Closed bug report

LrgMed medium pin fit requires 8 mm spans; 4 mm guidance accepted incompatible molecules

John Lauer · ·closed by John Lauer

Environment: installed adom-aiflow0.1.90; source base5d2acb188de3be4ec3190892cde35e01d1720c57; Linux x86_64. Found2026-10-11. This is an offline geometry/acceptance defect, not a UI or Bridge failure.

The molecule skill said LrgMed medium contacts form a4mm grid. checks.rs only warned for non4mm spans. The actual current adom/scaffold-lrgmed-96x96 circuit.json has256 medium0.78mm holes on8mm pitch. Electronic-load board29 has60x68mm corner spans and has zero valid placements among1024 cardinal rotation/contact-anchor combinations. A64x72mm candidate has224 nominal XY placements. Signal contacts retain their independent2mm grid.

Direct source correction5db21db90b505f2bd35a9e9667a9ef04acb42e48 fixes the skill and rejects incompatible all-medium corner-pin spans as an error. Regression covers60x68 failure and64x72 success, preserving2mm signal placement. Production cargo check/build and152 workspace tests pass; component149-reference, widget, silkscreen contact-policy and obstacle checks pass. New CLI returns2 for actual board29 with: ERROR pin span 60.00 x 68.00 mm cannot seat on the LrgMed 8 mm medium-contact grid It returns0 for the isolated candidate. Native candidate DRC reports0violations/0opens.

Source is pushed and skill readback verified. Installed/registry runtime remains0.1.90; package release/deployment is still required. No hardware gate bypass: board candidate is unadopted; carrier clamps, real fixture Z seating, model frames and thermal/manufacturing qualification remain open. XY span acceptance is not full scaffold-fit qualification. Evidence lives in adom/adom-custom-electronic-load checkpoints/scaffold-fit-review (project publication pending).

2 Replies

John Lauer ·

Environment: installed AIFlow0.1.90; tested source d33b2fbe552bd64ab90b9c0632347fb59e333cc7 on Linux x86_64. Native validation uses KiCad10.0.5 / KiCad Bridge1.0.55. Registry/installed executable unchanged; isolated production build used explicitly.

Follow-up after adopting the reviewed64x72mm interface: the gate still rejected MP2/MP3/MP4 because it compared against the original60x68mm board, even when explicit spec.molecule.fixed coordinates were supplied. Exact error began ERROR fixed references moved since the run started: MP2 (160.00, 100.00, 0 deg) -> (164.00, 100.00, 0 deg).

Direct source fix d33b2fbe552bd64ab90b9c0632347fb59e333cc7 makes explicit MP1-relative positions authoritative while preserving the absolute MP1 anchor, all rotations, untouched references,5micrometre position tolerance,8mm scaffold pitch and other molecule checks. Malformed contracts/missing references fail. Skill now requires an evidence-backed interface correction, not a movement waiver. Original run boards remain unchanged.

Production check/build and154workspace tests pass (one existing native test ignored), plus component149-reference, widget and silkscreen regressions. Real Board32 isolated-run CLI rejects the old spec and accepts the explicit revised contract. Actual scaffold circuit search finds224nominal placements. Live run molecule check0errors/0warnings; native DRC gate0new/0inherited/0unconnected. This is not physical seating or thermal/manufacturing qualification.

Evidence: adom/adom-custom-electronic-load checkpoints/board32-interface-contract and checkpoints/scaffold-fit-review/board32-fit-review.json. Source readback matches both changed files; registry release/deployment remains open.

John Lauer ·

Shipped: 5db21db9 and the follow-up are in every release since 0.1.91 (current 0.1.94). Closing.

John Lauer closed this ·

Log in to reply.