kerfmaster pin list · The spec — what we build to
recordStates what the code does today. Auditable against the code — this is the only type that can go stale.
Content last changed 2026-08-07 — computed from the item itself, not typed.
Contract — design stage
This contract describes the code as it runs today.
takes vector ?
makes contours ?
fails if Cutting geometry we drew instead of geometry Aristide drew; silently re-laying out his slot pattern.
Contract last changed 2026-08-07 — computed, not typed. Dated separately from the text above, so neither date can speak for the other.
Aristide, 2026-08-07: “prepare cuts only for 3 items (elements) from layer 2 and then what is on layer test.” mvt_cut.py is exactly that and nothing else. Every element comes out of the one file he drew it in, read by name with [[ai-layer-reader]], so nothing is stitched together from two sources.
The subpaths go into bezfit as their true Béziers, not as sampled points, through the same fitter the boards use. Worst fit deviation 0.00141 in, worst fit-induced kink 0.029°. Output is 566 primitives over 30.000 × 22.277 in: viewer/mvt/cut/mvt_option_c.dxf with one layer per element, plus a DXF per element and cut.png.
Width 0.28, minimum metal to the bracket edge 0.40 — agreed earlier, applied here, not re-argued. As drawn the 55 fastener slots are 0.075 and 0.111 wide and sit 0.045 from the edge. They are cut as real stadiums: two lines and two 180° arcs, never a polyline.
The width is applied IN PLACE. The first version relieved the edge by sliding each slot inboard; on a bracket with slots 0.045 from the edge in dense rows that walked one slot 4.6 in across the part and overlapped its neighbour. A slot's position is the design — moving it is a re-layout of the pattern, not a width fix. So the width goes on and what still fails is reported: at 0.28 wide, 12 of 55 slots have under 0.40 in of edge metal, 4 break out of the bracket, and the closest slot-to-slot gap is 0.135 in. That is a re-layout waiting for Aristide, and it is on the page rather than hidden by a nudge.
One thing is dropped: a 0.108 in stray dot on layer test, 0.01 in², under the pierce standoff. It is named in the report rather than deleted quietly.