Every visible surface arrived.
The art is still unfinished.

V34 reshaped A’s connected jacket without remeshing, raised the hem, suppressed the waist and added editable lapel piping. B’s existing fringe was shortened and moved away from the face. The result improved the black-and-gold silhouette and made more of B’s face visible.

The limits stayed visible. A retained a closed collar, weak closure construction and unresolved garment contact. B still depended on coarse inherited hair islands and unresolved legacy texture provenance. Its corrected whole-hair inward estimate returned only to the inherited 12.845 mm state. V34 ended at 0 of 11 accepted.

A received a recut center neckline, low ivory blouse panels, open collar leaves, notched lapels, an overlap facing, a gold button and bound piping. B’s legacy curtain fringe and ponytail were replaced by five editable scalp, sweep, gathering and framing sections with an original locally generated fiber texture.

The matched B native renders show real changes under retained cameras and lighting. They also show why the model did not advance to review: the regular radial cap, crossing locks and high hollow knot did not form a credible loose updo. A’s progress remains text-only here because its visible legacy hair has conflicting license metadata and an independently undeclared texture.

V35’s narrow changed-mesh checks looked promising: neutral discrepancies were 0.000635 mm for A and 0.000716 mm for B. But a fresh complete-character import exposed unwanted body and helper surfaces. The native evaluated bodies contained 19,370 triangles for A and 22,352 for B; each imported body contained 36,972.

The cause was an export-only mask and helper stack left unapplied. Earlier subset checks also missed unparented static shoes, a belt and trim. The V35 GLBs remain useful failed evidence because they demonstrate exactly why a selected-component comparison cannot certify a whole character. The B counterexample is paired with the repaired import below without substituting the rights-blocked A render.

V36 bakes the visible rest surfaces in an export-only derivative while leaving the editable bodies, groups, masks, UVs, shape keys, weights and 163-bone rigs untouched. It also preserves the authored heel pitch, moves A’s thickness before skinning and retains tiny native weights through export.

The fresh full-scope result contains all 40 visible A meshes and all 33 visible B meshes. Source and imported body triangle counts match at 19,370 for A and 22,352 for B. Maximum neutral surface error is 0.000750 mm for A and 0.000835 mm for B. Across the declared posed checks, the maxima are 0.001568 mm and 0.001263 mm, both below the retained 0.05 mm transport limit.

A’s explicit sewn rows improve from 3.043078 mm to 0.916744 mm and pass their 2 mm scoped check. The stronger all-surface attachment probe still reaches 4.116058 mm at frame 81 and fails. The stiff neck edge, weak white collar construction, closure stump and shoulder creasing remain visible.

B’s root distance is 1.282910 mm and its frozen cranial coverage sample is 99.751984%, but 251 neutral raw body pairs remain. Those counts are diagnostics rather than a collision pass. The scalloped hairline, nape, parting and C-shaped gathering still need authored repair.

Both sources preserve their original bodies and 163-bone rigs. Neither whole hairstyle nor garment is accepted. V36 proves that the model can be transported completely enough to inspect; it does not prove that the transported art is ready.

A’s next correction is the shared support and fold response of the failed lapel surface, followed by collar roll and closure topology. B needs local inspection of temple and root orientation, then rerouted locks and a natural asymmetric boundary rather than another blanket density increase.

V33 sustained foreground performance remains pending. Full self contact, physical drape, continuous final motion, GAME and Unreal integration, runtime and mobile qualification, and owner acceptance remain outside this evidence.