[Oliver-Johnson]

Requested by Oliver · project thread

What this changes

Before. On a solid floor, with an H-clip or snap clip put in from above, some magnet sizes from beneath left piece A1 with 3 to 14 open edges at the joint's housing, and Download stayed on. The issue's case is a 6.3 mm magnet at 34.5 mm (4 open edges with the H-clip, 3 with the wall snap clip); a 6.25 mm magnet at 37.5 mm left 14. Wall keys put in from above did the same at some sizes.

After. These plates build watertight. No magnet size is refused that was taken before, and no cap, Checks wording or field limit changes. Of 40,884 designs swept (38,013 of them sizes the field takes), 176 came out open here before, every one with its joint put in from above, and all 176 come out closed now with no fold; exactly those 176 build other bytes. 600 random designs (510 built) build main c0dc948's bytes (984 pieces), and none takes the new path.

How. The issue guessed the pocket's cap shared a plane with the housing. It does not: a magnet from beneath stops 2.2 mm under an H-clip's cup (2.1 under a snap clip's slot). What lines up is a flat of the pocket. The BSP carries its plane across the whole cell and splits the cell's faces where it crosses them; at 34.5 mm with 6.3 mm the nearest flat crosses the side of the cell's region 0.019 mm from the seam face, beside the corner the region's top face is fanned from. The spokes there are microns apart, the weld in healCsgSeams chains and moves the side's top corner 3.4 microns off its plane, and the housing, cut afterwards, crosses that face with its end wall 1.9 and 2.5 microns off it at its two ends, which the weld joins at one end and not the other. Moving the housing, as cutAgain does, leaves the face as it is. So in buildPiece's top-insert pass, a whole cell whose housing still comes out open has its pockets cut again the ways settle already does (moved 1.7 microns along a diagonal, turned a 28th of a turn, or both), the housing cut out of each try, and the first that comes out closed, with nothing turned over and no more edges shared with the shells beside it, is kept. A housing that comes out closed is built as before. The cost is at most seven pocket cuts and housing cuts for one cell, on designs that came out open. 20 of the 176 are wall-key cups from above, which the issue does not name. The report is /mnt/project-files/issue-batches/issue-85-report.md.

Two defects the sweep found, the same on main, are not touched here and are filed as #89: a wall bowtie from above at tight tolerance, 37.5 mm, 7.8 mm magnet (2 edges used four times), and a wall puzzle key from above at loose tolerance (2 folds at 39.5 mm with 11.2 mm, the cap, and at 42.5 mm with 5.8 mm).

Closes #85.

Checks

• I edited the sources in src/, not the generated index.html / bins/index.html
• I ran node build.js and committed the regenerated pages alongside the sources
• npm run test:all passes (run here on 3140d62: npm test (the bin and plate audits among them) and node test/ci-sim.js clean; the plate UI specs plates-input, plates-robustness, export, keys and half-cells, 186 passed; CI's check and ui jobs, the whole UI suite, both passed on 3140d62)

If this touches geometry

• Bins: node test/bin-audit.js reports all cases clean
• Baseplates: node test/plate-audit.js reports all plates watertight, and no case has picked up bad edges that were not already named as known
• I added a case covering the new shape: four audit rows, the issue's two 34.5 mm links, a 6.25 mm magnet at 37.5 mm (14 open edges on main), and an H-clip control at 42 mm with the 13.6 mm cap; the first three fail on main with 4, 3 and 14 bad edges

🤖 Generated with Claude Code

https://claude.ai/code/session_01Uqqb1cvhXxrpb3P6y6ykBY