8c4d8aad20efdfda3424726d65adf92e, prev=1), REGISTERED on the MCP surface and installed to _offc (OFFC-INSTALL-PROMOTED installed=1). This is the job behind the door: rig <jobdir> [cells] converts through nx_gltf2mesh or nx_fbx2nxa, rigs through the promoted nx_autorig_mesh, renders through nx_nxa_texbake view, exports through nx_mesh2glb, hashes through nx_filehash, and writes status rows at every stage with done last; anim <jobdir> <donor> <pose> <key> retargets and bakes a clip. NOT YET PROVEN BY ITS GATE: nx_motion_serve_gate is written and its build is queued behind admission (run queue 14 of 8); the daemon build was refused the same minute and is paced. The presence of the binary is a receipt, not a proof.
2026-09-06MO3measureFIRST REAL JOB through the promoted nx_motion_rig (house mesh, 1,014,256 B bare NXA, 96 cells, 2026-09-06): rig OK in 16,928 ms -- limbs 7, joints 111, branches 6, verts 14164, tris 28092, GLB exported -- and two stages refused BY NAME, each a composition fact the first run measured: nx_nxa_texbake view exit 4 (TEXBAKE-REFUSE no TEXC section, run nx_nxa_texc first) and nx_filehash exit 127 (the elf lives at the serving root, not in _offc). Fixed at the roots in the job organ (an elf resolver that tries _offc then the root; a UV-unwrap stage before the render) -- rebuilt 54,168 B sha bf4d6fd7, staged, promote in flight. The gate was renamed nx_motion_gate (it proves the door AND the job, and the async launcher's deny-by-class bound refuses any name segment starting with 'serve' -- debt 1787686197, a THIRD measured recurrence after nx_deployjrnl; the correction row 1787707354 already records that the guard is correct and the declaration file is knowledge/status/organ_kind.conf, where all three organs are now declared). nx_motion_serve_gate.nx could not be retired -- nx_retire_path protects any name carrying 'serve' (same class) -- so it stands superseded beside nx_motion_gate.nx; the rebuild under the new name is owed.
2026-09-06MO3landnx_motion_rig v2 PROMOTED 2026-09-06 (54,168 B, sha256 bf4d6fd72809f31a72f30a0b76bf1e5564b2641b46acefbf6974a70797566f2c) over v1 7b8b6a30. The promote guard first refused CAPLOSS 76 permil (10 of 131 live tokens absent); nx_contentdiff enumerated every lost run and all ten were the eight _offc/nx_<organ>.elf path literals the new elf resolver replaced (lost_srcpath=0 lost_symbolname=0 lost_other=10, gains rig_tex.nxa, log_texc.txt, _tex.nxa), so the loss was intended and named and the promote was re-sent with allow_capability_loss=yes. No matched-vintage control was needed: the cause was read from the diff, not inferred. _offc twin re-installed from the promoted root.
2026-09-06MO3measurenx_motion_gate (staged 94,756 B decb5cb9, run as nx_motion_gate_staged through nx_job_run, 2026-09-06): 43 of 44. Every ruler tooth passed on planted inputs -- typing by magic incl. the PNG negative control, the multipart parser incl. the cut-body refusal, form and query decoding incl. the truncated-escape refusal, path shape incl. traversal refusals, line-anchored status rows, the report-line fields, both renderers (no script tag, UNGRADED and ABSENT rendered, the cap printed from the one constant) -- and the box fixture rigged end to end through the promoted nx_motion_rig v2 in 1,658 ms: limbs 2, joints 12, mixed_verts 8, rig.glb 3,124 B, rig_view.png 331,814 B (the UV-unwrap-then-render stage now works), a 64-hex digest, and the OBJ control refused at stage convert with 415 and done written. The one FAIL was the determinism tooth, and the cause was the GATE's fixture, not the pipeline: the length pointer was re-read after run a's status read had overwritten it, so run b's box was written truncated and refused at the skeleton stage (its log: no valid VERT section). Fixed by pinning the fixture length once with a fixture-reached-the-condition tooth beside it, and T17 split so a second run that never rigged cannot be read as a digest mismatch. Rebuild under the new name and the re-run are next; the real-mesh determinism check (the house mesh rigged twice into two job directories) is running beside it.
2026-09-06MO3measureDETERMINISM ON THE REAL MESH, 2026-09-06: the house rig's bare mesh (1,014,256 B, 14,164 verts, 28,092 tris) rigged twice through the promoted nx_motion_rig v2 into two different job directories (1788706300 and 1788706301) at 96 cells printed ONE rig digest, e5fa2f5497c07a351fe8a981f8797e005600897b37c994b03c8f860e153fc601, with identical counts (limbs 7, joints 111, branches 6, mixed_verts 14164); wall time 26,089 ms cold and 4,189 ms warm, so the cost was the box's page cache, not the arithmetic. This is the mo_rig_digest row's claim met by measurement on a real character, beside the gate's box.
2026-09-06MO1measurenx_motion_gate v2 (95,011 B, sha256 005cfbdf10ee5adbbd800d9f72c7a083dc3256d4ebda7962e89db3a021ee7d19) staged and run as nx_motion_gate_staged: 46 of 46 GREEN. The two fixture-reached-the-condition teeth added after the first run (the box read back with the bytes the writer reported; the second run rigged with exit 0) both pass, and T17b now proves determinism honestly: the same box rigged twice through the promoted job organ prints one 64-hex digest. Emitted values: box_limbs 2, box_joints 12, box_mixed_verts 8, rig.glb 3,124 B, rig_view.png 331,814 B, 1,738 ms, upload cap 8,388,608 B. The static fallback was emitted by the staged daemon's own emit verb (MOTION-EMIT OK bytes=5602 sites/nishifamily/motion/index.html, sha 4a684627): the same renderer the live page will serve. Promote of the gate and the health-checked deploy of the daemon are in flight; the supervisor row, the route row and nx_page_verify on the live page remain.
2026-09-06MO1landSHIPPED AND GATE-PROVEN, PUBLIC DOOR BLOCKED, LIVE UNTOUCHED. nx_motion_gate PROMOTED+REGISTERED+_offc (95,011 B 005cfbdf, 46/46 GREEN incl. the real-mesh determinism tooth), declared oracle; nx_motion_rig v2 PROMOTED+REGISTERED+_offc (54,168 B bf4d6fd7), declared oneshot; the static fallback emitted by the daemon's own emit verb (sites/nishifamily/motion/index.html 5,602 B 4a684627); deploy_targets.conf and daemons.reg rows added and daemons.reg re-signed. BLOCKED: /api/deploy nx_motion_serve ROLLED-BACK NEVER-LISTENED twice, so the serving root has NO nx_motion_serve.elf (catalogue BUILT-UNPROMOTED) -- the never-brick rollback fired correctly and LIVE is untouched. The startup cause is UNCAPTURED: the sanctioned launcher for a daemon is /api/deploy, no deploy spawn-log is exposed under knowledge/status or /tmp, and a daemon cannot be nx_job_run (jlb_denied) to capture its own stderr, so I cannot read the LISTEN-FAILED/listening line. Two hypotheses, neither hammered further under the active I/O storm (procs_blocked 5-6): (a) another listener already holds loopback 18132 -> EADDRINUSE at bind (nx_http_probe returned 301 on BOTH 18132 and 18133, which means it is answered by the host redirector and is NOT a valid bind test either way); (b) a startup crash. THE PERMANENT FIX, rung MO14: a bind-smoke TOOTH in nx_motion_gate that forks the daemon binary on a throwaway port proven free by nx_port_audit, connects for a 200 on /motion, and kills it -- turning 'does the daemon listen' from a deploy gamble into a gated repeatable check, and capturing exactly the stderr the deploy hides. Until MO14 or a calm window: the daemon source and staged .elf.new (92,397 B 18205859) stand ready; one clean /api/deploy off-storm, or a port off tier_ladder.conf proven free, opens the door. Everything the surface needs is built and proven; only the listen remains.
2026-09-06MO14measureDAEMON PROVEN HEALTHY, DEPLOY PROBE IS THE FALSE NEGATIVE (2026-09-06, rung MO14 core shipped as nx_motion_smoke, built 22,906 B 8731c7fe, registered staged). nx_motion_smoke forks the daemon binary on a throwaway port with gk_run_capture_ms and captures its own stdout+stderr: on port 28139, on 18132, and on the installed serving-root ./nx_motion_serve.elf, ALL THREE print '[nx_motion_serve] listening on loopback' and are killed by the 2.5 s watchdog (rc=137). So the binary binds 18132 in under a second and the two /api/deploy NEVER-LISTENED rollbacks are the health-probe missing the REFUSED->ACCEPT edge under load -- the documented deploy-lane fragility, not this daemon. Bring-up: the proven binary is installed at the serving root (nx_motion_serve.elf 92,397 B, mode 0755) and daemons.reg carries a signed revive row, so the fleet supervisor -- which does NOT use the deploy edge-probe -- starts and keeps it. /api/restart has a FIXED service allowlist that does not include nx_motion_serve, and nx_http_probe answers the host redirector's 301 on any loopback port, so neither nudges nor verifies it; the live URL waits on the supervisor cycle. MO14 finishes when nx_motion_smoke also does the HTTP GET /motion round-trip (200 + the upload form) and becomes a tooth in nx_motion_gate, so the daemon's listen is gated repeatably and a NEVER-LISTENED deploy is diagnosable from the gate.
2026-09-06MO10queueWHAT CLOSES THE WEIGHT-TOPOLOGY GAP (operator asked 2026-09-06 whether a code miner helps). The gap is top-1 weight agreement ~233/1000, a WEIGHTS-quality gap, not a skeleton gap (coverage 923). A rival SOURCE miner is LOW leverage here: SkinTokens' method is already published and stable (Skeleton Tree Tokenization + Bone-Point Cross-Attention + an FSQ skin-token VAE, read from its README) -- a standing source miner earns its keep only where the frontier is opaque and churning across many repos, which is exactly the langintel lane for compilers (nx_langintel_gh_fetch/llvm_ingest/rust_ingest feed /compare/lang from 1,651 banked oracle bugs). Ranked leverage: (1) FIX THE REFEREE'S METRIC FIRST -- top-1 LABEL agreement penalises us for naming a joint differently even when the deformation is identical; the honest metric is posed-DEFORMATION agreement (md_measure on a played pose), which is degenerate on ref9d because its one pose moves zero joints, so get a real posed comparison (a donor with a POSE library or a synthesised swing); cheap, and it may show the true gap is far smaller than 233. (2) THE ORACLE-IMAGE LANE (MO10) -- run SkinTokens/UniRig published WEIGHTS in an nx_container image the estate spins up and down, feed the same upload, score BOTH with our referee on a shared licence-clean corpus (MO11); this turns their marketing number into our measured delta, the consensus the progress law requires, and it is the operator's own 'repos as container oracle images' direction generalised estate-wide. (3) A SOVEREIGN LEARNED WEIGHT-REFINER (modding MD4 autorig_humanoid, a Local-model rung) trained on our OWN rigged corpus plus CC0 sets, no third-party checkpoint in the build path, with the pre-declared accept rule that it must beat the geodesic control on the referee. The investment that actually helps is therefore NOT a code miner but (a) the estate-wide ORACLE-IMAGE contract on /compare so any published rival becomes a measured competitor, and (b) a licence-clean ASSET miner to build the training corpus (Mesh2Motion-assets CC0, Quaternius, MakeHuman, VRoid), each carrying a modding-MD9 provenance row. The oracle and its weights stay OUTSIDE the build and run path by construction.
2026-09-12MO9queueOperator acceptance 2026-09-12: one user action and matching API/MCP workflow must ingest a referenced asset, discover body/outfit/hair dependencies, preserve source units/materials/provenance, recover and qualify reference motion, fit interchangeable catalogue garments with physical tightness controls, qualify hair appearance/dynamics and publish a reviewable Beach candidate without root-agent manual coordination. Current implementation does not satisfy this end-to-end requirement. Dark Elf Beastmaster https://modbooru.com/mods/dark-elf-beastmaster.2182 is observed as outfit/accessory listing with body variants; full pictured character and hair inclusion unknown. Linked Nexus189114/189845 returned content-preference/login restriction; archive contents and permissions unverified.
2026-09-12MO9measureReference strut.gif SHA256 2d17175184e088f93cff9f165644c93f0a9bbfb77757ae154d28e5602b747811: browser ImageDecoder36 frames268x402 over1800000us; retained knowledge/gates/strut-gif-reference-20260912.json. Native nx_gif_decode currently first-image-only and skips timing/transparency/disposal; differential repair under review, not delivered motion recovery. Public GIF cannot establish hidden geometry or physical parameters.
2026-09-12MO11lessonFrontier qualification requires dated representative comparative results, not newest publication date or symbol existence. Do not interpret historical roadmap training proposals as training authorization. Character workflow success is intermediate; photoreal Beach requires separate terrain/material/light/sky/water/vegetation and contact evidence plus actual served visual review.
2026-09-23MO3landDETERMINISM TEST REPAIR ACCEPTED 2026-09-23: the prior RED 43/44 did not prove nondeterminism because the repeated fixtures were not byte-identical (run A 592 B SHA 23336a627f1ff562a3d954e538f213070891790d69ef0262207d5fa7dc60a119; run B 238 B SHA f8c979d6ab879ea59e4cc97249c7d07b3dd8e9c79aeb7bbd52f1067e3c485b50). The canonical source now pins the fixture length and gates T16h byte identity. Staged job 1790174357, promoted live job 1790174780 and compatibility alias job 1790174894 each passed 45/45 with T16h and T17 GREEN. Source SHA 3001731f29587b8626b08d819c62e12fb243863c9da03aabd50f399afa3d4140; ELF SHA 9bfbd8fcbe8525f340c1247360627019c238e9fcb196db0f602065b300325131; retained receipt buildroot/knowledge/gates/motion-determinism-20260923.json; PM correction triaged as 1790175588. Test-gate and release-record work only; no visual or production-motion change.
2026-09-23MO3measureTORRENT GATEWAY REQUEST-LIFETIME REPAIR QUALIFIED 2026-09-23 in the canonical owner (not installed or promoted): source buildroot/runtime/_hdl_build/nx_torrent_gw.nx 14,246 B sha256 25d4f820068ec70812bd70ff27c02b5f56fd84638ac1ddb037ad482e3f5c5184; shared dependency buildroot/runtime/nx_sha512.nx 19,081 B sha256 bc430061bf7f335f6c0d262520fd6586ecc8d1e834da9f3a3bccb0fa31242fd7. Root cause was per-request arena scope plus Ed25519 and SHA-512 schedule mappings that survived request return; canonical owner now closes request scope/arena, wraps Ed25519 scratch and places sha512_compress W[0..79] in a nested scratch frame. Candidate build job 1790177568 produced buildroot/_build/nx_torrent_gw.sov.elf 167,256 B sha256 c595f401a7128ae351cdfa013daf14e2617bacc2047f85387962d192d6c66f70. Isolated gate job 1790177585 passed 15/15 over 30 repeated mixed requests with VmSize/RSS delta 0; extended job 1790177609 passed 15/15 over 100 repeated mixed requests with VmSize 1,808->1,808 KiB and VmRSS 1,500->1,500 KiB, both deltas 0. Gate source 9a639c5859bc1e32e7775815f19fd299620924dc58697bad4dd93e2356dc6a6f and gate ELF a28a32aa276bc5cbb3a87a7205ae7373b6c7233f6e82890d133f86518361c0e6. Prior RED job 1790177354 was 13/15 at 10 rounds and 360 KiB growth; this causal delta is the accepted evidence. Live PID 21376 was not restarted; publisher/registration remains a separate owner action after source review.
2026-09-23MO3reclassifyThe preceding 1790177609 torrent lifetime row is retained as historical cross-reference only; torrent gateway/resource evidence belongs canonically in buildroot/knowledge/compare/loadgov.plan. Motion owns only the dependency cross-reference: nx_request_scope lifetime contract is consumed by nx_torrent_gw. No history was deleted.
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/motion.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 motion, 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).

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 18 capabilities: 16 exclusive to it and 2 shared with other boards. Shared rows on this page: 2 = home here 0 + home elsewhere 0 + ambiguous 0 + undecided 2 + refused 0 (partition 2); undecided = unmapped 0 + unproven 2 + unevidenced 0; homonyms 0; since the previous census: new 0, resolved 0. Census FRESH age_s=7199.

CapabilityOrganBoards grading itHomeRuleEvidenceStanding on this board
m_jointsnx_bvhmotion.nxmangagen, charsim, graphics, koikatsu, gameengine, motion—noneunprovenundecided
mg_convert_nxanx_mesh2glb.nxmodding, motion—noneunprovenundecided

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 1396304 s), sources read 4 of 7 declared — a BLIND census is a FLOOR: unread sources can only add rows.

kindnameboardsourceevidence
directivenx_langintel_gh_fetchmotionplan-queuerung=MO10

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.
  1. [mesh2motion] Mesh2Motion, Scott Petrovic (site read 2026-09-06): Animate Animals, Monsters and Humans -- a 3D animation library and auto-rigging tool for web and game engines; over 150 animations; imports GLB, GLTF, DAE and FBX; animation rigs for Birds, Dragons, Spiders, Kaiju, Foxes and Humans with intuitive skeleton positioning and undo; exports multiple animations packed into a single GLB; code MIT, animations CC0, human animations partly from Quaternius; runs in the browser with no LLM in the tool. publisher · read in our library knowledge/fetched/cmp_motion_mesh2motion.html · pin ha0dec1a6181356965f86913710d218291536d878c6cb588b7388b259ba8f4b53 · accessed 2026-09-06 · vendor-docGrounds: SURFACE: upload a model in the browser and get it rigged
  2. [mixamo] Mixamo (Adobe), Wikipedia article read 2026-09-05 by the modding lane and reused here: an online Auto-Rigger described as the industry's first online rigging service, plus a library of motion-capture animations retargeted onto the uploaded character. publisher · read in our library knowledge/fetched/cmp_modding_mixamo_wiki.html · pin hf280b3e1d4e03af6ac01325d88e57757d49babb7f7862b8a22e3e9e971f952ca · accessed 2026-09-05 · mirrored-indexGrounds: MOTION: apply a library clip to the fresh rig
  3. [tripo3d] Tripo (VAST) product page read 2026-09-06: Auto-rig with AI in seconds for any humanoid characters and avatars; rigged characters ready for an auto-animation pipeline like Mixamo; Auto-Animate; turn concept art into 3D characters, creatures and props, auto-rigged and ready to animate right away; the hero upload accepts png, jpeg and webp images, so the entry is an image, not a mesh. publisher · read in our library knowledge/fetched/cmp_motion_tripo3d.html · pin h68d9bfc1ceccc5feedc2155bb81796a49047ee8dab04673231c619d81c0a1671 · accessed 2026-09-06 · vendor-docGrounds: RIG: skeleton and weights from the volume with no archetype hint
  4. [vastai] VAST-AI-Research GitHub organisation page read 2026-09-06: pinned TripoSG, TripoSF, MIDI-3D, ComfyUI-Tripo, tripo-mcp and tripo-python-sdk; also TripoSplat, AniGen, StereoWorld, LegoACE and the Rust, JS, Java and Go SDKs; UniRig and SkinTokens are the rigging line. The generation repos belong to the generation boards, not this one. publisher · read in our library knowledge/fetched/cmp_motion_vastai.html · pin hfd09dff6b9631cc6f9e824c12b35d08944f3c0c3de9970b90edf0422d64818eb · accessed 2026-09-06 · mirrored-indexGrounds: RIG: skeleton and weights from the volume with no archetype hint
  5. [arxiv_autorig_2026_09] arXiv API listing fetched and read 2026-09-06: search auto-rigging OR automatic rigging OR skinning weights OR skeleton extraction OR rigging and skinning, sorted by submission date descending, 40 of 141 results; newest entries Make-It-Poseable v3 (2512.16767, updated 2026-09-02) and Kirin (2609.01823, 2026-09-01); newest skinning-weight paper A Geodesic Cut-Cell Prior for Neural Skinning (2608.11272, 2026-08-11); SkinTokens (2602.04805) dated 2026-02-04. This is the current-month bar source the operator's rule requires. publisher · read in our library knowledge/fetched/cmp_motion_arxiv_autorig_2026_09.xml · pin h6402b93cc09bfbab9027a57ac0cab32b2cbd8167ca60557560382d35a4c72c0d · accessed 2026-09-06 · mirrored-indexGrounds: RIG: skeleton and weights from the volume with no archetype hint
  6. [skintokens] SkinTokens, VAST-AI-Research, README read 2026-09-06: unifies skeleton prediction and skinning into a single autoregressive sequence via learned discrete skin tokens, yielding 98 to 133 percent improvement in skinning accuracy and 17 to 22 percent improvement in bone prediction over state-of-the-art baselines; the TokenRig model is trained on ArticulationXL 2.0 plus VRoid Hub plus ModelsResource and refined with GRPO; weights published on Hugging Face. publisher · read in our library knowledge/fetched/cmp_motion_skintokens.html · pin ha4a596b7707973dbd9a8d49414ac5aedc0596e3aed1a07741e6eb4dfbdd34539 · accessed 2026-09-06 · source-readGrounds: RIG: skeleton and weights from the volume with no archetype hint
  7. [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 by the modding lane, mirror reused): two stages, an autoregressive transformer predicting a topologically valid skeleton through Skeleton Tree Tokenization, then Bone-Point Cross Attention predicting per-vertex skinning weights; inputs obj, fbx, glb and vrm; the README announces SkinTokens as its successor. publisher · read in our library knowledge/fetched/cmp_modding_unirig.html · pin h8d77b9eca87d2e91030ccd214a4e39027f3fe86b3ff706689af3bfa61ec6dad6 · accessed 2026-09-05 · published-paperGrounds: REFEREE: the rig grade travels with the surface
  8. [cascadeur] Cascadeur (Nekki), site read 2026-09-06 from its own licence text: computer software for physics-based character animation, distributed as a desktop application; an animation authoring tool with AI-assisted posing, not a mesh rigger. publisher · read in our library knowledge/fetched/cmp_motion_cascadeur.html · pin he56cc7b28b440001d3e1054139efd8c07375c15d5aaa7edc469b1f5e3fa33ca4 · accessed 2026-09-06 · vendor-docGrounds: MOTION: apply a library clip to the fresh rig
  9. [accurig] AccuRIG (Reallusion ActorCore), witnessed 2026-09-05 by the modding lane and again 2026-09-06 by this one: the vendor's auto-rig page answers a JavaScript single-page-app shell with no static prose (3,291 B) and the manual URL answers a Reallusion not-found page, so NO MIRROR IS CITED -- a stub is worse than none. What the vendor states through its app: a free one-click character auto-rigger for humanoids and quadrupeds exporting rigged FBX to Mixamo, iClone and the engines. Its only artifact is a closed desktop binary, which is not ingestible and cannot be an oracle image; it stays a rival column, never a source. publisher · accessed 2026-09-06 · vendor-docGrounds: RIG: skeleton and weights from the volume with no archetype hint
  10. [arxiv_physchar_202609] arXiv API listing, (character OR humanoid) AND (physics-based OR physically based) AND (animation OR motion), newest first, 20 entries, read 2026-09-16: InstantMimic 2609.09821 and CHIP 2609.06591, both SIGGRAPH Asia 2026 publisher · read in our library knowledge/fetched/cmp_motion_arxiv_physchar_202609.xml · pin h9421d3fa7995fca122eac1471158a607d90eea4964732f8942c8cefa7a9d1583 · accessed 2026-09-16 · mirrored-indexGrounds: SB-PHYSCHAR-2609
  11. [meshy_feature_20260919] Primary read2026-09-19; scope and execution limits in motion-competitor-acceptance-20260919.json publisher · read in our library knowledge/fetched/rf_8ba8e1c2.raw · pin ha902a0083c364362d9446f478ba6ad5a9f41302ef76d00b8b8a7b9c423ca6f3d · accessed 2026-09-19 · vendor-docGrounds: Candidate rig-motion-export acceptance; no competitor run
  12. [mesh2motion_home_20260919] Primary read2026-09-19; scope and execution limits in motion-competitor-acceptance-20260919.json publisher · read in our library knowledge/fetched/rf_a82828aa.raw · pin ha0dec1a6181356965f86913710d218291536d878c6cb588b7388b259ba8f4b53 · accessed 2026-09-19 · vendor-docGrounds: Candidate rig-motion-export acceptance; no competitor run
  13. [meshy_api_20260919] Primary read2026-09-19; scope and execution limits in motion-competitor-acceptance-20260919.json publisher · read in our library knowledge/fetched/rf_e3123506.raw · pin h41c7382857a1ddcf46d9a40a4cb5f4d813659e78d7bbf4dbaa802dec0cbf063a · accessed 2026-09-19 · vendor-docGrounds: Candidate rig-motion-export acceptance; no competitor run
  14. [mesh2motion_repo_20260919] Primary read2026-09-19; scope and execution limits in motion-competitor-acceptance-20260919.json publisher · read in our library knowledge/fetched/rf_2cd4b336.raw · pin h597dd159d6b69493579a4aa9a6f7ee9fcf89c6d80fc8d977c67965a8c1e1e215 · accessed 2026-09-19 · vendor-docGrounds: Candidate rig-motion-export acceptance; no competitor run
  15. [mesh2motion_assets_20260919] Primary read2026-09-19; scope and execution limits in motion-competitor-acceptance-20260919.json publisher · read in our library knowledge/fetched/rf_4ed31b28.raw · pin h27025a120582a5606eac68942a05e8b5bc3a75cb7e4a6ac6ffaca0ef830fb652 · accessed 2026-09-19 · vendor-docGrounds: Candidate rig-motion-export acceptance; no competitor run
  16. [mesh2motion_desktop_20260919] Primary read2026-09-19; scope and execution limits in motion-competitor-acceptance-20260919.json publisher · read in our library knowledge/fetched/rf_8c76fa72.raw · pin hbcfc628c477988e1560e216a0452897a497c3e03ae79806124fe8fef2f3faab0 · accessed 2026-09-19 · vendor-docGrounds: Candidate rig-motion-export acceptance; no competitor run
  17. [meshy_mcp_repo_20260919] Primary read2026-09-19; scope and execution limits in motion-competitor-acceptance-20260919.json publisher · read in our library knowledge/fetched/rf_a5936747.raw · pin hc31c0474d3f4c2dc61de39ec3adb90bf249d34d9e2eb2e946debf1096a90a118 · accessed 2026-09-19 · vendor-docGrounds: Candidate rig-motion-export acceptance; no competitor run
  18. [mesh2motion_loader_20260919] Primary read2026-09-19; scope and execution limits in motion-competitor-acceptance-20260919.json publisher · read in our library knowledge/fetched/rf_aa49b7d0.raw · pin hfe4b43978a49a6898e12f6028bce3423696b66be5070a8467563d27b90b82b42 · accessed 2026-09-19 · source-readGrounds: Candidate rig-motion-export acceptance; no competitor run
  19. [mesh2motion_export_20260919] Primary read2026-09-19; scope and execution limits in motion-competitor-acceptance-20260919.json publisher · read in our library knowledge/fetched/rf_73b2a0a5.raw · pin h6b5bbda141416bc54b22bda2ee59a3db8a72e9a602aaef07fc6d5ec538b29b43 · accessed 2026-09-19 · source-readGrounds: Candidate rig-motion-export acceptance; no competitor run
  20. [mesh2motion_customclip_20260919] Primary read2026-09-19; scope and execution limits in motion-competitor-acceptance-20260919.json publisher · read in our library knowledge/fetched/rf_7700c5cc.raw · pin hee74d707bdacfdade8e50a01a03666bc2688d776ff93fc6496b20e11fa17049e · accessed 2026-09-19 · source-readGrounds: Candidate rig-motion-export acceptance; no competitor run
  21. [meshy_mcp_source_20260919] Primary read2026-09-19; scope and execution limits in motion-competitor-acceptance-20260919.json publisher · read in our library knowledge/fetched/rf_cbbc8259.raw · pin h4ae443fab02d0efe12f6cf9f7e40b14c4e6376ea6319be86dd0db541b437165c · accessed 2026-09-19 · source-readGrounds: Candidate rig-motion-export acceptance; no competitor run

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