Aesthetic Digital Twin

Inspect a rung and its prerequisites

Declared nodes 49. Rank input binding: STALE_INCONSISTENT. Dependency order is authored. Implementation, acceptance evidence, authority and resource readiness are unverified. No action is recommended or dispatched here.

Plan SHA-256 4bade160452a70f1e00aa9032f729c969f583a1fe9c40aa6ed4a92a7994f6500. Target rows 3; role rows 49. Existing risks and release worklog retain their own scope; no node completion is inferred.

Runtime observation coverage: 0 of 49 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. AT1 — Evidence-graded aesthetic target rows

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

   Contract: bp_evidence_grade

   Acceptance: Every aesthetic target the canon carries becomes a row of knowledge/aesthetic_canon.conf with region, marker, band, evidence class from REPLICATED or CONTESTED or REFUTED, and the key of a source pinned in aesthetictwin.refs, consumed by nx_body_proc beside body_dimorphism.conf rather than by a second loader. Done-rule: a gate proves every row resolves its key against the refs register or is REFUSED by name, that a REFUTED row may be an axis to measure but is refused as a steering target, that the counts of the three classes SUM to the row population so the grade is a partition and not a headline, and a neg-control proves a row citing an unknown key does not load. The first rows written are the ones the matrix header already graded: adiposity REPLICATED [@tovee1999], averageness and female-face femininity REPLICATED [@lee2025], WHR 0.70 CONTESTED [@voracek2002] [@katzmarzyk2001] [@hubner2024], symmetry CONTESTED [@pentonvoak2001] [@vandongen2012], phi REFUTED [@naini2024], neoclassical canons REFUTED as prescriptions [@farkas1985].

   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.

2. AT2 — Outline curviness metric

   Prerequisites: AT1 (acceptance unverified)

   Contract: am_outline_curvature

   Acceptance: A curvature-based measure over the silhouette nx_anatomy_measure already extracts, reported beside WHR and never instead of it. Done-rule: a gate proves the metric ORDERS two silhouettes of equal WHR and different curviness the way the source's raters did [@hubner2024], that it is invariant to uniform scale, and a neg-control proves a straight-sided silhouette reads zero. CORRECTION FILED 2026-08-28 AND IT IS AGAINST THIS RUNG'S OWN DELIVERY: am_outline_curvature, shipped 2026-08-28, DUPLICATES nx_curvebench.cb_measure, which has existed since 2026-08-07 and already computes the integrated absolute SECOND DIFFERENCE of a half-width outline profile, normalised to figure height and emitted BESIDE waist, hip and WHR -- which is verbatim this rung's own pre-declared done-rule, including the never-instead-of-WHR clause. Two agents reached that finding independently, and nx_capsearch ranks cb_measure at rank 2 for the capability with corpus_complete=1, so the incumbent was retrievable before a line was written and was not retrieved. That is a RETRIEVAL FAILURE, NOT A DISCOVERY, and it is recorded here rather than tidied away because the ranker was simultaneously reporting this rung as LIB-GATE-ONLY, i.e. the estate was carrying a duplicate ruler AND an adoption gap on the same symbol. WHICH RULER SURVIVES IS DELIBERATELY NOT DECIDED HERE: the proposed consolidation was itself refuted for repeating the same check-before-build error, and note that two independent implementations AGREEING on one fixture is the STRONGEST evidence of duplication rather than a refutation of it, so a comparison tooth must not be written the other way round. The adjudication belongs with the nx_curvebench owner. ADJUDICATION SHAPE SETTLED 2026-08-29 BY READING BOTH CONTRACTS, without usurping the owner: they are ONE METRIC OVER TWO SUBJECTS -- cb_measure extracts the half-width profile from an EMITTED MESH in 128 stature-normalised bands, am_outline_curvature extracts it from a RASTER MASK in 16 height-normalised segments, and the bend arithmetic between them is the duplicated part. Neither subject should die: the mesh leg is what the generator aims at and the mask leg is what an image referee reads. The consolidation is therefore a SHARED PROFILE-INDEX FUNCTION taking a width-profile array, with two extraction front-ends that each own their representation -- one arithmetic home, two honest subjects, and AT2's ordering done-rule then proves the shared arithmetic ONCE instead of twice. Also load-bearing and worth carrying here: cb_measure's own header is where the muse doctrine lives -- what a generator needs from a reference figure is an AIM POINT ON A METRIC, not pixels; a stored image is a liability with an expiry date, a curvature target is a parameter -- which is the licensing-clean form of every reference-model conversation this board has.

   Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

3. AT3 — Averageness against a population mean face

   Prerequisites: AT1 (acceptance unverified)

   Contract: fa_averageness_axis

   Acceptance: Derive the mean face from the canon's own dimorphism rows sampled over seeds, then a landmark distance from it. Done-rule: a gate proves the mean face scores zero, that the score is monotone in morph amplitude for each of the seven face morphs, and that a distinctiveness the source found to LOWER attractiveness [@lee2025] [@little2011] reads as a higher distance rather than being folded into a beauty number.

   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.

4. AT4 — Canthal tilt and lip ratio as measured axes

   Prerequisites: AT1 (acceptance unverified), AT16 (acceptance unverified)

   Contract: pm_canthal_tilt

   Acceptance: Tilt in degrees from the medial and lateral canthus landmarks of the emitted mesh, banded by the source's classes [@lee2020canthal]; vermilion heights, the upper-to-lower ratio and the lip-area-to-lower-third ratio reported against the witnessed 1 to 2 and 9.6 percent figures [@popenko2017], the latter landing on fa_lip_ratio in the same organ. Done-rule: a gate proves both axes are read from mesh landmarks and not from slider positions, that a mirrored face reads the same tilt with opposite sign, and a neg-control proves a face with no eye landmarks reports UNMEASURED rather than zero. DONE-RULE DEFECT CORRECTED 2026-08-28 AND IT WAS VACUOUS AS PUBLISHED: the mirror tooth above, same tilt with OPPOSITE SIGN, is satisfied by the degenerate constant zero because minus zero equals zero. On the geometry this rung was ranked against, tilt was identically zero for every face by construction, so a stub returning zero for a present eye and UNMEASURED only when the eye parts are deleted satisfied BOTH published teeth while measuring nothing whatever. That is a completion signal keyed on a name, and it sat at the head of this board's critical path. ADDED AND MANDATORY: a tooth that plants a KNOWN tilt in the lid geometry, asserts the measured tilt tracks it within tolerance, AND asserts it is NOT ZERO. That tooth is unsatisfiable without a palpebral aperture, which is precisely why AT16 is now a declared prerequisite and ranks ahead of this rung rather than behind it.

   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.

5. AT16 — Palpebral fissure so that a canthus exists to be measured at all

   Prerequisites: AT1 (acceptance unverified)

   Contract: fa_palpebral_fissure

   Acceptance: PROMOTED FROM A WATCH ROW TO A RUNG ON 2026-08-28 BECAUSE THE RANKER RANKS RUNGS AND COULD NOT SEE IT. AT4 is ranked FIRST on this board at priority 1666 and is NOT BUILDABLE, so the board's own critical path has been pointing at unbuildable work. A canthus is the commissure where the UPPER and LOWER eyelid margins MEET; faceanat_build emits 27 parts whose only eye-region rows are 16 and 17 carving sockets, 18 and 19 full eyeballs, and 25 and 26 lashes. There is no lid margin and no aperture, so one of the two boundaries whose junction DEFINES a canthus has no geometric representative and there is nothing for any measurement to find. WORSE THAN ABSENT, THE TEMPTING PROXY IS VACUOUS: parts 16 and 18 are axis-aligned and share cx exactly, so their intersection curve is mirror-symmetric and a tilt read off it is identically zero for every face forever -- and an axis that cannot fail is not an axis. This also means AT4's pre-declared mirror tooth, that a mirrored face reads the same tilt with OPPOSITE SIGN, passes VACUOUSLY on the current geometry because minus zero equals zero, so a stub returning zero would satisfy the declared done-rule while measuring nothing. Done-rule: an upper and a lower lid-margin part per eye whose medial and lateral ends MEET, so medial-canthus and lateral-canthus landmarks exist as surface features resolvable by marching the blended field off-midline; a gate proves the two landmarks are DISTINCT points, that they MOVE when the lid geometry moves, and a neg-control proves a face with the lid parts removed reports UNMEASURED by name rather than zero. SECOND LAYER, PRICED SEPARATELY AND DELIBERATELY NOT FIRST: sdf_ellip_rot is PITCH-ONLY, rotating the y and z pair and passing px through untouched, so even correct lid geometry stays axis-aligned and would read zero; O_ROLL was reserved in the arena for exactly this and today is declared at two lines with NO setter and NO reader, so the expression half is half-landed and unwired. Wiring roll FIRST would yield a measurable-looking number with no anatomical referent, which is worse than a refusal.

   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.

6. AT15 — Lip protrusion against the Ricketts esthetic plane

   Prerequisites: AT1 (acceptance unverified)

   Contract: fa_eline_offset

   Acceptance: Both vermilions measured against a line from the MEASURED pronasale to the MEASURED pogonion, read off the blended surface so the axis cannot be satisfied by editing a slider. This rung was SHIPPED BEFORE IT WAS PLANNED and the row is being written after the fact, which is recorded here rather than tidied away: it was mislabelled AT5 in its own gate for one session, and AT5 is the neoteny axis. Done-rule: a gate proves the offset moves when the LIP moves and, independently, when the CHIN moves, which is what separates a line from a depth and which a ruler that had degenerated into lip depth would pass on the first and fail on the second; that a mirrored face reads identical offsets, the profile being a midline quantity; and a neg-control proves a face absent from the sampling band refuses BY NAME rather than returning the most flattering reading available. Raw units are published beside the millimetres because the mm conversion divides by MMU, and MMU disagrees by ~2.2x with four independent witnesses of this model's own declared geometry, so a consumer reading only mm is reading a contested constant.

   Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

7. AT5 — Neoteny and maturity blend axis

   Prerequisites: AT4 (acceptance unverified)

   Contract: fa_neoteny_axis

   Acceptance: OPERATOR DIRECTIVE 2026-08-28 AND IT IS THE BINDING CONSTRAINT ON THIS RUNG: SEXUAL MATURITY IS THE FLOOR. The axis measures the juvenile-to-mature continuum and the generator half is bounded so nothing below an adult subject is emitted. THE MECHANISM IS DELIBERATELY NOT pg_check, AND THAT WAS MEASURED RATHER THAN ASSUMED: pg_bound resolves only 8 BODY axes and returns -1 for every face axis, so pg_has_bound reads 0 and pg_check returns PG_NO_BOUND unconditionally -- a floor routed through it would pass on every possible input and prove nothing, which is a checkbox wearing a control's name and is worse than no tooth because its existence is counted as coverage. The floor therefore lives IN THIS AXIS: the emit path refuses BY NAME below the floor, with a named reason code like every other refusal in this family, never a silent clamp that rewrites its input and reports success. AND THE FLOOR CANNOT BE DEFINED AS GROWTH COMPLETION: craniofacial structures are never static, and the jaws and dentition undergo subtle LIFELONG positional change [@cfgrowth2025], so there is no age at which facial form finishes and any floor phrased as growth having stopped is unmeasurable by construction. The floor is defined by ADULT MARKERS BEING PRESENT on the emitted face, measured, never by an age number. THE SUB-FEATURES ARE CONTINUOUS MORPHOLOGY AXES AND ARE DELIBERATELY NOT ETHNIC TYPES (operator 2026-08-29, who named the soft rounded look and explicitly rejected the population-label framing). A three-bucket phenotype classifier would be abandoned racial typology rather than a measurement, it is scientifically wrong because craniofacial variation is clinal and within-group variance dominates between-group, and a system that infers ethnicity from a face is a PROHIBITED biometric categorisation practice under the EU AI Act, so it would also fence this board out of a market it is meant to serve. The continuous axes give strictly MORE variety than buckets, which is the point. Each is a knob the model already has or needs, listed with its part rows: malar and mid-face soft-tissue volume (faceanat_morph_cheek, present); lower-third proportion, subnasale to gnathion over total face height (NEEDS a per-third y-scale morph, which is the one generator symbol this rung actually adds); brow-ridge projection, parts 8 and 9 cz, flush rather than deep-set; lateral zygomatic flare, parts 6 and 7 cx; nasion depth and bridge length, part 4; nasal tip rotation, part 5; inter-eye separation, parts 16 to 19 and 25 to 26 cx; and cephalic index, which is ALREADY a canon row graded CONTESTED as an AXIS with no target [@hamdan2025]. fa_neoteny_axis SCORES a face on this combination and does not drive it: the morphs are the gate's FIXTURES and the axis is the ruler, which is why the done-rule below asks the two extreme morphs to score at opposite ends. An eye-area-to-face-area ratio, a lower-third height and a zygomatic projection combined into one axis on which a child-like and an adult-like morph land at opposite ends, carrying the measured sex differences in lower face height and palpebral fissure height [@babacan2025] and the trait definition the review literature uses [@lidborg2025] [@perrett1998]. Done-rule: a gate proves the two extreme morphs score at opposite ends, monotonicity along the blend, and that the axis is reported as a number with its inputs printed rather than as a verdict.

   Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

8. AT6 — Cervicomental angle of the neck

   Prerequisites: AT1 (acceptance unverified)

   Contract: bp_cervicomental

   Acceptance: The angle derived from hyoid and menton landmarks of the emitted body, monotone in the submental fat parameter, with the surgical anatomy of the axis carried as citation and no target [@ellenbogen1980] [@abdulshakoor2021]. Done-rule: a gate proves the angle falls as submental fat rises, that it is read from landmarks the canon already places, and a neg-control proves a body with no hyoid station reports UNMEASURED.

   Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

9. AT7 — Rib-cage floor under the waist

   Prerequisites: AT1 (acceptance unverified)

   Contract: bp_thoracic_floor

   Acceptance: The waist minimum DERIVED from the chest coefficients and the bone layer so a waist below the skeletal envelope is REFUSED by name and one at the floor emits. Done-rule: a gate plants a sub-floor waist and proves the refusal fires, proves a waist at the floor emits with the bone layer intact, and a neg-control proves the floor moves when the chest coefficients move rather than being a typed number.

   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.

10. AT8 — Referee calibration across render fidelity

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

    Contract: cjp_tier_of

    Acceptance: LANDED 2026-08-27. The panel carries a fidelity tag per image (silhouette, clay, textured, photoreal), floors may be declared per tier, a lens votes only with a floor about that image's tier, admission is earned per tier and the receipt binds the tier set, the admitted subset and the per-tier floors. Proven by nx_cjc_panel_gate 19/19 (T13 additive, T14 cross-tier refusal, T15 ratchet, T16 receipt binds the tier set, T17 unknown tier refused by name) and bite-proven killed=1 not_reached=0 with both deployed twins restored byte-identical. ⚠THIS DONE-RULE WAS CORRECTED AGAINST ITS OWN EVIDENCE RATHER THAN GRADED AGAINST THE WORDING IT SHIPPED WITH. It demanded that the existing single-tier receipt be reproduced BIT-IDENTICALLY; that is unachievable by construction because the receipt carries an epoch, and it was also the wrong thing to ask -- a receipt is a summary, and reproducing it says nothing about the verdicts consumers actually read. What was proven instead is strictly stronger on the part that matters and is stated as what it is: the FULL panel output over the live 13-row corpus is unchanged value for value, vote for vote, composite for composite and verdict for verdict, with only the new tier fields added, measured by banking the live run BEFORE the promote and re-running after. The receipt gains three tier lines and its epoch moves; nothing else in it changes. A done-rule that cannot be met should be corrected out loud, not quietly met with a weaker proof.

    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. AT9 — Licensed marker-vector oracle for a named model

    Prerequisites: AT1 (acceptance unverified), AT2 (acceptance unverified), AT4 (acceptance unverified), AT10 (acceptance unverified)

    Contract: rp_marker_oracle

    Acceptance: Behind a resolved licence chain and consent record, compute a named model's marker vector (waist-to-hip, curviness, leg-to-body, canthal tilt, lip ratio, lower-third height, skin ITA) from licensed reference captures, publish a band per marker, and report per marker whether an emitted body lands in-band. private-ref:b7a71b7a is the named oracle class; the estate holds no licensed capture, so the oracle is a refusal today. Done-rule: a gate proves refusal by name without a licence row, proves under a synthetic licensed fixture that the vector is computed and the in-band verdict flips with the morph, and proves the vector carries the evidence class of each marker from AT1 so a CONTESTED marker cannot decide the benchmark alone.

    Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

12. AT10 — Person licence chain and media classes

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

    Contract: al_person_chain

    Acceptance: BUILDS ON the phototwin lane's landed al_license_chain (chain row with owner, rights, term, territory, royalty and three named exits) rather than re-implementing it: the widening is the SUBJECT as owner of her likeness and the MEDIA CLASSES (film, game, VR room, print) as separately licensable rights rows, with the emit-for-publication path refusing without a resolved chain. Done-rule: a gate proves a publish path refuses by name without a chain, that a chain missing any conjunct names the missing one, and that the measurement path still runs so the oracle lane is not destroyed by the fence.

    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.

13. AT11 — Gynoid fat pooling and ptosis

    Prerequisites: AT7 (acceptance unverified)

    Contract: sd_fat_pool

    Acceptance: Fat that pools, sags and spreads on contact for the gynoid distribution -- waist, flank, hip, thigh and breast -- so the brief's contouring markets become a distribution parameter on a body that behaves. Shared with the procgen board. Done-rule: a gate proves volume conserved through pooling, that a higher fat parameter lowers the breast lower-pole point monotonically against the incompressible tissue truth [@breasttissue], and a neg-control proves a zero-fat body is bit-identical to today's emit.

    Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

14. AT12 — Male evidence rows in the canon

    Prerequisites: AT1 (acceptance unverified)

    Contract: bp_shr_axis

    Acceptance: The male body already emits at sex factor zero; this rung writes its graded rows: shoulder-to-hip ratio REPLICATED as more attractive when larger [@pazhoohi2023], leg-to-body about 0.5 standard deviations above baseline [@versluys2018], arm-to-body no effect [@versluys2018]. Done-rule: a gate proves a shoulder-to-hip axis is read from the acromion and trochanter stations, that the male rows carry classes and keys under AT1's schema, and that no female-row target leaks into the male emit.

    Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

15. AT13 — Animals and fantasy through the same canon

    Prerequisites: AT1 (acceptance unverified), AT12 (acceptance unverified)

    Contract: bp_species_canon

    Acceptance: A second species and a stylised human through the identical canon, generator and emitter path, shared with the procgen board. The aesthetic canon for a non-human is an OPERATOR DECLARATION of which human markers survive; this rung refuses to invent one. Done-rule: a gate proves the species canon loads through the same loader, that a fantasy character's graded rows are a declared subset of the human rows rather than a new table, and a neg-control proves a species with no declared rows emits with no aesthetic verdict at all.

    Authored effort: 4. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

16. AT14 — Reverse-direction referee for the licensed oracle

    Prerequisites: AT9 (acceptance unverified)

    Contract: rp_bench_report

    Acceptance: A monocular mesh recovery used as an ORACLE ONLY: recover a metric body from one licensed photograph [@multihmr2] in the Anny representation [@anny], compute the same marker vector from the recovered mesh, and publish the disagreement between the two readings beside the benchmark, so a fidelity claim about a named person carries its reference uncertainty. Done-rule: a gate proves the recovered and emitted vectors are compared per marker, that the oracle-to-emit disagreement is a distinct published figure never folded into the score, and that the learned model is invoked only on the measurement path and is ABSENT from every emit path by the same fence as AT10.

    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.

17. AT17 — The model scale is DERIVED from the population survey, never typed

    Prerequisites: AT1 (acceptance unverified)

    Contract: fa_mmu_x10

    Acceptance: MEASURED 2026-09-02, and it is the unit contradiction this board has carried on every millimetre since 2026-08-28: nx_faceanat declares MMU = 35 (3.5 units per mm) from a comment that took a 150 mm HALF-width for a skull that is about 145 mm WIDE, while the model's own landmarks against ANSUR II female means (interpupillarybreadth 516 u, headbreadth 1040 u, bizygomaticbreadth 980 u, the menton-sellion span) imply about 7 to 8.5 units per mm -- a 2.2x error that makes every published mm a factor of two wrong and every tissue relief drawn through fa_tissue_z about half as proud as the cited depth, which is one MEASURED cause of the painted-flat face the operator rejects. Done when (a) nx_ansur_regress's mean verb publishes the female means for the witness columns with n and the publisher's population bound (verb landed in source 2026-09-02); (b) knowledge/faceanat_scale.conf carries one WITNESS row per landmark pair (model units, ANSUR mean mm, n, refkey ansur2) and the derived units-per-mm with its SPREAD; (c) fa_mmu_x10() returns the derived value beside the witness spread, the E-line and every other mm this organ prints convert through it, and the legacy 35 is renamed and quarantined to the relief path until the relief re-baseline is accepted visually -- doubling every relief silently would change the emitted face under gates that were green on the old one, so that is a separate rung with its own acceptance; (d) nx_faceanat_scale_gate recomputes the derivation from the conf, asserts the shipped literal sits inside the witness band, asserts the spread is published, and its neg-control asserts the legacy 35 sits OUTSIDE the band -- the tooth that could not have passed before this rung

    Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

18. AT18 — The composite prototype is PROCGEN: landmarks, shape mean, texture mean and PCA modes computed from a population, never hand-sculpted

    Prerequisites: AT3 (acceptance unverified), AT9 (acceptance unverified)

    Contract: fa_composite_prototype

    Acceptance: OPERATOR 2026-09-02: photorealism is procgen, not hand-crafting, outside benchmarks and oracles; use the research to identify what places the named top-list oracle class where it is, as DATA. The published bar [@asjbeauty2023]: a 2,870-face aesthetic database from 15 years of top-100 lists against a 10,000-face baseline, PCA over 158 parameters, clusters uncorrelated with race, a shared aesthetic proportion that differs from the baseline. Done when ONE sovereign organ (a) places the estate's landmark set (the face canon's own named points, 80 to 150 of them) on every input face by rule, (b) computes the Procrustes-aligned shape mean and the thin-plate-spline warp to it, (c) averages texture in the aligned frame through nx_perf-class integer arithmetic, (d) computes the PCA modes over the population and publishes each mode's variance share, and (e) writes the prototype and its modes as plane rows so AT3 measures a face as a distance to the prototype and AT9 measures the licensed oracle class as a cluster against the general baseline -- with a neg-control prototype from a permuted landmark set that must NOT match, mixed-population prototypes as data rows rather than a claim, and every population row carrying its consent or licence class (nx_realperson fails closed on a named person without one)

    Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

19. AT19 — The facial-proportion canon is a TABLE OF HYPOTHESES the ruler tests, never a rulebook it enforces

    Prerequisites: AT18 (acceptance unverified)

    Contract: fa_canon_table

    Acceptance: OPERATOR 2026-09-02 supplied the hard-math canon (vertical thirds, rule of fifths, facial index, intercanthal index, nasofrontal and nasolabial angles, nasal index, lip and mouth ratios, mandibular angle, sex-specific frame). Landed as DATA in knowledge/facecanon.conf: 30 rows, 2 PINNED to a read primary [@pallett2010] (eye-mouth 360 permil of face length, interocular 460 permil of face width, and the paper's own finding that these are the AVERAGE face's ratios, with little support for the golden ratio), 28 BRIEF rows that cite nothing until their primary is fetched and read. Done when nx_faceanat reads the table instead of typing canon positions, computes every row from its landmark set, publishes per-row distance-to-band for the licensed top-list population [@asjbeauty2023] AND the general baseline, and writes status=refuted on any row the top-list population itself violates -- the Farkas control: a canon most attractive faces fail is a description of a statue, not of the population we are fitting. Neg-control: a permuted landmark set must fail the two PINNED rows.

    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.

20. AT20 — Operator-tagged ingestion ranges: the beauty bands come from faces and bodies the operator flags at ingestion, fitted by the same procgen machinery

    Prerequisites: AT18 (acceptance unverified), AT19 (acceptance unverified)

    Contract: at_operator_ranges

    Acceptance: OPERATOR 2026-09-02: the operator will flag or tag faces and bodies at media ingestion as part of the genetic research; those tags are DATA (accept-plane rows, full-path prefix, actor operator) and the ranges are FITTED from that tagged population through AT18's landmark, shape-mean and PCA machinery, then published beside the licensed top-list ranges as a second population column. EXPLICITLY LAST-MILE BY OPERATOR ORDER: this rung must not derail the engine, graphics and physics gap; it is placed in M6 at the end of the version ladder so the ranker never puts it on the critical path ahead of them. Done when the ingestion lane writes a tag row per flagged asset with the tagger's identity and consent class, the fit refuses populations under a declared minimum n, and the published band carries its n and its population name.

    Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

21. AT21 — Longitudinal reference: one declared person across decades fits the ageing trajectory of the canon axes

    Prerequisites: AT22 (acceptance unverified)

    Contract: rp_age_series

    Acceptance: OPERATOR 2026-09-16 (example given: a declared model photographed from her early career to today): a reference series of ONE person across ages lets the ruler fit how each canon marker moves with age (skin, fat distribution, ptosis, facial thirds, body ratios) instead of assuming a static canon; feeds anatomy AN10 (lifespan canon) and AN12 (ageing prediction). Done-rule: the intake accepts a name plus DATED media, landmarks each item (nx_facemark, nx_headcrop), fits per-marker slopes against a published ageing band, refuses an undated item by name, and publishes the slopes as aggregate rows only; beta-usable, production gated on the licence row.

    Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

22. AT22 — Reference intake as an estate capability: name in, mirrored media with licence state, landmarks and canon vectors out, aggregated into ranges

    Prerequisites: AT20 (acceptance unverified)

    Contract: ri_discover

    Acceptance: OPERATOR 2026-09-16: build the collection into the estate and let it run; grow what is missing from the first byte. Composes nx_research_fetch and nx_model_get for the mirror (licence state REVIEW beside the bytes: beta-usable, production-refused), nx_facemark and nx_headcrop for landmarks, the AT19 canon table for axes, rp_range_canon for aggregation, and the operator tags of AT20 for the bands. The intake measures faces and bodies on DECLARED MORPHOLOGY AXES (malar projection, bizygomatic-to-bigonial ratio, midface height, canthal tilt, facial thirds, waist-to-hip and bust ratios); it carries NO ethnic, racial or religious label and refuses any such tag by name, because those are not measurable axes and the estate does not classify people by them. Done-rule: one declared name yields mirrored media with licence state, per-item landmark vectors (intermediate, never published) and a range contribution; a planted undated or unlicensed item is refused by name; a planted group label is refused by name (neg-control).

    Authored effort: 4. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

23. AT23 — Twin card: every value on a declared axis with side, unit, method and source; face axes are the canon ids; features per freckle; rulers censused

    Prerequisites: AT22 (acceptance unverified)

    Contract: tc_card_check

    Acceptance: The card the operator asked for beyond an Illusion card (a PNG of about a hundred unitless sliders with no provenance). knowledge/twincard.axes declares 14 sections and 114 values: 21 paired axes recorded per side, 13 region charts for one-row-per-freckle feature rows, 11 ruler rows naming the organ and symbol that measure each axis; nx_twincard check prints coverage, the method partition and the completed left-right pairs, and refuses by name (an undeclared axis, a group label, a value outside its band, a class word not declared, a side that contradicts the pairing, a duplicate, a bad feature, a shot naming no ref). Done-rule: nx_twincard_gate GREEN on the NAS and a card for a declared subject passing check. First card 2026-09-16: private-ref:b7a71b7a, 43 valid values, 0 measured, 11 of 21 pairs, 377 permil coverage.

    Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

24. AT24 — Video intake through the clean-serve lane: KVS player pages, the function/0/ hash permutation undone, the clip mirrored and classified by its ftyp magic

    Prerequisites: AT22 (acceptance unverified)

    Contract: ri_kvs_real_url

    Acceptance: OPERATOR 2026-09-16: thothub pages through clean-serve. The transform is gate-proven against an outside-oracle known answer (nx_refintake_gate 72/72); the live clip fetch is the referee and the fetcher body reserve is the open question. Feeds the gs_avatar_from_video watch.

    Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

25. AT25 — Tetrahedral volume mesh from the layered anatomy SDF (bone, muscle, fat, skin as nested boundaries)

    Prerequisites: AT23 (acceptance unverified)

    Contract: sdf_tetmesh

    Acceptance: The substrate every tissue solver needs: nested signed-distance layers (nx_skullsdf, nx_myo, nx_fascia, nx_skinwrap already emit surfaces) meshed into tetrahedra with per-layer material ids. OpenVDB [@sota0916_openvdb] and the Infinigen layered procedural volumes [@sota0916_infinigen] are the oracles; a mesh whose layers do not partition the body volume is refused by name.

    Authored effort: 6. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

26. AT26 — Sovereign hyperelastic FEM step: Ogden and Mooney-Rivlin constitutive models on the tet mesh, implant as a rigid or soft body

    Prerequisites: AT25 (acceptance unverified)

    Contract: st_hyperelastic_step

    Acceptance: The soft-tissue reaction to an implant or a reshaped bone, computed from the first byte on the AT25 mesh. Referees outside the build path: FEBio [@sota0916_febio] for the constitutive laws, SOFA [@sota0916_sofa], SlicerTissue CAPS [@sota0916_slicertissue] and SkinFlaps [@sota0916_skinflaps] for the surgical cases; the done-rule is a displacement field within a stated tolerance of the FEBio oracle on a declared fixture.

    Authored effort: 8. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

27. AT27 — GPU tissue kernel: the same step lowered to the WGSL compute lane, thousands of variants in parallel

    Prerequisites: AT26 (acceptance unverified)

    Contract: wg_tissue_step

    Acceptance: NVIDIA Warp [@sota0916_warp] is the September 2026 bar for GPU-native tissue physics; the estate lowers ONE source to the WGSL lane (scale from a sensor to a supercomputer) and the CPU step is the bit-exact referee.

    Authored effort: 6. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

28. AT28 — Surrogate predictor trained on FEM runs: instant post-operative shape prediction with a stated error against the solver

    Prerequisites: AT27 (acceptance unverified)

    Contract: st_surrogate_predict

    Acceptance: The PINN class the field uses for consultation-speed prediction, trained on AT26 runs over AT25 cohorts through the sovereign nofloat engines; it ships only with its error against the solver printed beside every prediction.

    Authored effort: 8. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

29. AT29 — Landmark-driven face morph: alter declared canon landmarks on a fitted head mesh and re-measure every card axis

    Prerequisites: AT23 (acceptance unverified)

    Contract: fm_morph_by_landmarks

    Acceptance: 3D-face-morph [@sota0916_facemorph] is the open research tool; the estate morphs on nx_faceanat channels and nx_facemark re-measures, so a rhinoplasty proposal is a card diff, not a picture.

    Authored effort: 5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

30. AT30 — Sovereign face and eye boxes from a still: the Viola-Jones engine grown on the incumbents (integral tables, rect features, variance-normalised stumps and trees, feature scaling, grouping) reading the published cascade schema as DATA [@haar_face_default] [@haar_eye]

    Prerequisites: AT22 (acceptance unverified)

    Contract: ff_find

    Acceptance: BETA-ONLY BOOTSTRAP ARM: the cascade files are third-party data under a permissive licence [@haar_face_alt2] [@haar_licence], mirrored beside their licence and never a product path; the engine is sovereign and the same flat layout takes a cascade the estate trains itself. Done-rule: on the oracle-judged beta set the face box contains the oracle canthi and nose and each eye box its iris on at least 12 of 15 stills, gate GREEN with a flat-field neg-control

    Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

31. AT31 — Outside landmark oracle rows as the liar-killer for the photo route: a face-mesh model run OUTSIDE the estate over the mirrored bytes, its canthi, iris, lip and nose points landed as intake data (private-ref:496d5e6c;slug>.oracle) with the derived axes the ruler prints

    Prerequisites: AT30 (acceptance unverified), AT22 (acceptance unverified)

    Contract: pm_referee

    Acceptance: The oracle never enters a build or run path; only its numbers do. A sovereign referee verb reads the rows and scores every measured canthus as permil of the oracle's outer-canthal span (the field's NME shape). Done-rule: the referee prints per-still NME and a partition (FOUND, PLACED-OFF, UNMEASURED, UNJUDGED) that sums to the row count

    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.

32. AT32 — Sovereign learned landmarker: canthi, iris centres and lip points regressed by the estate's own trained model (the nofloat training lane) on oracle-labelled beta references plus the estate's own renders, replacing the licensed cascade in every product path [@arxiv_facealign_202609]

    Prerequisites: AT31 (acceptance unverified), AT23 (acceptance unverified)

    Contract: pm_learned_landmarks

    Acceptance: CONTENDER against SB-facelandmark: the September listing's methods are learned heatmap regressors; ours must land within the referee's NME bar on the beta set before NishiOS or the browser carries any landmarker, and the cascade arm AT30 is retired from product paths on that day

    Authored effort: 8. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

33. AT33 — One file door for the reference still and the estate's own render: the photo ruler reads a PNG (nx_anat_sov output, any sovereign render) through the same entry as a JPEG still and announces the codec, so ONE ruler measures the reference and the twin and a twin-fit loop can score a render against the card

    Prerequisites: AT4 (acceptance unverified), AT30 (acceptance unverified)

    Contract: pm_png_rgb

    Acceptance: The twin loop needs the same instrument on both sides of the comparison; a ruler that reads only the photograph can never score the render. Done-rule: the gate proves the fixture encoded through the sovereign PNG writer and read back through the file door reproduces the in-process canthi, tilts, ratios and stomion EXACTLY, announces codec PNG, refuses a non-image file by name and reports an absent file unreadable; the first estate render measured through the door is journaled with its honest verdict

    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.

34. AT34 — The twin loop's first rung, the reachable axis envelope: a face built from explicit morph percents over the inside-out canon face, rendered by the sovereign SDF renderer and measured IN-PROCESS by the same photo ruler that reads the reference, swept gene by gene to its declared floor, middle and ceiling; per card axis the envelope and a target's REACHABLE or UNREACHABLE verdict, which names the morph the generator lacks

    Prerequisites: AT33 (acceptance unverified), AT30 (acceptance unverified)

    Contract: tw_reach

    Acceptance: The fit loop stands on this question and on nothing else: which of the reference's scale-free axes can a render reach at all under the morph basis the generator exposes today. Done-rule: the gate holds the target reader to a planted file with a bogus row, the reach predicate to both sides of its band, the level arithmetic to the declared ranges, the seed-42 render to in-process measurement, byte-identical re-render and PNG-door parity; the envelope over the seven persongen genes is journaled with its partition and its verdict, PARTIAL when the ruler cannot measure a render

    Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

35. AT35 — The estate's own face render read by the photo ruler and judged by the outside oracle: nx_twinface renders a persongen identity to a PNG through the file door, the ruler measures it as it measures a photograph, and the outside landmark oracle on the same render is the render-side liar-killer, so every ruler change is judged on the photo referee AND the render referee

    Prerequisites: AT33 (acceptance unverified)

    Contract: tw_render_png

    Acceptance: The render is the first subject whose true geometry is known by construction (a symmetric canon face), so it exposes placement bias no photograph can: the first render read the medial canthi 12 px high on a 110 px span. Done-rule: the ruler measures the render (face, both eyes, mouth), the oracle row for the render is intake data, and the referee's NME on the render is journaled beside the photo referee's; the accept bar for the ruler on renders is the same FOUND bar the photo referee uses

    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.

36. AT36 — Media submission to the vault and through the beauty rulers with a report the operator can check: any still (JPEG, PNG), GIF, APNG and video (MP4 with H.264, VP8 keyframes) submitted through the estate's own doors lands in the vault with a content id, every frame the estate decodes is judged by the same rulers (the photo ruler's canthal and lip axes, skin sampling, bilateral symmetry), and the annotated frame publishes beside the numbers and the outside oracle's row as an asset page on nishifamily.com

    Prerequisites: AT33 (acceptance unverified), AT31 (acceptance unverified), AT24 (acceptance unverified)

    Contract: mj_judge_file

    Acceptance: OPERATOR 2026-09-17: submit media to be stored in the vault and reaccessible, run through the beauty evaluator, with output to confirm and triangulate, and not just photos but gif and video. Incumbents, all sovereign: nx_content_put (the laptop door) and nx_vault_capture (a URL) bring bytes in, nx_mvault classify gives the content id, nx_gif_decode, nx_apng_decode, nx_mp4_demux with nx_h264_decode_frame and nx_vp8_kf yield frames, nx_photomark, nx_skinsample and nx_image_beauty judge, nx_asset_page publishes a judged manifest as the three-tier page. Stills are the first honest unit for GIF and video because the meta-analysis puts static against dynamic ratings at r = 0.70 [@pmc_static_vs_dynamic], and the live-streaming benchmark [@livebeauty_2501] is why the dynamic residual is journaled per frame rather than hidden. Done-rule: a submitted JPEG, PNG, GIF and MP4 each yield a judge manifest with one row per sampled frame, an overlay PNG per judged frame drawn by the estate's own PNG writer, a page verified by nx_page_verify, and the oracle row beside ours whenever one exists

    Authored effort: 5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

37. AT37 — Outside beauty oracles and a licensed calibration set for the beauty axes: the published regressors run OUTSIDE on the laptop over the same stills as liar-killers (AttractiveNet [@attractivenet_repo], the deep-network and landmark studies [@pmc_expression_dcnn] [@pmc_fbp_landmarks] [@frontiers2025_deepgeofusion], the 2008 per-user regression [@whitehill2008], the press-reported Tehran model [@techxplore2023], the clinical critique of AI beauty scoring [@jcad_ai_beauty]), SCUT-FBP5500's human mean ratings calibrate through beta under its non-commercial terms [@scut_fbp5500_repo], and the current listing sets the bar [@arxiv_fap_202609]

    Prerequisites: AT36 (acceptance unverified)

    Contract: mj_oracle_row

    Acceptance: The field's number is a Pearson correlation against human mean ratings on a non-commercial dataset; the estate's bar is that correlation from a sovereign ruler over a LICENSED set with the outside models as oracles, never as dependencies. Done-rule: every judged still carries our axes beside an outside regressor's score; over a beta calibration set the sovereign axes' correlation with human mean ratings is journaled with its N; a licensed set (the Face Research Lab London Set, published under CC BY 4.0, whose figshare page answered the fetcher with a WAF challenge on 2026-09-17) is fetched through a browser vantage before any production claim

    Authored effort: 4. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

38. AT38 — Body-side oracle for the anatomy fit: SHAPY's body shape and anthropometric measurements regressed from photographs [@shapy_repo] [@muelea_profile] with Human Bodies in the Wild as the ground-truth validation set [@shapy_hbw] over the SMPL-X body [@smplx_site], run OUTSIDE over the reference stills as the liar-killer for the body axes the twin card carries

    Prerequisites: AT36 (acceptance unverified)

    Contract: mj_body_oracle_row

    Acceptance: The face side has its landmark oracle and its render referee; the body side has neither. Done-rule: an oracle row per reference still with height-normalised chest, waist and hip measures and shoulder width, the twin's body fit judged against them the way the face fit is judged against the landmark oracle, and the sovereign body ruler's NME published beside it

    Authored effort: 4. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

39. AT39 — The eye as its own piece: a lid shell whose opening's corners ARE the canthi, six eye genes drawn per seed, and the render's own landmark truth judging the photo ruler per canthus with a signed error

    Prerequisites: AT35 (acceptance unverified), AT16 (acceptance unverified)

    Contract: tw_truth_landmarks

    Acceptance: OPERATOR 2026-09-17: the optimal in all our work is three different systems agreeing within a statistically acceptable range, refined without overfitting; if proc gen fails at a larger size, generate pieces at smaller sizes and integrate; some left-right symmetry within reason; hand crafting only where it lands as math the generator reuses. MEASURED CAUSE: two ellipsoid lids have no margin, so the canon face had no recessed aperture. The generator's own fissure scan read NO-APERTURE on 19 of 21 genomes and a tilt of minus 77 degrees on another, nx_fissure_gate read 1 of 8 and had never been promoted or run on any beat, and the photo ruler paired the eyes on 6 of 21 renders. Done-rule: (a) a renderer primitive, op SDF_OP_LIDSHELL, a spherical shell over the globe with an almond opening (the intersection of two discs through both canthi, tilt = the part's roll), that leaves every arena without the op byte-identical; (b) the face seats one shell per eye from canon rows (palpebral fissure height 11 mm, intercanthal index 400) and the width and length morphs MOVE an eye unit without stretching it; (c) six eye genes (tilt, opening height, width, upper share, left-right tilt and opening differences) with ranges declared once, drawn per seed AFTER every earlier draw so existing identities keep their morphs and tone; (d) landmark truth = the shell's stated corners projected through the renderer's own camera, CONFIRMED by a surface scan in a window around each stated corner with the gap published; (e) a judge that scores the photo ruler per canthus with signed dx, dy and NME against that truth over the gene sweep and persongen seeds, partition summing; (f) gates GREEN on both hosts with a no-lid-shell neg-control that refuses by name. OPEN beyond this rung: the outside oracle on renders as the third system with seeds 9 to 16 held out (queued as the next rung), the lips as the next piece (the generator has no complete lip truth even on the canon face), brows as hair rather than a dark slab

    Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

40. AT40 — The first fit: solve the face morphs and the eye genes against a reference card's axes from the generator's STATED landmarks, no render in the loop, a gene that ends on a range limit named, the answer confirmed on the surface

    Prerequisites: AT39 (acceptance unverified), AT34 (acceptance unverified)

    Contract: tw_fit

    Acceptance: The twin loop closes for the first time on the axes the generator can state: once the face states its own canthi, a card's eye axes are a function of thirteen genes (seven face morphs, six eye genes) that costs four lookups to evaluate. Done-rule: a pattern search over the declared ranges (single and paired gene moves, halving steps, a move kept only when the cost strictly falls); a cost that puts every axis inside its error band first and nearer its centre second; tilts and roll taken from milli-pixel projections so the objective does not move in plateaus; the stated corner's depth taken by arithmetic on the lid shell's outer sphere; a gene the card cannot use stays at the canon; a gene that ends on its range limit is NAMED, which is the range the generator would have to widen; an axis the generator cannot state is reported UNSTATED and the verdict is PARTIAL, never quietly dropped; the answer is confirmed once by the surface scan; the gate recovers a planted genome's axes, holds an unreachable tilt to UNREACHED with the tilt gene on its ceiling, and holds the cost to planted arithmetic. OPEN: the lip axis (needs the lip unit), sub-unit opening precision (one half-width unit moves the fissure ratio about 21 permil, the named limit on fitting closer than that), and more card axes as the generator learns to state 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.

41. AT41 — Results home: one page per board, every image click to open, nothing sensitive drawn on load

    Prerequisites: AT39 (acceptance unverified)

    Contract: cres_render

    Acceptance: The board's results file renders to /compare/aesthetictwin/results with zero media outside a click-to-open block; the gate holds a planted bare image, a planted explicit row and a policy file that tries to loosen the floor; the page is regenerated by a seeded plan and a daily clock row

    Authored effort: 2. Executor kind: runtime/nx_compare_results_lib.nx. 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.

42. AT43 — The lip unit: two swept vermilions meeting at the stomion, with construction truth

    Prerequisites: AT4 (acceptance unverified), AT39 (acceptance unverified)

    Contract: fa_lip_stated

    Acceptance: The generator states labrale superius, stomion, labrale inferius and the cheilia in closed form through the renderer's own outline function; on the canon face the surface ruler reads the vermilion ratio from the rendered midline with nothing refused and agrees with the statement within one refine step on each height; the stated ratio is the canon row upper_over_lower_lip within one unit of rounding; a fullness morph holds the stated ratio; lips past the sampled band are refused by name; each of those teeth is bite-proven by a planted mutant

    Authored effort: 2. Executor kind: runtime/nx_faceanat.nx. 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.

43. AT44 — The photo ruler reads each canthus from the eye fissure and the mouth from its seam, refusing every window-side reading by name

    Prerequisites: AT43 (acceptance unverified), AT31 (acceptance unverified), AT39 (acceptance unverified)

    Contract: pm_eye_fissure

    Acceptance: Ruler version 3 of nx_photomark_lib (shipped 2026-09-18, live nx_photomark 927dd890 285,280 B, gate 5d4152ec 79/79): each lateral and medial canthus is the extreme of the eye fissure grown from the pupil instead of the edge of its search band, and a canthus that lands on its window side is refused as PINNED naming the side; the mouth gives five landmarks from the lip seam (pm_mouth, pm_mouth_anchored on the oracle canthi, pm_lip_score) with seamless and short-seam refusals by name. DONE-RULE: nx_photomark_gate GREEN with planted asymmetric canthal tilts and a split mouth read within tolerance and every refusal firing on its decoy, and the referee over private-ref:b7a71b7a publishing FOUND, PLACED-OFF, UNMEASURED and PINNED as a partition that sums. Measured after this rung: FOUND 2 of 16 (mean NME 59 vs 54 before), the two remaining PLACED-OFF stills halved their errors, three stills now refuse as PINNED; the lateral-canthus bias on photographs (L_lat NME 101) is NAMED, not solved

    Authored effort: 3. Executor kind: runtime/nx_photomark_lib.nx. 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.

44. AT45 — Image-level likeness referee: same-person band derived from the reference set, every axis published, the rejected renders read outside

    Prerequisites: AT4 (acceptance unverified), AT31 (acceptance unverified), AT33 (acceptance unverified)

    Contract: lk_score

    Acceptance: Two stills are aligned into one canvas by the similarity that maps each face's two eye centres onto the canvas's (the oracle canthi for the references, the photo ruler's canthi for a render, its refused-pair candidates when that is all it holds, every source named on the receipt) and ten axes are measured between them and published one per line: landmark NME, canthal-tilt delta, the edge-orientation descriptor as L1 and as cosine, skin L a b, the shading layout, the texture layout, the texture energy. The band per axis is the WORST distance among the reference set's own same-person pairs (six pairs of four clothed frontal stills, printed with the pair that set it), the canvas is twice the smallest reference eye span and the grid cell is the set's own landmark spread; a subject passes an axis iff its nearest reference lies within the band, the verdict is the count of axes inside, outside and unmeasured, INSIDE only when no measured axis is outside, no weighted sum and no typed coefficient. Done-rule: nx_likeness_gate GREEN on the real set: band from at least three pairs, a reference against itself zero on every axis, a rolled and scaled copy INSIDE with its geometry within two canvas pixels, the two rejected renders OUTSIDE and measured rather than unaligned, the door agreeing with the in-process path. OPEN BAR carried to the structure contender: the structure family separates 1 of 4 renders today (the same person's photographs spread more on every classical structure ruler than a render differs from a photograph at an 84 px canvas), so the rejections rest on skin chroma and texture layout; the rung is not accepted until a structure ruler reads 4 of 4

    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.

45. AT46 — The skull-bound tissue-depth skin: the skin as an offset surface of the generated skull by the forensic tissue-depth field, controllable fat compartments, and named patches where no bone backs the feature

    Prerequisites: AT17 (acceptance unverified), AT39 (acceptance unverified), AT43 (acceptance unverified)

    Contract: sks_surface

    Acceptance: the skin is the generated skull offset along its own gradient normals by a depth field interpolated (Gaussian RBF, kernel width the median nearest-landmark spacing) from knowledge/tissue_depth.conf at landmarks LOCATED on the skull mesh, plus additive fat compartments whose magnitudes are a printed parameter vector; proven by nx_skullskin_gate: kernel arithmetic against closed forms, no vertex penetrates the skull (depth positive everywhere, min median max printed), neg-control-zero-depth-field (the skin equals the skull) and neg-control-doubled-cheek-fat (only the compartment support moves, partition summed), the photo ruler finds a face with both canthi of both eyes on the skin render and its intercanthal index sits inside the canon band widened by the ruler's own measured precision, and the skin's midline thirds equal within the table's spread (this last conjunct is SUBSTRATE: it reads the skull generator's proportions and is RED today); the NME against the outside oracle is unmeasurable in-lane until a render carries an oracle row

    Authored effort: 2. Executor kind: runtime/nx_tissuefield_lib.nx. 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.

46. AT47 — Auto photo-match: media placed in the vault inbox or tagged in the vault is judged, matched and scored with no seat

    Prerequisites: AT36 (acceptance unverified), AT40 (acceptance unverified), AT33 (acceptance unverified)

    Contract: pmx_beat

    Acceptance: Operator 2026-09-30: "i also want to get an auto ingestion system setup so that i can just place or flag media in the vault/gallery for auto nishi attempts at getting the photo matched via the ssytem". THE LIKENESS POLICY IS THE OPERATOR'S (2026-09-30): the estate reproduces the pictured subject, identity included, and flags every result ESTATE-ONLY; the only use a result has without a grant is a proposal shown privately to the depicted person, and every deployment (VR, cosmetology, fine nude art, a gallery, merchandise, an advertisement, an avatar) is a use scope that person grants under a two-party royalty contract. Done-rule: a still placed in the vault inbox or tagged photomatch:queued is admitted exactly once by content id (a second beat never duplicates it), classified by nx_mvault, judged by nx_mediajudge (AT36), measured and attempted within a declared wall budget and re-measured by the same rulers as the photo (nx_photomark through the AT33 file door, nx_refstill GR62, CIEDE2000 by cs_de2000), its private report carries every score or names it UNMEASURED, the result is flagged ESTATE-ONLY with a proposal-candidate row, and every outward door composes nx_rights_ledger default-deny per exact talent and use scope; proven by nx_photomatch_gate on a fixture vault (the door refused with no grant, opened by a planted grant for exactly that talent and use, refused for another scope or talent) and by a live beat on the mirrored reference with its report produced

    Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

47. AT56 — Attractive body-proportion range: published bands as data, sampled by the generator and measured on the emitted body

    Prerequisites: AT1 (acceptance unverified), AT2 (acceptance unverified)

    Contract: bp_propband_sample

    Acceptance: Operator 2026-10-01: get the right attractive proportions in a range like the glamour publications carry. knowledge/attractive_bands.conf carries one row per ratio with lo, hi, unit, refkey and the derivation from the source's own numbers (whr 650..710 permil from 0.68 +/- 0.03 [@katzmarzyk2001]; bmi 17.3..20.0 from 18.1 - 0.8 up to the stated optimum 20 [@voracek2002]; lbr -0.5..+0.5 sd from the canon row [@kiire2016] [@frederick2010]); nx_propband_lib loads it, refuses the WHOLE table on any malformed or ungrounded row, checks a ratio in nx_proportion_guard's vocabulary and samples inside the band by construction. Done-rule: nx_body_proc draws each body's WHR, BMI and LBR through pb_sample from a per-body hash; a gate emits N bodies from N seeds, measures waist and hip GIRTHS and leg and stature from the EMITTED mesh (girths, not silhouette widths), and every measured ratio lands inside its band with the per-ratio spread across the N bodies covering at least half the band; known-answer teeth per ratio on a fixture body; a neg-control body outside the band is reported outside by name. The lib half is landed and gated (nx_propband_gate); the emitter wiring and the mesh-girth measurement are this rung's open work

    Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

48. AT57 — Duplicate receipt: a reference still reproduced by the estate twin and scored per region and per landmark against declared bars

    Prerequisites: AT47 (acceptance unverified), AT45 (acceptance unverified), AT44 (acceptance unverified), AT31 (acceptance unverified)

    Contract: at_dup_receipt

    Acceptance: OPERATOR 2026-10-03 (ecosystem EC87, the duplicate half): prove we can duplicate an item to have evidence we have sota capabilities if not frontier, starting with digital twinning humans. The estate already holds every piece and no receipt joins them: the photo-match loop (AT47) renders a twin against a vault reference, the image-level likeness referee (AT45) publishes the same-person band, the photo ruler reads canthi and the mouth seam (AT44) with the outside landmark oracle as liar-killer (AT31), graphics GR62 measures per-region CIEDE2000 against a reference class through skintwin ST1 cs_de2000, and the private reference store (private-ref:81f02c0a, licence areas as rows, nx_refarea refusing any public use) holds the originals. Done-rule: for a reference still in the private store, the twin loop emits the estate reproduction and one receipt prints, per declared region (skin, hair, lips, iris and sclera, background), mean and 95th-percentile CIEDE2000; landmark error normalised by inter-ocular distance (the face-alignment field's NME) from the sovereign ruler AND the outside oracle side by side; the AT45 likeness distance and band; the AT56 proportion ratios where a body is in frame; each against a bar ROW whose value comes from a measured human calibration (JND for colour, pair-judge or two-alternative forced choice for the whole frame, cms CM30-class), never a typed threshold. Three method classes (ecosystem EC82): the receipt reproduced from its inputs, the referee arithmetic checked against its pinned sources' own known answers (Sharma 2005 CIEDE2000 test data for colour), and a planted render perturbed by a known amount read back at that amount. Estate-only: the receipt carries the store id and numbers, never an image; no real person reaches a production surface without the EC85 door

    Authored effort: 4. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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

49. AT58 — Variation receipt: one named twin parameter moved by a requested amount, the target axis measured to move by that amount and every other measured axis held

    Prerequisites: AT57 (acceptance unverified), AT56 (acceptance unverified)

    Contract: at_var_receipt

    Acceptance: OPERATOR 2026-10-03 (ecosystem EC87, the variation half): the ability to create variations of the original to show we have control of the math capabilities, not just observation and build capabilities. A duplicate proves the estate can match; only a controlled variation proves it holds the parameters that made the match. Done-rule: starting from an AT57 duplicate, a request names ONE parameter and a delta (first axis: waist-to-hip ratio inside the AT56 band, because it is the one with a measured emitter path today, nx_body_proc propband with the nx_girth_lib girth ruler, gate 24/24; then skin melanin through the skintwin chain and graphics GR59, a joint angle of the pose, and garment tension through softbody contact); the receipt prints requested, predicted and measured for the target axis with the absolute difference of measured and requested inside a declared tolerance row, and a held-axes table (per-region CIEDE2000 outside the region the parameter owns, landmark NME, the other proportion ratios, the AT45 likeness band) each inside its own tolerance. The gate plants a crosstalk mutant (a WHR change that also shifts skin colour or the other ratios), a wrong-magnitude mutant and a request outside the declared band, and proves all three refused by name while a clean variation passes; the same three method classes as AT57. Estate-only, numbers and store ids only

    Authored effort: 4. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.

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