kerfmaster pin list · Nesting
recordStates what the code does today. Auditable against the code — this is the only type that can go stale.
Content last changed 2026-08-31 — computed from the item itself, not typed.
Record. Engine: trial_lattice.py — search theta; unit cell = part at theta (A) + point-reflected pair (B) at offset (bx,by); tile on v1=(p,0), v2=(qx,qy); every contact found by slide-to-contact on the true sampled outline (shapely distance >= gap), never bounding boxes. Sheet fill tries lattice phase offsets. The winner is re-audited pairwise on finely sampled outlines and again on the EMITTED prims in the deliver script (builds 147/157) — the audit gates delivery. Build 158 adds the TOOLED mode (run_lat158_exm275_tooled.py + verify_lat158_leads.py): the outer lead/pierce is part of the arrayed toolpath, so the search also finds a common start point on the outer contour whose pierce keeps the full 0.15 standoff to every neighbour at EVERY placement (leads hunt the meshing voids; a straight-channel layout needs lead+standoff = 0.30 of gap, which is exactly the shop's measured 0.3033). Start-point rank: feasibility, then max worst-case clear-air — safety outranks the lower-right bias (flip-column precedent). The verify pass is independent of the constructor (3A76 lesson): every pierce/lead point vs every neighbour outline, no prefilters. Lesson from the first buggy run (2026-08-26): cell-vs-cell clearance must check ALL interacting lattice offsets, and per-job GAP carries a construction pad (+0.005) over the ruled value so sampling scatter cannot eat the audit floor.
Deliveries: build 147 (3A76-LAT-1X on 36x60 stainless, vs shop NC 24384) and build 157 (EXM275FG-0625 on 36x48 copper: 99 vs shop NC 66261's 94 — measured back out of his PDF: all 94 at 0/180, half-offset rows meshed nose-into-waist, min gap 0.3033, 0.2 margins; our theta=0 run reproduces exactly 94, theta=26 audits 99 at gap 0.2044 vs the ruled 0.2). Un-delivered benchmark: 12729704 114 vs 108 (their 66248).
Recorded gaps: per-job scripts (run_lat_3a76.py, run_lat_exm275.py), not a split_parts mode — multi-type kits still use the shelf packer with its full-bbox column advance; leads are not planned at nest time in lattice builds (0.2 gaps leave less than lead+standoff room — per the 2026-08-27 ruling the pipeline flags squeezed leads, the operator re-aims per contour in the nest-tooling loop).