[Oliver-Johnson]

Requested by Oliver · project thread

What this changes

Before. A screw's counterbore and shank were cut with their corners on the size typed, so a 6 mm head printed 5.85 mm across its flats and a 3 mm shank 2.90. The fit clearance stopped at 0.3 for a bowtie in the floor under 20 mm, and five audit rows one step past a ceiling built clean and said the ceiling waited on a sweep. A screw head a hair wider than its shank left hundreds of bad edges (#73), and a counterbore whose corner lined up with a flat of its shank left the plate open round every screw (#74).

After. Every screw bore is at least the size typed across its flats, as the magnet pocket has been since #61 (but a counterbore within 0.05 mm of the magnet pocket at its corners, which is cut as the pocket, up to 0.1 mm wider or narrower across its flats than typed), and the screw limits shrink to match: at 42 mm the default plate takes an 8 mm shank, from 8.3; its head still stops at the dovetail's tab, 13.1 mm, and the sizes refused just short of the notch move down from 12.83 to 12.85, 12.94 and 12.95 mm to 12.51, 12.52, 12.61 and 12.62. A bowtie in the floor takes clearances up to 1 mm at any pitch. A puzzle holds 0.3 up to 20.7 mm, not 20, since its leak line runs on past 20 mm and leaves the field at 20.53. A screw head too narrow to clear its shank's corners (under 1.0353 times the hole, plus 0.01 mm) is cut as no counterbore, as a head no wider than the hole always was, and a line under the screw fields says so; it left hundreds of bad edges at some sizes. Where a corner of the counterbore or of the magnet pocket would line up with a flat of its shank, the shank is turned 15/7 degrees (or less, so as not to line another corner up instead), by the pocket's corners alone where no counterbore is cut, and the plate no longer comes out open round every screw (136 edges a piece for the design in #74, 26 for #78's 41.24 mm jigsaw link, 12 for a 4.18 mm shank under a 5 mm magnet from beneath that holds a 5.12 mm head whole, and 30 for that shank under the magnet from above with no counterbore). Where a corner of the counterbore or of the magnet pocket would stand within 0.01 mm of the plane of one of the other's flats, the counterbore is turned half a facet (90/7 degrees) about its axis, its corners midway between the pocket's and its flats where they were, and the plate no longer folds there (the head at its 13.6 mm cap at 42 mm, 1.5 deep under the default 6 x 2 magnet from above, folded the default plate's mesh 160, 120, 128 and 96 times a piece). Turned so, two of its flats can stand square to a cell's x edges, and a cell whose pockets come out with a face turned over is cut again, as one left with open edges is: a 3.89 mm head 1 deep over a 1.01 mm shank under a 3 x 2 magnet from above folded 3 times a cell at 45.16 mm with a dovetail and at 44 mm on one cell, where main builds both closed. The limits are measured on the counterbore as cut, so a head that turns is held to its turned bore, and its cap and the sizes refused beside a joint's cut can differ from those of a head that does not turn: 14.0 mm against 14.1 at 46.45 mm with a jigsaw and a 3 x 2.4 magnet from beneath, for a 3.89 mm head that turns. The page measures the head as typed: it read the head against the limits of the one before it, so a 14.05 mm head pasted over a 6 there, under a 6.2 x 2.4 magnet from beneath, took the 6's 14.1 and built, and 14.1 pasted over a 14 was refused at every read after. A head held to its cap or a gap's end is measured and held again on its own limits until they take it, and the page names that size: an 8.86 mm head at 42 mm with snap clips in the floor and a 3.85 mm magnet from beneath turns and stops at 7.8, but 7.8 does not turn and stops at 7.6, and the page said 7.8 and then refused it. The hole is read last and can be held as well, to the ledge the magnet pocket leaves it, and a head's cap goes by the hole it is measured on, so a hole the read moves is another hole to read the head on: the head is read again from what was typed, on the hole as held, and the hole again on that head, until the hole stays. A 7.97 mm head over a 7.71 mm hole at 42 mm with a jigsaw and an 8.1 mm magnet from beneath builds, but with a 6.35 mm magnet pasted in the hole was held to 6.1 and the head left at 7.97, where its cap is 7.9, with Checks naming the hole alone. A screw hole too wide for the magnet pocket it runs through to clear, as a counterbore has to, is refused at the field: the two walls weave and no turn kept them closed (47 open edges for a 5 mm hole under a 5.1 mm magnet), and the magnet would have nothing to rest on over the hole. At 42 mm that holds the hole to 5.8 mm under a 6 mm magnet. The test tile with magnets and screws is watertight again at every counterbore depth. The default 6 mm counterbore's corners stood 0.022 mm inside the 6 mm magnet pocket's, and deeper than the magnet from beneath it left 6 open edges a cell at 42 mm, where main built closed. It is cut as the pocket now, one prism with it, and the shank's turn and the limits go by it as cut. The other ceilings stay, each with the leak that earns it built in the audit, and the page says "not watertight" where it said the joint opened holes.

How. One helper, MOUNT_BORE, decides each bore's sides and corner radius. The cutters read it, and mountLimits measures each bore by it as #69's polygon, the shank turned where it is cut turned (MOUNT_BORE.screw). The ceilings were swept every 0.01 mm of pitch and every 0.05 of the field on 2 x 2 and 1-cell pieces (about 160,000 builds); only the bowtie in the floor was clean a step past a new ceiling, and the puzzle's line was followed past 20 mm to where it leaves the field. Heads near the shank's corner circle were swept on 7,028 one-cell plates, and test tiles with screws and magnets on 2,986. The #74 sizes were swept on 3,846 one-cell plates with a magnet from above (569 closed, none opened), on 1,800 random designs a counterbore's corner turns (71 closed, 11 opened, 1,718 unchanged) and on 1,500 that only a magnet's corner turns (41 closed, 1 opened, 1,458 unchanged); designs it does not move build the same bytes. Of the 11 opened, 10 have a 6 x 2 magnet from beneath under a 3 mm counterbore, 9 of them at 42 mm. There the turn opens more than it closes (9 against 5), and main builds those sizes closed but one; that mount also leaks at sizes the turn leaves alone: of 300 random sizes at 42 mm, 21 of the 273 it does not turn leak here, and 17 on main. Counterbores near the magnet pocket's corners were swept on 240 random designs that reach the pocket: 2 of the 223 taken build bad here, against 12 at 6bdb7f4 and 7 of 225 on main. Of 600 random designs with screws, the 560 whose counterbore is not near the pocket build 6bdb7f4's bytes. The default 6 mm head is near the default 6 mm magnet: from beneath it stays inside the pocket and builds the same bytes, and from above it is cut with the pocket as one prism. A bare magnet pocket, with no counterbore cut and a corner within 0.0045 mm of a flat of its shank, was swept on 1,000 random designs from beneath and 800 from above. From beneath, 6 of the 859 taken built bad unturned and 2 turned (6 open edges each under a 4.16 x 2 magnet at 42 mm, as unturned). From above, 11 of the 566 taken built bad unturned (351 open edges) and 3 turned, test tiles with a face turned over and no open edge. The turn opens one size known from beneath, 2.58 mm under a 4.34 mm head the 4.16 x 2 magnet holds whole at 42 mm: 9 open edges, where main leaves 10. Of 700 random designs with screws, the 7 that build other bytes than b0f37a4 are the ones it turns, each with a magnet and no counterbore cut. The counterbore's turn off the pocket's flats was swept counting folds as the audits count them: two faces with area back to back in the mesh (test/orientation.js leaves out a face with no area). On 600 random designs with a head within 0.012 mm of a size that puts a corner of either on a plane of the other, on plates of several cells and split pieces (2 x 2, 3 x 2 in two, 3 x 3 in four, two pieces with dovetails or jigsaws, 2 x 2 on corner bosses), 48 of the 354 taken built bad unturned (b7ffea3) and 1 here, against 1 turned π/252 and 5 of 356 on main; that one leaks on all of them. An 8.798 mm head 0.592 deep under a 7.037 x 2.5 magnet from above at 42 mm, which π/252 and main build closed, had an edge shared four ways and two folds a cell until a cell whose pockets come out with a face turned over was cut again. On round 4's 900 designs (one cell, the test tile, 2 x 2, corner bosses, two pieces), held to what the page takes, 56 of 498 built bad unturned and none of 496 here, against 2 at π/252 and 2 of 504 on main. With the head at its cap, under each magnet that puts a counterbore corner on a flat's plane, every 1 mm of pitch from 34 to 56 mm on 2 x 2 and 3 x 2 in two, 18 of 108 from above and 8 of 186 from beneath built bad unturned and none here or at π/252 (6 and 2 on main); every 0.5 mm on one cell, 28 of 216 and 14 of 360, and none here or at π/252 (21 and 8 on main). Read back from the STL in single precision, 14 of these designs show a fold here, 30 at π/252 and 114 on main, but here and at π/252 every fold the mesh does not have is a triangle with no area in the mesh, which rounding gives a direction. Those are not folds in the solid: the 48, 36, 36 and 27 given before for π/252 on the default plate were 147 of them over a mesh with none. Half a facet is kept over π/252, both turned from 0.01 mm, because π/252 leaves designs open that half a facet builds closed: a 13.582 mm head 0.251 deep under a 6 x 3 magnet from above at 42 mm (12 open edges a cell), and three found in review (3 and 12 open edges, and two shells touching along 8). Half a facet moves the limits of a head that turns more than π/252 does: of 1,034 designs drawn with a head near a size that turns, 302 against 102, the head's cap by up to 0.4 mm up or 0.3 down against a tenth down. Measured both ways with the tighter kept, the limits would not move with the head typed, but 28 of 140 page designs at five pitches would lose 0.1 to 0.3 mm of the head's cap (8.1 to 7.9 on a 2 x 2 at 42 mm with a jigsaw) and 12 dovetails would refuse more sizes, so the limits stay measured on the bore as cut. The turns are taken where a corner comes within 0.01 mm, where the sweeps of #74's sizes above took them from 0.005: at 0.005 a 38.97 mm cell under a 3.96 x 1.5 magnet from above, a corner 0.006 mm off a plane, folded 3 times unturned, and π/252 left a 53.72 mm test tile open by 15 edges round a shank whose flat a corner missed by 0.00502 mm. Of 600 random designs whose turns that changes, the 511 the page takes build bad on one at either bound (6 open edges where 0.005 and main leave 3), and main leaves 4 open. Of 800 random designs with screws, the 54 that build other bytes than at 0.005 are the ones whose turn it changes (48 the shank's, 6 the counterbore's), and all 800 keep their limits. Cutting a cell again where its pockets come out with a face turned over was swept on 2,366 of the designs with screws above (both classes, the heads at their cap and the shank sweep): 49 were cut again, 47 build other bytes, 4 that folded build closed (the 8.798 mm head and three test tiles the page does not take), none came out worse in the mesh, and building them all took 1.5% longer. On the page, of 3,124 heads pasted near their cap over 781 random designs, 1,286 were held once and 33 twice, none more (four passes are allowed), and every one ends on a size its own limits take, where 100f9f1 held 33 to a size they refuse. Of 43,536 heads, magnets and holes pasted near their caps over random screw designs with each joint (eight seeds), 16,006 took a second round, none a third (four are allowed), and each ends on a head and a hole that their own limits take measured on each other, and a second read changes nothing, where 8bc63fe leaves 499 for a second read to change, 4 of them with the head above its own cap.

Designs with no screws build the same bytes as main 326cc48 (500 of 500). With #69's dovetail rule, of 1,400 random designs with screws into or beside a joint's cut, 7 built bad at 0e12f68, each a single size that builds the same bytes as main f319c83 with the same bores. On 2,000 random dovetail designs with screws, drawn the same on main f319c83, 9 of 2,083 built bad at 0e12f68 and 12 of 2,101 on main (6 and 11 at the notch, folded edges); each bad one here is a single size, clean 0.005 mm either side, that main builds with the same flaws from the same bores, and no measured feature sets them apart from the clean ones, so the limits add no guard for them. On 320 random designs with corner pockets and screws, with no joint and with each joint cut from beneath, none has an open edge, a fold or a turned face here; main f9873f5 builds 4 of them open at the same sizes. #80's design builds closed here at every head the page takes there (its default 6 mm head is now over the 5.8 mm cap), where main leaves 24 open edges a piece.

New top-level names: MOUNT_BORE in core.js (with MOUNT_BORE.screw), and headCutAsNone and headClearsFrom in ui.js, none of them on main 64f91bd (checked with #65 in). screwCutter, which nothing called, is removed. MOUNT_BORE is exported for the audit's row that still reaches the seam repair's guard. This branch contains main 64f91bd (#78, #79, #82, #84 and #65), merged in 8c7f8dd, 81c83ac, b0f37a4, ac8da72 and 8bc63fe; #82's one-line clash keeps both fields, jointCuts.push({ top: ceiling, bottom, segs, sites }). With #84 in, the rim's screwHoleD flag in mountLimits is now false where the magnet pocket the shank runs through sets the hole's cap, as beside's is, and ENGINE.md gives the rim's screw-hole cost with the hole sized across its flats (6.2 to 3.8 at 34.5 mm).

Closes #73.
Closes #74.

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 002df36: npm test, ci-sim, the plate UI specs plates-input, plates-robustness, export, keys and half-cells, 186 passed; the whole UI suite with two workers passed 826 of 826 on 002df36, 43 minutes)

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 screw rows read the bore through the same corner rule, and the clearance rows build each ceiling's earning leak, the bowtie's step to spare, the puzzle's line at 20.05 mm with the step under it, heads either side of the counterbore line (Baseplates: a screw head only slightly wider than its shank leaves hundreds of bad edges #73), the test tile with magnets and screws, and counterbores with a corner on their shank's flat, under a magnet from above and from beneath (Baseplates: magnets from above with a screw head a little wider than its shank can leave bad edges #74), a magnet pocket's corner on its shank's flat (Build every piece of the 39.07 mm dovetail plate, and name a piece that fails in Checks #78's 41.24 mm link, and a 7.4 mm head beside it), and shanks the magnet pocket does not clear, refused, and counterbores within 0.05 mm of the magnet pocket's corners (the test tile 2.5 and 3 deep, the 42 mm cell 3 deep, the 44.46 mm dovetail from above, and two rows that guard the turn and the one-prism cut), and main's bores at 41.24 mm cut apart from the magnet pocket, which still reaches the seam repair's guard, and bare magnet pockets with a corner on their shank's flat, five from beneath that hold the counterbore whole and three from above under heads too narrow to cut, and counterbores with a corner on the plane of the magnet pocket's flat or the pocket's on theirs (the head at its cap at 40, 42, 43, 44 and 46 mm under the magnet that puts its corner there, at 42 mm from above 1.5 and 1 deep and from beneath 3 deep, and two rows that hold the turn to its size, and plates of several cells, so that one cell cannot pass for them: the 13.6 mm head 1.5 deep on a 2 x 2, on a 3 x 2 in two pieces and on the default plate, and a 13.582 mm head on a 2 x 2, and five designs that the 0.005 mm bound or π/252 left open or folded: a 38.97 mm cell, a 53.72 mm test tile, two 42 mm cells under magnets from beneath and an 84 x 42 plate, and a 3.89 mm head under a 3 x 2 magnet from above at 45.16 mm with a dovetail and at 44 mm on one cell, which folded until a cell whose pockets come out with a face turned over was cut again); and four page tests, one that a screw head pasted near its cap is judged on its own limits (fails on e1e542c), one that the shank is then measured on the head as held, one that a head held to its cap is held again until it holds and named there (fails on 100f9f1), and one that a head is measured on the hole as held, so a pasted magnet cannot leave it above its own cap (fails on 8bc63fe)

🤖 Generated with Claude Code

https://claude.ai/code/session_01Uqqb1cvhXxrpb3P6y6ykBY