126 poses
The declared pose set stayed at or below 0.031877 mm after re-import.
Inside the workshop · October 23, 2026
126 poses passed.
One crown section still failed.
Four completed checkpoints repaired the lapel transport path and tested two approaches to B’s hair. The numbers moved. The characters did not pass.
Follow the evidence ↓Our previous report repaired the complete visible-character export. V37–V40 used that foundation to test harder questions: could A’s lapels survive the full pose set, and could B’s hair become a believable connected hairstyle without hiding failures inside a better silhouette?
The answers split. A’s lapels now survive the declared transport checks across 126 poses. B’s first whole-hair replacement regressed basic coverage, its second generated thousands of crossing witnesses, and a smaller isolated crown proof still missed its strict geometry limit.
The connected hairstyle closed a hole and opened two regressions.
V37 replaced B’s separated V36 ribbons with a connected whole-hair mass. The edit removed the obvious hollow gathering, but its frozen cranial coverage dropped to 72.519841% against the 95% minimum. Maximum root distance reached 1.706682 mm against the 1.5 mm limit.
The front read as a helmet and the broad nape remained mechanical. V37 was rejected, and V36 stayed selected as the fallback rather than promoting a fuller silhouette with worse measured attachment.
The V37 connected redesign closed the visible shell, but cranial coverage fell to 72.519841% and the maximum root distance rose to 1.706682 mm. The helmet-like front mass remained rejected.
The same rejected V37 checkpoint from behind. The connected nape and gathering remained a broad mechanical mass; V36 stayed selected as the fallback.
The lapels passed natively. The complete export missed by a fraction.
V38 gave A a shared support surface under the lapels. In the native file, the maximum lapel-to-support distance was 0.650717 mm and the explicit sewn-row distance was 0.374221 mm. Those scoped checks passed.
The first complete GLB comparison did not. Its maximum surface discrepancy was 0.065453 mm, above the retained 0.05 mm transport limit. The difference was small, but the gate is exact: a native pass could not excuse a failed derivative.
B’s rebuilt lock system recovered the numeric root and coverage limits, then found 3,865 new finite-segment triangle self-crossing pairs. The count records intersecting triangle pairs, not 3,865 unique visible defects. It still prevented an art pass.
V38 restored the numeric coverage and root limits, then detected 3,865 new finite-segment triangle self-crossing pairs. The front view is evidence of an unaccepted art checkpoint, not a collision pass.
The V38 back view exposes the dense routed locks behind the 3,865-pair diagnostic. Pair counts are intersection witnesses, not unique defect counts, and the hairstyle remained failed.
| Checkpoint | Scoped result | Disposition |
|---|---|---|
| V37 B whole hair | 72.519841% coverage; 1.706682 mm root | Rejected |
| V38 A native lapels | 0.650717 mm support; 0.374221 mm sewn | Native checks passed |
| V38 A complete GLB | 0.065453 mm maximum | Transport failed |
| V38 B locks | 3,865 new crossing pairs | Art failed |
V39 repaired the complete transport path across 126 poses.
V39 corrected the A export derivative while retaining the passing native attachment measurements. The complete import returned a 0.000750 mm neutral maximum and a 0.031877 mm maximum across the declared 126-pose set, below the retained 0.05 mm transport limit.
This is transport evidence, not a final controller or motion qualification. No runtime animation system, GAME or Unreal integration, final drape, or owner appearance review was exercised. B received no new selected art in V39; the exact V36 fallback remained in place.
0.000750 mm
The complete neutral import retained the selected visible surfaces closely.
0 / 11
The technical transport pass did not accept a garment, hairstyle or full character.
A smaller crown proof reduced crossings and still failed.
V40 kept A’s exact V39 transport result and used a saved native mesh-and-lattice edit of one existing B crown section for a narrower geometry experiment. No interactive brush session was performed. The section’s maximum root distance is 1.210775 mm and passes the scoped root limit.
The stricter barycentric sum error is 1.232326e-5, above its 1e-5 limit. Diagnostic self pairs fall from 153 to 82 and pairs against the other major locks fall from 397 to 257. Those are triangle-pair witnesses, not defect counts, and lower counts do not establish collision-free hair.
A’s V40 construction test added connected front fabric beneath the unmoved button and a 2.052 mm sewn shank, while restoring the exact V38 white-collar geometry locally. Transverse facing/jacket crossing pairs increased from 310 to 510, so the garment art still failed.
The V40 B section proof is a temporary diagnostic isolation, not a whole hairstyle or GLB. Its smaller clay render cannot replace a complete-character review.
The retained V36 crown section in the temporary V40 clay-proof setup. This 480 × 600 image is an isolated section baseline, not the complete hairstyle or a GLB.
The isolated V40 crown section reduces diagnostic self pairs from 153 to 82 and other-major-lock pairs from 397 to 257. Its strict barycentric sum error is 1.232326e-5, above the 1e-5 limit, so the proof still fails.
The result: a useful local direction remains a failed checkpoint when its own strict limit is missed.
Keep the transport. Rebuild the visible decisions.
A can retain the 126-pose transport repair while the collar roll, closure and complete garment response continue. B’s next step is to repair one existing section fold or reversal and preserve strict source-triangle correspondence before expanding the groom or bun.
V33 sustained foreground performance remains pending. Full self contact, physical drape, continuous final motion, runtime and mobile qualification, GAME and Unreal integration, and owner acceptance remain outside this evidence.
The smaller proof found a direction. The failed limit kept it honest.
Explore the full character workshop ↗All development updatesEvidence, image and rights notes
This first-party report covers completed R4-V37, R4-V38, R4-V39 and R4-V40 records. Numerical claims are scoped to the saved native, export and focused-result evidence; the page does not rerun Blender.
The six figures are actual unretouched local Blender renders of B. The first four preserve their 640 × 800 full compositions; the two clay crown diagnostics retain their native 480 × 600 compositions. Every 448 × 560 mobile file is a proportional reduction with no crop or upscale. No image generation was used.
An independent component audit qualified these six exact B render hashes for factual public illustration under the retained V36 dependency basis and the recorded V37–V40 changes. This does not grant model or source-asset redistribution. A remains text-only because its visible legacy hair has conflicting license metadata and an independently undeclared image texture. Reference panels, raw assets, comparison composites and the unchanged V39 B fallback are excluded.