Repository navigation
Description
A variable round's stations are accepted only on its rails, so nothing bounds its skin between them. The progressive cubic round of fork_ea59e9c_section_orders::wr711_exact_circles_of_one_arc_count_keep_the_progressive_round_on_its_balls is built 3.07e-8 (relative) from its balls' envelope volume, past the test's 3e-8.
What is measured
The body is the 10 x 20 x 30 box with the edge at x = 10, z = 30 rounded by the progressive cubic law [2, 2, 6, 3, 0] (holds_case), on the SolidCraft kernel. The edge tolerance is 1e-5, so the solver's approximation tolerance is T = 4 x 1e-5 = 4e-5.
- Volume. 5957.718684883223, against the envelope's
HOLDS_VOLUME= 5957.718502: 3.0697e-8 relative. Two integrators agree on it (adaptive Gauss-Kronrod at the test's 1e-7 and forced to 1e-9 through 1e-12, and the Runge-Kutta integration forced to 1e-12: 5957.718684883148, 1.3e-14 apart). So the gap is the body's, not the integration's. - Skin. The signed distance to the union of the balls,
min_s |x - c(s)| - r(s), over a 200 x 200 uv grid of the face, is -6.33e-5 to +8.67e-5: 2.17 T. It is about 1e-11 to 1e-15 at the trace's stations, and peaks mid-span (the spans 35.288-37.137 and 39.18-40.76 both reach 8.67e-5). - It converges only with T. Setting only the filleted edge's tolerance:
| edge tol | T | skin off | / T | (V - V_env)/V |
|---|---|---|---|---|
| 1e-5 | 4e-5 | 8.67e-5 | 2.17 | +3.070e-8 |
| 5e-6 | 2e-5 | 3.68e-5 | 1.84 | -1.415e-9 |
| 2.5e-6 | 1e-5 | 1.38e-5 | 1.38 | -5.906e-10 |
| 1.25e-6 | 5e-6 | 3.7e-6 | 0.74 | -7.224e-11 |
- Other laws on the same box (interior y in [2, 18]):
[2,2,6,3,1]6.1e-5 (1.5 T),[1,3,5,3,1]8.7e-6 (0.2 T),[2,4,2,4,2]4.07e-4 (10.2 T),[3,3,3,3,1]3.16e-4 (7.9 T).
Where
- T is
TOL_FACTORx the edge tolerance (blend/fillet_executive/p2.rs:557, used at:796). - Its only acceptance test is
compute_next_point(blend/fillet_intersector.rs:1602-1640): the Hermite rails lie on the faces within T at the five test shares. Where the radius varies, the contacts solved inside the span also lie within T of the rails (:1643-1700). Nothing tests the skin. - The skin is built later, in
flush_curve(blend/fillet_intersector.rs), from all stations at once.CircularCrossSectionFsg::create_surface_from_curves_of_balls(blend/fillet_standard_solver/p1.rs) puts each ball's touch circle at a station and skins them with a cubic in the trace direction. Between stations the skin is that interpolant, and the rail test doesn't see it. - These rounds are bare B-splines (no definition), so no certificate covers them either.
Proposal
Hold the skin to T where the construction decides its stations. In flush_curve, after the skin is built, solve the round's section at each span's middle (as the span equalisation already does). Then measure the skin's iso-curve there against that section's touch circle. The distance from a point to a circle has a closed form. The circle lies on the balls' envelope, so that distance bounds the point's distance to the envelope. Where it exceeds T, add the solved section as a station and build again. Keep the rounds and the station count bounded, and refuse with a typed cause past them.
Only rounds whose skin is this circle generator's are affected: the ball's radius varies and the circles are exact. Their geometry moves wherever a span fails today, and every other blend is unchanged.
Found while validating #15021, which waits on this.
Written by Claude Code (Anthropic) on behalf of @brett19.
Activity
github-actions commentedon Nov 12, 2026
Claimed by @brett19 through #15257. The claim lapses after 14 days without pushes.
solidcore: a variable round holds its skin to its tolerance between s…
solidcore: measure a law-corner round's pieces, and place a skin stat…
solidcore: claim: hold a variable round's skin to its tolerance betwe…
solidcore: a variable round holds its skin to its tolerance between s…
solidcore: measure a law-corner round's pieces, and place a skin stat…
solidcore: claim: hold a variable round's skin to its tolerance betwe…
solidcore: a variable round holds its skin to its tolerance between s…
solidcore: measure a law-corner round's pieces, and place a skin stat…
solidcore: goldens: six variable rounds' mass properties move as thei…
PROPOSED (maintainer decides): wr646's default-kernel test asserts th…
solidcore: claim: hold a variable round's skin to its tolerance betwe…
39 remaining items
solidcore: type the skin check's refusals, and split a span whose sec…
solidcore: keep a variable round's span only on a proved bound over t…
solidcore: expectations: six law rounds' goldens move; wr646 holds it…
solidcore: keep a variable round's span only on a proved bound over t…
solidcore: expectations: six law rounds' goldens move; wr646 holds it…
solidcore: claim: hold a variable round's skin to its tolerance betwe…
solidcore: a variable round holds its skin to its tolerance between s…
solidcore: measure a law-corner round's pieces, and place a skin stat…
solidcore: prove the moved law rounds against their balls in closed f…
solidcore: type the skin check's refusals, and split a span whose sec…
solidcore: keep a variable round's span only on a proved bound over t…
solidcore: expectations: six law rounds' goldens move; wr646 holds it…
Solve a skin span's middle section once and cap stations after equali…
solidcore: keep a variable round's skin only on a bound over each spa…