kerfmaster pin list · Big capabilities
recordStates what the code does today. Auditable against the code — this is the only type that can go stale.
Content last changed 2026-08-10 — computed from the item itself, not typed.
Jordan green-lit bridge emission for the daisy cross (2026-08-10: "This is the design the customer wants. Can we make it work and add bridges? We will need a MetaCam friendly dxf with the bridges added."). The same message settled the thin-web question: the design is as the customer wants it — the sub-80-mil letter webs stay as drawn.
A bridge is an aligned gap pair in the two facing contours joined by two connector cuts across the moat — the crescent pattern the shop's own bridged files carry (a bridged O = two crescent contours). Every contour stays CLOSED: the waste between bridges still drops as normal closed slugs through the existing pipeline (slug tabs, leads, cut order all apply unchanged). The earlier proposal message estimated ~18 extra pierces from open-span posting; that construction was wrong and never built — closed crescents cost far fewer (measured on the board).
place(prims) — for every UNBRIDGED island: boundary contour = smallest contour containing the island's centroid; facing contour = nearest neighbour; N bridges = one per 10" of perimeter, clamped 2..8, spread evenly, each snapped to the narrowest local moat, rejected within 30° of turn across the span (corner rule) and forced apart along the path. Width 0.150" for islands over 2" span, 0.125" under (the approved daisy proposal).apply(prims, placements) — cuts the gap spans out of both contours (exact arc/line splitting, native precision kept at untouched ends), adds the connector cuts, re-chains everything into closed contours.build_bridged.py <src> <label> [tol] — the board: bridged healed DXF, MetaCam shop DXF (R12 bulge profile, bridges in), production post + slug report + cut-path review, and two verifications: QC closure (0 open contours) and the raster island re-test (every island must now separate ONLY when its bridges are cut).Sizing bands and pitch are house numbers pending shop feedback; they sit as constants (BR_W_SMALL/BR_W_LARGE/BR_PITCH) flagged for the material tables.
Jordan reviewed build 017: placement must follow the DESIGN's symmetry ("Aesthetic is a priority. In this case symmetry counts"), and the widths — which were cadmaster's own assumed numbers, floor-at-thickness logic — were too wide, attacking the letterforms.
_mirror_axis() detects a vertical mirror axis (98th-pct reflection error, axis refined around the bbox centre, aesthetic tolerance 0.10"); when one exists, bridges are chosen on ONE side (narrowest moat, corner-avoiding, evenly spread) and MIRRORED, so pairs sit level. No axis → even perimeter spread as before.Jordan's red ink on build 018: the arms must get the same treatment as the shaft — a pair flanking the limb on its flats, not just a tip bridge. Rule added to place(): a protruding limb whose bbox runs ACROSS the mirror axis (wider than tall) is never straddled by the global mirror, so it gets its own pair — one site scored onto its flat, reflected across the limb's own centreline, both then mirrored to the opposite arm. Detection: concave-corner (crotch) segmentation of the island contour; a segment over 3" whose ends sit close together relative to its length is a limb.
Measured on build 019 (daisy_cross_t2brg3): 16 bridges — the 12 of build 018 plus 4 arm straddles at (2.51, 16.56)/(2.49, 13.65) and mirrored (12.10/12.13), pairs level within 18 thou. Raster re-test 0 loose, all contours closed, 21 pierces.
Jordan on build 019: "you added the bridges, but never eliminated the ones that were in the wrong spot." Four circled — the two tip bridges out at the arm ends and the two window-fill picks hugging the shaft crotches. The straddle pair REPLACES a limb's window bridges; it does not join them.
BR_LIMB_CLEAR (0.30") past each crotch corner, is barred to window fill.