Skip to content

Build every piece of the 39.07 mm dovetail plate, and name a piece that fails in Checks - #78

Merged
Oliver-Johnson merged 8 commits into
mainfrom
claude/dovetail-counterbore-build
Oct 9, 2026
Merged

Oliver-Johnson merged 8 commits into
mainfrom
claude/dovetail-counterbore-build

Conversation

@Oliver-Johnson

@Oliver-Johnson Oliver-Johnson commented Oct 9, 2026 •

Copy link
Copy Markdown
Owner

Requested by Oliver · project thread

What this changes

Closes #76.

Before.

  • A dovetail plate at 39.07 mm with a 3 mm screw, a 2.34 mm counterbore and a head of 7.3 mm or more listed only A1. B1, and every piece with a seam on its left, never built.
  • Download was off, and the checks under the cut map said nothing.
  • The engine's seam repair threw on B1's corner cell. Two facets of the socket's sloped corner cross the counterbore's ceiling about 2 microns from an edge of its own. That leaves three corners just over the weld distance apart, each within it of the line through the other two, and the repair kept putting them into each other's edges until it ran out of passes.
  • It was a point and not a band: 2.33 and 2.35 mm built, as did 39.06 and 39.08 mm, other shanks and other tolerances. Main before Plates: keep magnet and screw pockets out of a joint's cut in the floor #69 did the same, so it is not new with Plates: keep magnet and screw pockets out of a joint's cut in the floor #69.

After.

  • Every piece builds, watertight and with every shell facing outwards: the plate as filed, 164 mm wide, three seams, and heads from 7.3 mm to the cap.
  • When a piece does fail to build, the checks say so and name it: "Piece B1 could not be built, so the build stopped there and Download is off." Before, only the piece table, the preview and Download's tooltip said so.

How.

  • healCsgSeams does something new only when it runs out of passes.
    • It makes one point of the points it was still putting into edges, where they lie within two weld distances of each other.
    • It moves every vertex that had collapsed onto them.
    • It runs once more, and keeps that run only if it comes out closed.
    • A second run that does not settle, or settles open, throws as before. A jigsaw plate at 41.24 mm with magnets from above is one: it fails to build, as on main, rather than going out with 55 bad edges a piece. (Its family builds open on main at nearby sizes; that is Baseplates: a magnet pocket and a screw bore of nearly the same width can leave bad edges #77.)
  • A repair that settles in its passes builds the same bytes as before. A retry that threw and was passed over can now come back closed and be kept. Checked piece by piece against main:
  • The checks list carries the failed piece. It is drawn as an error, but not one that stops the next build, which clears it.
  • healCsgSeams is shared with the Bins page, so bins/index.html is rebuilt too; the Bins audit is clean. No new top-level names.

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: npm test, ci-sim, and the plate UI specs, 166 passed; not the whole UI suite)

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: the Baseplates: a dovetail plate with a 2.34 mm deep counterbore fails to build one piece at 39.07 mm #76 plate as filed, 164 mm wide and with three seams, the smallest head that threw and the cap, and the step either side in head, depth and pitch, now checked for orientation too (10 rows; the first five throw on the base). Also the 41.24 mm jigsaw plate and its 2.264 mm magnet variant, which must throw in the seam repair or build watertight and oriented (2 rows; both fail on 2767936).

The page tests: a new case opens the #76 link and needs "2 ready" with Download on, and the stubbed-failure case checks that Checks names the piece, says Download is off, and that the next build clears it. Both fail on the old page.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Uqqb1cvhXxrpb3P6y6ykBY

claude added 4 commits October 9, 2026 15:53
…piece

Every piece with a seam on its left threw in the seam repair and was never
built. Two facets of the socket's sloped corner cross the counterbore's
ceiling 2 microns from an edge of its own, which leaves three corners just
over the weld distance apart, each within it of the line through the other
two; the T-junction pass put them into each other's edges until it ran out
of passes. When the passes run out, the repair now makes one point of the
points it was still putting in that lie within two weld distances of each
other, and runs once more. A repair that settles in its passes, which every
one did across 1,100 random plates, is unchanged. Audit rows for the plate as
filed and the edges of where it threw.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Uqqb1cvhXxrpb3P6y6ykBY
A piece the engine threw on stopped the build and turned Download off, and
the piece table and the preview said so, but the checks under the cut map
said nothing. They now say which piece could not be built and what usually
gets past it, drawn as an error, and the next build clears it. The page test
forces a failure for that, since no design is known to throw now, and the
plate from #76 is opened and has to build both pieces.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Uqqb1cvhXxrpb3P6y6ykBY
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Uqqb1cvhXxrpb3P6y6ykBY
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Uqqb1cvhXxrpb3P6y6ykBY
claude added 4 commits October 9, 2026 17:13
… as before

The merge #76 added ran the repair once more and returned whatever that came to. A
jigsaw plate at 41.24 mm with magnets from above and a 3.126 mm counterbore runs out of
passes on both pieces, and the second run settles with edges open: the page built A1 and
B1 with 55 bad edges each and Download on, where main fails A1 and says so. With a
2.264 mm magnet the pieces came out closed but folded, with NaN normals in the STL. The
second run is now kept only if it comes out closed; otherwise it throws as before.

plate-audit.js holds that plate and its 2.264 mm variant to "threw in the seam repair,
or watertight and oriented", built as the page builds them, and the #76 rows are now
checked for orientation as well as leaks.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Uqqb1cvhXxrpb3P6y6ykBY
…e that gets past it

"The plates cannot be downloaded" overstated it: the pieces built before the one that
failed keep their own STL buttons in the piece table. And "a size changed by a
hundredth of a millimetre usually gets past it" can send someone to a size that builds
with holes: the 41.24 mm jigsaw plate fails with a 7.504 mm head and builds with 57 bad
edges a piece at 7.502 mm, on main too. The forced-failure spec's comment said no design
was known to throw; that jigsaw plate does again, and the spec now reads the new wording.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Uqqb1cvhXxrpb3P6y6ykBY
Before #76's merge, the turned cutters of a cell cut again threw at healCsgSeams' pass
limit on a few engine-only cells, the build passed the throw over, and the nudged try
after it closed the cell. Now that try can settle after the merge, come back closed,
and be kept, so "same bytes unless it threw" holds per cut, not per piece. The audit's
"a retry that throws" is that case: the limit is reached ten times over its five
pieces, every piece's bytes change, and every piece is still watertight and oriented.
Its comment and the fastener cut's now say so, and that no audit row reaches the catch
any more (traced over the whole audit: 19 merges, 2 throws, neither passed over).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Uqqb1cvhXxrpb3P6y6ykBY
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Uqqb1cvhXxrpb3P6y6ykBY
@Oliver-Johnson
Oliver-Johnson merged commit 5da18ce into main Oct 9, 2026
3 checks passed
@Oliver-Johnson
Oliver-Johnson deleted the claude/dovetail-counterbore-build branch October 9, 2026 18:09
Oliver-Johnson pushed a commit that referenced this pull request Oct 9, 2026
Only sw.js conflicted, on its VERSION line: taken from main and rebuilt
with node build.js, which regenerated both pages.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XjNf7SQF9o4ZVeWiSbWQmk
Oliver-Johnson pushed a commit that referenced this pull request Oct 9, 2026
Only sw.js conflicted; the pages are rebuilt from the merged sources.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Uqqb1cvhXxrpb3P6y6ykBY
Oliver-Johnson pushed a commit that referenced this pull request Oct 9, 2026
sw.js and plates-robustness.spec.js conflicted: the spec keeps #73's cases and takes
main's comment and name for the failed-build case. The pages are rebuilt from the
merged sources.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Uqqb1cvhXxrpb3P6y6ykBY
Oliver-Johnson pushed a commit that referenced this pull request Oct 9, 2026
…for a counterbore's

MOUNT_BORE.hole.turn looked at the magnet pocket's corners only once the counterbore's
already called for a turn. So a shank no counterbore corner came near was cut unturned
even with a magnet's corner on a flat's plane, and the BSP split there: the review's
41.24 mm jigsaw link, a 3 mm shank under a 7.3 mm magnet from above whose corner stands
0.4 microns off a flat, built both pieces with 26 open edges and Download on, where main
fails to build. The magnet's corners now call for the turn as the counterbore's do.

The link, its 2.264 mm magnet and the 7.4 mm head beside it build whole, and #78's rows
keep their test, with the 7.4 mm head added. Heads 7.3 to 7.7, depths 3 to 3.25, magnets
2.15 to 2.35 thick and pitches 41.2 to 41.34 close, but for 7.31 mm, #77's coincidence.
The turn moves none of the 3,846 one-cell plates or 1,800 designs it was measured on, and
of 1,500 random designs it now turns, 54 close and 1 opens, at that one size.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Uqqb1cvhXxrpb3P6y6ykBY
Oliver-Johnson pushed a commit that referenced this pull request Oct 9, 2026
…1.24 mm link builds

With its shank turned, the review's 41.24 mm jigsaw link no longer runs out of passes
here, so #78's rows build whole and no longer reach the second run's guard. A fourth row
cuts main's bores for it, 7.504 cos(pi/14) and 3 cos(pi/12), which still run out of
passes and throw, here as on main. The comments that said the plate fails here now say
where it still does.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Uqqb1cvhXxrpb3P6y6ykBY
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Baseplates: a dovetail plate with a 2.34 mm deep counterbore fails to build one piece at 39.07 mm

2 participants