One seam repaired.
A blouse still to build.

V21 had left the pitch junction enclosed by occupied triangle corners. Adding detail nearby had not given the route a way out. V22 planned the routes together instead of letting each one consume space greedily. One conforming midpoint split allowed four compatible paths to reach their anatomical destinations while retaining correspondence to the original body.

That was a genuine routing milestone. The next image also shows its limit: the first emitted atlas still contained 21 reversed triangles. Finding four paths did not make the mapped garment surface valid, and no fitting followed.

V23 removed those 21 coarse mapped inversions using a more exact surface representation. It then exposed a subtler failure: two charts could agree at their stored boundary stations while tracing different physical curves between them. The continuous audit found a gap of about 2.85 mm. The matched shoulder views below show why an endpoint-only check had been too reassuring.

V24 inserted the intervening route corners into the source boundaries and made the incident maps share the same physical curve. Across 84 paired intervals, the saved native checks passed the existing position, tangent and normal limits. The repaired atlas uses 792 vertices and 1,502 triangles; the original pattern remains the material reference.

The exact atlas and the coarse solver interface are different representations. V24 tested whether the old simulation triangles could reproduce the mapped shape faithfully. They could not: the saved-file recheck found a maximum discrepancy of approximately 17.93 mm against a 0.01 mm limit, with 25 reversed faces.

V25 tried one bounded topology candidate. The result improved some counts but still failed: approximately 17.91 mm of transfer error and 15 reversed faces remained. The failed candidate was kept as a diagnostic. It was not fed into cloth simulation, substituted into the character or called a finished garment.

The rest-pattern study made an important distinction. Anatomical landmarks help describe a body, but that does not make each one a physical pin holding fabric in place. Treating every atlas partition landmark as a fixed material attachment created three named endpoint contradictions. Even the apparent 8 mm alternative constrained endpoints to artificial balls; it did not represent general tangential sliding.

V26 classified 2,716 relationships in a separate derivative specification, retaining all 34 real shoulder seam stations. The original pattern, body and historical definitions stayed unchanged. The authored bust point, inherited atlas bust guide and dart were kept distinct rather than silently replacing one with another.

The principal search still failed: 18 of 903 tested endpoint pairs exceeded the upper threshold. Independent review found a feasible combination in the unchanged eligible contact pools. The single bounded repair selected a simultaneous sliding witness with zero upper-threshold contradictions across all 903 tested endpoint pairs. Its largest endpoint-pair distance-to-rest-path ratio was 1.491119×, below the 1.5× screen.

This is a necessary endpoint test. It does not measure the full garment’s strain, prove the 0.5× lower bound, establish ease or collisions, or construct a continuous wearable surface. The sparse colored paths below describe selected contact trajectories; they are not a blouse and are not sewn seams.

V26 stopped before static initialization. The original front shoulder rest length and the placed V15 partner differ by 1.353048 mm, exceeding the existing 1 mm criterion. Their original rest stitchlines agree closely; the mismatch is not evidence that the pattern lacks material. A separately versioned correction of the posterior shoulder’s placed pose was the proposed next step at that cutoff, preserving pattern dimensions, station correspondence and the existing posterior checks.

The solver-interface failure also remains separate and unresolved. No fitting, seam joining, garment export or complete assembly acceptance followed the endpoint witness. Hair experiments during these checkpoints still failed artistic review; the selected predecessor was preserved, with no failed study propagated to the second receptionist.