Files
solidcraft/docs/perf.md
T
ea772a0438 Merge develop into main: SolidCraft on the SMLib kernel (v1.1.0-alpha) (#11351)
* UI: rim and wall picking tests for a circle of arcs that is one wall

A circle drawn as two half arcs now extrudes to one full-turn wall with its
seam edge, and each rim is one closed edge (tangency-labs/smlib#67). The tests
keep their intent on that body: a rim clicked once is one item, measured and
highlighted whole and picked from either side; a wall picked either side of
its seam is the same surface, lit whole, and toggles off from the other side;
every face is its own patch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Fillets: a later round at a corner where earlier rounds meet blends them again with it

The owner's ruling of 2026-10-30: a later fillet on an edge at a vertex where earlier
rounds already meet in a mitre makes the corner rounding all of them at once makes, one
smooth patch, the mitre gone. A body made by constant-radius rounds keeps what they were
made on and which edges they rounded at which radius; a later round reaching such a corner
blends the earlier rounds' edges (from the earliest of those meeting there on) with the new
edges in one SMLib blend of the body before them. The result's history is put over the body
the later round was given: an earlier round's face is that face trimmed anew (its name
kept), the corner patch and the new rounds are the later fillet's. A picked mitre seam has
no round of its own: it is refused unless a new edge of its radius meets it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Tests: a later r3 round at an r2 mitre, through the Fillet feature, makes the all-at-once corner

Two box edges rounded r2 in one fillet, then the upright edge at their mitre r3 in a second
fillet (tangent chain on, so the seam comes with it): 10 faces, 7928.743041 mm³ as
fillet_radii of the box gives, the r2 rounds' names kept, the r3 round and the corner patch
the second fillet's.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel: a body's record of rounds goes with its history, and can be read back

The rounds that made a body are over the body its history is over: a new history (an
operation carrying a step's result back to its own operands) leaves the record out, so a
later round never reads one history against another body. Body::rounds gives the record and
Body::with_rounds_of takes one back, checked (edges of the body before them, a fillet's
radii, the body's history over that body), for a saved model to keep it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Model cache: a body made by rounds keeps them, so a later round blends them the same after opening

A design gave another body when opened from its model cache: the rounds a body was made of
lived only in memory, so a later round at their mitre was built sequentially (or refused, as
too large) after opening and blended with them when evaluated anew. The saved model now
records, per kernel body made by rounds, the body they were made on (a place in the bodies,
stored once though no state holds it) and the edges with their radii; reading it attaches
them back down each chain, and a thinned model keeps those bodies with the bodies made from
them. Model cache format 7.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Tests: a corner patch of rounds 2/3/4 takes no more than the strips leave in its block

three_rounds_at_a_corner bounded the patch's extra cut at 3 mm³ beyond the strips, a figure
with no derivation; SMLib's all-at-once body at radii 2, 3, 4 takes 3.099 (V 7878.579574
against strips 7881.678). That body is the strips' outside the block its patch lies in: its
rounds are the strips' cylinders ending at a's section y = rc, b's x = 20 − rc and c's rails
where they meet a's and b's (the patch's four corners, now asserted), so the patch takes at
most what the strips leave in that block (11.86 at 2/2/3, 32.69 at 2/3/4). The lower bound is
that block; the upper one, at most the strips' volume, stays.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Tests: the io crate's hand-built saved models list their rounds, as format 7 records do

Model cache format 7 (f3cf0c4f3) records a `rounds` list in every saved model, and reading one
without it is refused ("saved model: rounds"), as is a model without any of its other lists:
the states, the preview and the thinned model all came back empty. The io tests' saved_model
and timeline_model wrote format-6 records, so six tests (the round trip, the repeated part
stored once, and the four too-big-for-the-file cases) read nothing. Their bodies are none made
by rounds: the list is empty, and every expectation stays as it was.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* SMLib: pin eaa998a (tangency-labs/smlib#68, #69, #70)

Batch 4: drafts and loops, mitre corners, rim chamfers, rulings, tight box, stitch, point
classification, eased corners. Batch 5: a face's tight box divides its cells at the surface's
knots (thread boxes exact); a cylinder, cone, sphere or torus is extended along its own circles.
Batch 6: a revolution's extension closes its turn; a dropped point's foot is the converged one.
The C API is ABI 5 (smlib_blend_dropping and smlib_brep_classify_point are new; nothing existing
changed), so the binding expects 5.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Tests: a strip's tangent and curvature run-offs are the distance long across the surface

SMLib measures an extension's distance across the surface (smlib_extend_sheet corrects
CreateExtendedSurface until the parameter line through the side's middle is that long), so each
run-off of the cylinder strip, swept 20 along z, adds 5 x 20: 10 pi/2 20 + 100 = 414.159 for
both. The parabola-length formula (418.18) was truck's.

Ruling d2 (2026-10-30).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* kernel: one smooth angle, and a round too small to change the part is refused

The blend's tangent angle is the shaded view's half degree (smooth_edges::ANGLE), one value
for smooth. A constant round lies r (sec(d/2) - 1) inside the crease it replaces (about
r d^2/8); below the edge's tolerance it changes nothing and is refused before SMLib (which
answered SM_ERR_INVALID_INPUT), naming the smallest radius that rounds,
tolerance / (sec(d/2) - 1). d is the least turn along the edge
(tangent_edge::least_angle_along, at tangent_all_along's stations).

Ruling d3 (2026-10-30). The block's tolerance is 2e-5 mm, not 0.1 um: at 1 deg the smallest
radius is 0.526 mm, so R2 and R20 reach SMLib, which still calls an edge within 2 deg smooth
(SmEdge::GetBlendEdgeuse); that test is ignored until SMLib uses 0.5 deg.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Fillet rig: the knife-edge chamfers name SMLib's cause, a cusp's ball radius of 0

smlib_chamfer turns the distance into a ball radius d sqrt((1 - c)/(1 + c)), c the faces'
binormals' dot (capi chamfer_radius); at #851's cusp the binormals agree, the radius is 0 and
the call is SM_ERR_INVALID_INPUT before the fillet executive runs (gdb: chamfer_edges sees
radius 0; the dump's why is empty). SolidCraft hands every chamfer to SMLib, so the
ball-contact chord (the ball of radius d: 3.822 wide at d = 2, 159.2365 removed) is SMLib's to
build.

Ruling d4 (2026-10-30).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* kernel tests: a corner round wider than its faces overflows and takes the corner's other edges

R21 on a 20 mm cube's edge builds (6107.529370, closed form, to 2e-12). With R5 on the
corner's other edges the R21 takes the faces they lie on, so the body is R21's alone; SMLib
fails that blend (status 1001, fillet error 4, Unable to Trim Rail Curve) and the test is
ignored until it builds. It replaces the refusal test that expected TooLarge.

Ruling d8 (2026-10-30).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Tests: a 1 degree crease rounds, and the knife edge's chamfers build, on SMLib smlib-batch-4-rulings

With SMLib's smooth limit at half a degree (422e6f1) the 1 degree crease rounds R2 and R20
(the test is no longer ignored), and with a cusp's chamfer the ball of radius d (d231547) the
The 0.5 one stays, for its own cause (the chamfer face left apart from the body). Two corner
cases the same SMLib (replay/1.0.1 and those commits) builds come off too.

Needs SMLIB_REV at that SMLib once it is pushed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* kernel, doc: an overflowing round goes first; an edge it removes is gone, and the fillet says so

SMLib (smlib-batch-4-rulings 0e83cca, SMLIB_ABI_VERSION 5) rounds the edges whose rounds overflow
their faces first and then what is left of the others; smlib_blend_dropping says which picks'
edges the overflow removed whole. The smlib crate binds it (Brep::blend_dropping), the kernel
returns those picks (fillet_removing), and a fillet notes them as its warning: "the edge at
(0, 0, 10) was removed by the larger round, so it is not rounded" (a 50 x 50 x 20 block, R21 on
its bottom front edge with the front left upright: the block is R21's alone, in closed form).

Ruling d8: [21, 5, 5] on a 20 cube builds as R21 alone (6107.529370); the test is no longer
ignored. R25 with R2 at a corner (blend_corner) builds the same way, R25's alone in closed form,
where it was refused. The rig's 1080 notched block outline at R8 builds now and comes off the
known-failing list.

Needs SMLIB_REV at that SMLib once it is pushed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Fillet rig: the knife edge's 0.5 chamfer builds

SMLib's trim took a fillet meeting its face at more than 78 degrees for one with no side
(smlib-batch-4-rulings 971e567); the chord of the knife edge's 0.5 chamfer meets the flat face at
81 degrees. The case comes off the known-failing list.

Needs SMLIB_REV at that SMLib once it is pushed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Tests: a 15° tapered text builds, every letter grown, exact where its outline gives it

Owner ruling d6: the letter t of "tres long" was drafted inward by SMLib (its
loop's normal, summed from the cross products of its chords, pointed down;
fixed in SMLib). With the fix every letter grows, and the join builds at 3°
and 15° by the Round rule and by the default rule (tangent for spline
outlines), where before 15° was expected to be refused.

Exact where the outline derives it: a letter's section at height z is its
outline offset by e = z tan a, by the Round rule of area A + P e + C e² while
the offset trims nothing (Steiner; C half each loop's turning less
tan(t/2) - t/2 at each reflex corner), so the letter is
A h + P k h²/2 + C k² h³/3. Each extruded alone, the t at 3° and s, l and o
at 3° and 15° are checked to 1e-6 (SMLib measures them within 1.3e-9; its
mass properties integrate to 1e-8). The concave joins of r, n, g and e are
tighter than e already at 3° (their swallowtails trimmed: r 1.1e-3 under the
formula), the t's by 15° (9.9e-5 over), and on the block the letters grow
into each other, so the whole word is checked for what holds exactly: one
closed body with the block, and half-way up every point of each straight
letter's outline inside the solid (the t drafted inward failed that).

With SMLib's fix (smlib-batch-4-text-loops 82133b2).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: a sloped basin's mitred rim rounds are checked tangent along their contact lines

1080-fillet-sloped-basin-1-mm-rim-r0-5-4-edges asked G1 at every blend seam. Its front wall
leans 19 over 20 and the others 10 over 20, so at each rim corner the two mitred rounds meet the
sloped corner edge at different heights (0.155625 and 0.276393 below the top): the deeper round
runs on past the corner edge and is cut by the plane of the other's wall, a crease by
construction, the gap below the shallower round filled on that plane. The check now measures
the rounds' contact seams, on the top and on each wall, where the rounds are tangent. The
volume is unchanged.

With SMLib smlib-batch-4-mitre (1c15364): the case passes (V 144530.427946 against
144530.427909, 15 faces, sound); before it, only the four corner creases failed the check.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* smlib: Brep::classify_point says exactly where a point lies against a body

Binds the fork's smlib_brep_classify_point (SmBrep::Point3DClassify at the body's own zone
tolerance): inside, outside, or on the boundary within that tolerance of a face, an edge or a
vertex. A box and a cylinder just off its curved wall in measure's tests.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NvoxKQ2rEZvcXNQnhE3Sgs

* kernel: Body::classify_point decides inside, outside or on the boundary exactly

SMLib's point classification at the body's own tolerance, on the exact faces and edges, in
place of a mesh at a chord of 2e-3 of the body's size (about 0.4 mm on a 200 mm part). A point
within the tolerance of a face, edge or vertex is OnBoundary, never decided by a ray. A mesh
body's triangles are its exact shape: on them within the record's tolerance, else by ray
crossings. The kernel's own probes move to it; Body::contains goes once its last callers have.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NvoxKQ2rEZvcXNQnhE3Sgs

* doc: a hole decides whether it exits at its drill point's shoulder exactly

Material past the depth needs a body with the point inside, air before it needs every body to
have it outside; a point within a body's tolerance of the exit face is neither.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NvoxKQ2rEZvcXNQnhE3Sgs

* engine tests: probes ask for the exact point class

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NvoxKQ2rEZvcXNQnhE3Sgs

* ui, cli tests: probes ask for the exact point class

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NvoxKQ2rEZvcXNQnhE3Sgs

* kernel: Body::contains, the mesh test, is gone

Every caller asks Body::classify_point now.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NvoxKQ2rEZvcXNQnhE3Sgs

* kernel tests: the rim chamfers running off their rims build, on SMLib 737c392

chamfer_wider_than_sloped_basin_rim_builds and a_rim_loop_chamfer_running_off_its_rim_builds
waited for SMLib to build the cliff overflow (CloseRemovalRegion refused it,
SM_FILERR_NO_ROOM). The pinned SMLib builds it (eddf0ede4, bba9914b6): both pass on aarch64
Linux and are no longer ignored.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Fillet rig: the cases SMLib 737c392 builds come off the known-failing list

The gate on aarch64 Linux reports 16 fillet cases FIXED with the pinned SMLib: the sloped and
equal-slope basin rim chamfers and rounds past their rims (1080-*, 5835-*), the pocket rim
chamfer past the floor, and the three pocket rim rounds now mitred at the pocket's corners
(every-pocket-*). No other area's gate reports a case fixed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* kernel: the re-blend's removed picks formatted as rustfmt lays them out

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* kernel: blend_smlib laid out by the 2024 edition's rustfmt

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Skills: the tests skill's unit-test example stays under the 3600-character limit

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Fillet: the radius-too-large choice in plain nouns

"If too large: Stop and say so / Largest that fits / Narrow where needed"
becomes "Face limits: Exact radius / Largest radius / Narrowed radius", in
every language of the catalog, with the error and the tip that name them.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* kernel: a round that would cut through its wall into a hollow is refused

The owner's rule of 2026-10-30: a round running past a rim of its faces
into open air builds; one whose section would remove material through to
a hollow behind its edge (a slit into the hollow, a cut through the part)
is refused as too large, so the fillet names the largest radius that fits.

Decided exactly in the section of a straight convex edge between planar
faces: the round's centre is at depth r below both planes, and it opens
where its arc uncovers a boundary point strictly inside the wedge with no
material beyond it toward the centre (the body's exact point
classification). Looked at: straight edges parallel to the round's (a
hollow's corner), cylinders about parallel axes (a rounded corner), and a
picked concave parallel edge's own round, which fills the corner.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Fillet rig: 1668's R7 past the inside corner is refused at 4 + 2 sqrt2

The round's centre (32 - r, 42 - r) leaves the hollow's corner (30, 40)
outside its circle, opening the wall, where sqrt2 (r - 2) = r: the case
expects the refusal naming 6.828. 2852's R30, refused naming 25.24, now
passes too: both come off the known-failing list.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* xtask: known-failing grades a nextest junit report against tests/KNOWN_FAILING_TESTS

`cargo xtask known-failing --junit PATH` reads nextest's junit (a file, or every .xml under a
directory, so all of a run's junit artifacts at once) and fails on a test that failed and the
list does not name, and on a listed test that passed (FIXED: take its line off, as the kernel
rig's gate does). A listed test that did not run is neither, since a pull request tests only the
packages it touches. The list has the rig's KNOWN_FAILING format (`<test> <cause>`, sorted, the
rig's own parser), with `<test>` nextest's `<binary id>::<test name>`; `cargo xtask lists`
checks it is sorted. The list starts empty.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* CI: the test jobs grade their junit against tests/KNOWN_FAILING_TESTS

Native Linux tests and the GPU path test shards no longer fail on any failed test: nextest's
"tests failed" exit (100) goes to `cargo xtask known-failing --junit`, which fails on a failure
the list does not name and on a listed test that passes. Any other nextest exit (a build error,
a bad filter) still fails the step at once. The tests skill says to list a failing test with its
cause, in place of ignoring it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Tests: the known-failing list takes develop's 10 failing tests

Develop 38fd4a4f6 (CI run 41157648345, after the SMLib eaa998a pin) fails 10 cargo tests; each
is listed with the cause its failure message gives. Seven are SMLib refusals or results off
their derived volumes (fillet, chamfer and shell), the cli action matrix and naming corpus
among them; the two corner_g2 tests are the fitted G2 corner patch bulging past the box, which
the explicit G2 corner fixes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Reference tests: the eight cases that fail on SMLib eaa998a name their cause as known failures

Develop 38fd4a4f6's Reference tests fail 34, 35, 77, 78 and 81 (STEP export finds fillet edges
4.6e-5 to 6.1e-5 off their surfaces against 2e-5), 37 (the apex blend 0.193 off), 80 (the
taper-to-fit area) and 83 (thicken status 1001). Each expected.json names its cause in
known_failure, which the oracle already reads: a known failure that passes fails the job.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Sweep path: a curve of several pieces is turned round whole

The path chain oriented its first piece, not its first curve: a first curve
of several exact pieces (a fit spline, an ellipse, a projected free-form
edge) was never turned round, so a path whose next curve meets it at its
start was refused as "not connected end to end". The chain now orients
curve by curve: the first runs toward the second where they meet at one
end only (else from the end nearest the profile), each next one starts
where the previous ended, with one joint tolerance for both questions.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* SMLib: pin 2fc375f (tangency-labs/smlib#71, #72, #73)

A sheet's side is extended exactly the distance asked, or the call says it cannot (#71);
NVIDIA's -Werror build passes again (#72); a body splits by a sheet of several faces as it
is, smlib_split_body_by_sheet (#73). The C API is ABI 6 (the new entry), so the binding
expects 6.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* kernel: Extend grows a surface of several faces face by face

A surface body of more than one face (a fit-point spline's spans extruded, #3830) was refused.
Each face with a picked edge is now extended alone by SMLib's extension of its surface, and the
faces are sewn back together, each the face it was. A side that lengthens onto a face meeting it
tangent takes that face along, past its edge continuing the picked one; across a crease the
neighbour grows only when its edge is picked too, otherwise the extension is refused. Faces whose
lengthened sides no longer meet are refused. Up to a target keeps the pieces every grown face
lies in.

3830's extensions now build; its split still fails (the tool point picks one span, and a face
tool cuts with that face's surface alone), so it stays on the list with that reason.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* kernel: Split Body by a face of a smooth sheet cuts with the whole sheet

A face tool cut with that face's surface alone, so a face of a surface body made of spans joined
tangent (a fit-point spline extruded, #3830) cut nothing where its span ends inside the body.
Split Body by a face of a surface body now takes the faces joined to it tangentially (the exact
check Extend uses, within the one angle for smooth) and slices with them as they are, through
SMLib's slice by a sheet (smlib_split_body_by_sheet, tangency-labs/smlib 7a01f1b), its sides those
of the picked face. A sheet that only nicks the body cuts nothing, as before; a lone face, or a
face of a solid, still cuts with its surface extended.

3830 comes off the booleans list. Its answer counted the two halves of the cup and not the surface
that cut it, which stays as a body of its own (4879 removes its surface for that reason): three
bodies, the volume unchanged.

Needs SMLib 7a01f1b pinned (SMLIB_REV, prebuilt; its SMLIB_ABI_VERSION 6, which crates/smlib now expects) to link.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Tests: a cylinder strip's run-offs are 100 pi + 100 to 1e-6, SMLib's exact area

The tangent and curvature run-offs of the quarter cylinder r 10 strip, 5 across the surface,
are its section swept 20 along z: exactly 10 pi/2 20 + 5 x 20 = 414.159265. measure is SMLib's
integration on the exact surface (1e-8), not a mesh, so the run-off check is 1e-6 relative and
the 'measured on meshes' notes go. Red until SMLib's extension is solved to its distance
(smlib-extend-converges 590d8a1): eaa998a's curvature run-off is 4.986 long, area 413.882.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* kernel: an extension SMLib cannot make exact says so, naming the length asked

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Sketch: a control spline can carry its own knots and weights

A sketch spline over control points was always the clamped uniform B-spline
over them, so a free-form edge whose knots are not evenly spaced (or whose
weights are not 1) could only come into a sketch as one curve per Bezier
span. A control spline now may save `nurbs`: its clamped knots and its
weights. Its points, polyline, exact pieces (one a knot span; conics for a
rational quadratic), the solver's distance and curvature, links, parallel
re-projection and inspect.sketch all read them; a loaded design's are
checked. One pending format file (spline-knots).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Sketch projection: a free-form edge comes in as the one exact spline it is

A projected spline edge of several Bezier spans came into the sketch as one
curve per span (#1508), so a G2 round's bottom edge was 16 curves and a path
naming the corner named one span of it (#3604). The kernel now reads such an
edge (a spline, or an elliptical arc of several spans) as the one clamped
B-spline it is over the edge's interval, its knots and weights its own, and
the sketch takes it as one control spline whose ends sit on the edge's
vertices. Planar regions still read edges span by span. The 3604 case builds
with its original path and leaves KNOWN_FAILING.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Fillet rig: blends past an outer rim of their faces build, on SMLib smlib-blends-past-rims

The owner's ruling: a round or chamfer that runs past an outer edge of its face into open air
builds, tangent where it has room and cut by the face beyond where it runs out. SMLib's branch
smlib-blends-past-rims builds them by the rig's analytic answers. Off the known-failing list:
the disc's rim past its 2 mm wall (190), the hole's top rim R12 past the plate's bottom, the
chamfer past the cylinder's bottom (2893), each closing on the wall continued as itself, and
926's chamfer ending on the boss's wall, which now builds to its answer too.

The sloped basins' R4 rim rounds (1080, mitred rounds built together where one overflows) build
to their analytic volumes (to 6e-9), and stay listed with that reason: their continuity check
(g1 at every blend seam) takes in the mitres' creases, G0 by the mitre ruling.
2893-chamfer-two-distances-elliptical-rim is graded as a solid (the owner: it builds) and stays
listed, waiting on the ruling on unequal chamfers along a rim whose angle varies.
5158-fillet-round-just-past-a-10mm-face stays listed with its cause found: the round passes
5e-6 under the box's top back edge, inside the body's tolerance.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Naming: test that a round of a round's end has no seam with that round's faces

The r1 round of a block's top edge, chained onto the end curve of the r2
round of its arc at a cylinder (rig fillets/every-block-cylinder-crease-
top-edge-after-arc-round-r1), ends where it runs into the cylinder. That
edge is no seam of the r1 round: the cylinder is a face of the r2 round,
not of the r1. blend_seams takes it for one today (the r2 round's name
F3:blend:F1:+z|F2:side holds a | of its own), and the rig reads it as a
contact seam at 60 degrees.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Naming: a blend's seams split its edge's name at the | between its two faces

A blend face's name holds the edge it replaced, <face>|<face>, and a side
of that edge can be a blend face whose own name holds a |. Whether a face
is a contact of the blend was read as the edge's name starting or ending
with the face's name next to a |, so the r2 round's own face F2:side read
as a contact of the r1 round F4:blend:F1:+y|F3:blend:F1:+z|F2:side. The
edge's name is now split into its two faces, a blend side read whole as
the two faces of its own edge.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Fillet rig: five refused blends build, off the known-failing list

926 (the chamfer ending on the boss's wall), the oblique cut's r0.1
round, the every-block r1 round (its seam with the cylinder is no
contact seam, after the blend_seams fix), the up-down squares' concave
chamfers and the shelled box's lip chamfer build with SMLib branch
smlib-blend-refusals (f8650b8 and later). 1207 r0.5 passes SMLib's check
on its rails now and stays listed for the next cause: its end rib on the
boss, which starts where the round and the boss are tangent.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the trough and the pocket whose floors fall to a drain build

SMLib now offsets a vertex where four faces meet (two walls and two facets at the floor's
corners, four facets at the drain) to two vertices joined by a new edge (tangency-labs/smlib
branch smlib-shells). Both volumes match their closed forms. The pocket's lining, hanging from
the top's rim round the pocket, stands apart from the slab's outer walls with the cavity rising
between them to the open top: the case says two shells in the one body.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the plate with a hole through its arc side shells open at its far end

SMLib's merge now intersects the offsets of an edge's faces where they could not be mated along
it (the hole's offset poking out through the half-round side's; tangency-labs/solidcore#1795):
one closed valid solid with the one face open, as the owner ruled.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the ring cut frame and the up-down join shell

SMLib's inset now keeps only the regions that are cavity, not the one the offsets close where
they overlap (between the two faces of #545's 0.5 floor, round the up-down join's two empty
octants). The frame's core under the recess is hollowed to a sealed void, so the case says two
shells and its closed form, 21044 mm³.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: #616's lid shells open at a rib's side, with the 0.15 lip

SMLib keeps the skirt top's offset where its loops wind the other way under the opening, the
cavity's floor (the owner's ruling of 2026-10-31: build it; a lip thinner than the wall is no
break-through).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Known failing tests: the action matrix's shells and chamfers as they fail with SMLib batch 7

The ring cut frame's and the up-down join's shells build now, and so do the up-down join's and
modern-plastic's chamfers (SMLib smlib-blend-refusals). #616's lid in the corpus still fails
SMLib's loop check, the shelled box shelled again at an inner wall is unchanged, and the
oblique cut of a cylinder's fillet still fails SmSurfOfRevolution's Test #7: 3 actions of 7.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* SMLib: pin 26cc451 (batch 7: rounds past a rim, blend refusals that build, shells)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Known failing tests: the shell after a cut whose end touches an arc side builds on SMLib batch 7

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Surface fair test: the energy falls strictly, the face moves, and the edges are read after the fair

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Sketch scaling test: the first dimension sets l1 to 40 and scales both joined rectangles about the origin

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Frames tests: a pick on a moved component or an instance removes the exact round and bore volumes

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Issue 1080 tests: the rib's rounds and the mitred corner chamfers and rounds have their exact volumes

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Engine tests: lip_after_boss checks each lip's and groove's exact band, hole_as_deep_as_the_plate each hole's exact bore

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Robustness tests: lip_after_boss and hole_as_deep_as_the_plate live once, in the engine's tests.rs with exact volumes

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Engine tests: loft_cut goes; loft_spacing's hollow elbow covers its section counts and more

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel loft test: the square-to-circle loft has the oracle's exact ruled volume, without printing

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Tessellation lens test: the lens's triangles fill exactly the rhombus through its arcs' midpoints

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Pattern test: six chamfered holes take exactly five more holes' bore and chamfer rings

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel pocket floor fillet test: the corner rounds and the floor loop round add their exact volumes, by Pappus at the corners

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Shell regression: the saddle-corner shell opened on the top box's side has its exact volume

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* SMLib history test: every operand's order is the identity on its record's faces and edges

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Boolean history tests: the SMLib module's box-minus-sphere, box-minus-cylinder and box-union-box copies go; the top-level ones check more

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Modify history tests: shell, offset and thicken are one test each through the public API, with the volumes, refusals and history the SMLib copies checked

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Chamfer history test: the chamfer after each box cut is exactly one face on the painted block too

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Grid test: a pole side's face is the five triangles fanned from its one grid node, none degenerate

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel tests: the twelve ignored print-only debug tests and surveys go; coplanar_curved_booleans checks the flush-capped cylinder cuts and unions they printed

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel: the #883 board fixture generator is an example, not an ignored test

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* STEP import test: a seam sphere whose meridian stops short of the pole imports as the whole 2 mm ball

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Material tests: the nine copies in material_properties go; the persistence file checks the library copy's card and formats 20 and 21 in one loop

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Sample test: one loop checks every sample builds cleanly and without a panic; the robustness copy goes

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Known failing: the non-antipodal seam sphere import, refused for a loop of no (u, v) area

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Tests: rustfmt the board fixture example and the coplanar asserts

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the two 3871 cone cases were byte-for-byte the 258 cases; the 258 cases carry their sources

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: a volume or area can be graded as a hand-derived range where analysis only bounds it

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: fillets keeps one case per regime; 17 replications of a case in the same regime go

The sloped-basin chamfers and rounds at 0.5 mm rims repeat the 1 mm rim's regimes; two of the
three knife-edge chamfer tips, the r1 and box r3 curvature corners, the r1 tangent corner, the
1 mm pocket-rim and hole-rim chamfers short of the floor and the r1 together-round each repeat a kept case's regime
and grades. #854 moves to 851-chamfer-arc-top-edge-9-5, and the cases that cited a dropped one
cite the kept one or state its formula.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: five cases outside fillets repeat a kept case's regime and go

A square cut is the slanted plane's ordinary crossing; the cam's half turn is the quarter turn's
contact at a symmetric angle; the r15 slab stands inside the same coincidence tolerance as the r10;
the 90 degree unfold and the 40 degree flange repeat the 72 degree and 20 degree cases' formulas.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 3243 rim and tooth join graded on its volume (rim plus the trimmed tooth less their 0.01 mm overlap)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 3782 emboss on a cylinder graded on its volume, area, faces and box (the wrapped 4 x 5 badge raised 1 mm)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 3604 G2 swept-tool cut graded on a volume range (filleted box minus the swept corner section)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 4505 coil joined through a cylinder wall graded on its volume, area and box (cylinder plus the wire's screw-swept segment outside r 76)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 4879 window cut through a thickened spline revolve graded on a volume range and its faces

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 4623 join of crossing pleated loft skins graded on its volume and box (both skins less their overlap)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 831 tapered glyph extrude graded on a volume range and box (the font's S drafted in by 10 deg)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 2357 ring of 90 cylinder segments graded on its volume and box (the segments less their 90 joint overlaps)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 4542 fluted wall joined on a revolved base graded on its volume and box (the base by Pappus plus the fit-spline ring)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 1985 and 2243 STEP-export enclosures graded on their volume and box (the outline less the cavity, plus gusset and standoffs)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: both 4092 loft, ribs and pad unions graded on their volume and box

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 1525 graded on its volume (the convex body's exact 2 mm inward offset), faces and box

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 1494 graded on its volume (box and cylinder less the 1 mm-offset cavity, sharp to rounded concave edges) and box

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 4492 graded on its box and a volume range (the face's 19 mm sweep added, within the block's sides)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 745 graded on its volume (the end moved 1 mm plus the spline boss's exact area), faces and box

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 4567 graded on its volume (the revolved spline less its revolved 2 mm offset), faces and box

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the up-down squares shell graded on its volume (the columns less the 0.346 mm-offset cavity, sharp to rounded concave edges) and box

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 545 graded on its volume (the frame less its 0.5 mm-offset cavity, sharp to rounded concave edges), its sealed pillar void and box

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the shell-again case graded on its volume (the shelled box less the 0.5 mm-offset cavity in its walls) and box

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the arc-sided plate shell graded on its volume (sharp to rounded at the hole's bottom), faces and box

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 616 graded on its volume (the lid less its 0.5 mm-offset cavity, sharp to rounded concave edges) and box

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 391 graded on its volume (the shelled housing, sharp to rounded at the flange, less the letters' 0.6 mm deboss) and box

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: shell/3576 was the ladder's drafted box with rounds shelled, recipe for recipe; the ladder case grades it and carries #3576 and #3573

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the two 851 chamfers of 2 graded on their volume (the cone's sweep, and its union with the front edge's prism)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the chamfered corner of the lipped shell graded on its volume (the chamfer's triangle through the wall)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the concave round into the peg's cylinder graded on a volume range (the torus it must add, the band it can't leave)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the two folded 72 degree trays graded on their volume (flat area x thickness) and the walls' reach

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the flanged rhombus graded on its volume (flat area x thickness, all four reliefs cut) and its walls' ends

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the base with two adjacent flanges and a hole graded on its volume (base, bends and panels folded)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the flange beside the ledge graded on a volume range (the flange's two readings of its length)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the two G0 window patches graded on the box the part they close keeps

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the hub's six swept grooves graded on its volume (the revolved hub less each tube's part inside it)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the cut mirrored loft graded on the box its sections pin

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the three modelled-thread joins and the fine M48 graded on their volume (the basic profile's groove, Pappus)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the oblique-cut round graded by its seams (tangent, not curvature-continuous); the flanged rhombus says what it tests (its reliefs stay apart at 60 mm)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 3841 is 168's rim touching its coin along a circle, and 4092's reduced recipe is its report less a flange clear of the failing join; the kept cases carry their sources

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the oblique-cut round keeps its old grade until the new blend algorithm lands

Its seams grade needs the continuity check to skip the body the round is not on. Fillet cases
wait for the new algorithm.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: 1985's enclosure is 78 x 77 (74 and 73 inside, 2 mm walls); its volume 38063.827 by hand

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the three spline walls' face counts wait for the one-wall decision; their volumes grade them

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* STEP import: a face bounded by seams alone runs its loop the other way, never turns over

A loop that uses each of its edges once each way (a whole torus within its tube and ring seams)
is the same set of uses run backwards, so it says nothing of the face's side. One written
clockwise in (u, v) turned the face over, against its same_sense: the torus came in inside out
with a warning. Such a loop is now reversed and the face keeps its side.

Test: seam_only_torus_imports_as_a_whole_ring, both orders of the two seams, volume 2π²Rr².

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* STEP import: a face bounded by a slit alone is the whole of its closed surface

A loop that runs out along a chain of edges and straight back (a meridian from a sphere's pole
stopping short of the other, out and back) encloses no (u, v) area, yet on a closed surface it
bounds all of it less a cut of none. The reader only read such a face when the slit already ran
pole to pole (and moved the sphere's frame onto it only then), so the 2 mm ball whose meridian
stops at 2.5 rad was refused for its empty loop.

The slit is now made the face's seams, so its loop bounds the surface's whole domain: on a
sphere, a revolution whose profile ends on its axis, or a spline closed round u that collapses
at both ends, the slit's meridian is run on to both poles, out and back (a sphere's frame put
on the slit first, whichever two points it joins); on a torus, the circle the slit lies along is
closed and the other circle through its start is the second seam. The seams on a sphere are its
own meridians (angle 0 on the equator), so their pcurves are exact lines on its form.

Tests: non_antipodal_seam_sphere_imports_as_a_whole_ball passes (its known-failing line goes);
seam_only_torus_imports_as_a_whole_ring adds a torus bounded by a slit along its tube.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Rig: the test-peg's concave arc and front top edge rounded together stay within the body's box

A new fillet case, report-fillet-concave-arc-and-front-top-edge-together-solver-residues:
the reported part's bar (with the sketch solver's residues) whose sides run tangent into the
cylinder, its concave arc and front top edge rounded r2 together. The plane of the front face
bounds the whole body, so every blend lies within the unblended body's exact box (SMLib's tight
box); graded to 4e-5, the corner patch's own tolerance. On SMLib 2fc375f the corner patch
stands 0.0374 out of the front face (box y min 0.0205): listed in KNOWN_FAILING.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Rig: the test-peg corner at r5 and r7

Two more fillet cases, the reported part's concave arc and front top edge rounded together at
r5 and r7, with the same exact-box containment check (graded to 4e-5, the corner patch's
tolerance). On SMLib 2fc375f the corner patch stands 0.128 (r5) and 0.230 (r7) out of the
bar's front face (box y min -0.0699 and -0.1723): listed in KNOWN_FAILING.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Rig: blends that end where their face runs tangent into another

Fifteen fillet cases on the test-peg's shape with clean numbers: a bar 40 x 16 x 10 whose far
end is a half disc R8, a disc R8 standing 10 on its top, joined, so the front face runs tangent
into the cylinder along x = 40 and the front top edge (convex) meets the concave arc of the top
tangentially there. Each blend ends on a face tangent to the one it rolls on.

- Rounds: both edges together at r2 and r5; one after the other in either order at r2, and at
  r1.5 and r3; the front edge alone as a curvature round, whose end on the cylinder is a
  semicubical cusp. The containment invariant (the unblended body's exact box, to 4e-5) and
  G1 seams, or the analytic volume where the blend has a closed section.
- Chamfers: the front edge alone and the concave arc alone (analytic volumes: the section along
  x up to the wall, the cone's spandrel ring clipped by the sides), both together, and one after
  the other (1 then 2).
- STEP: the peg written to STEP and read back, its front edge rounded (analytic) and both edges
  rounded together.
- A cone: the peg drafted 5 deg, whose front face runs tangent into a cone along a generator,
  its front edge alone and both edges together.

The quadratures are tanh-sinh, converged to 1e-13; SMLib's own volumes of the passing analytic
cases agree to 4e-9. A sphere is left out: a plane meets a sphere tangentially at a point, not
along a seam, so it is not this ending.

On SMLib 2fc375f eight fail, listed in KNOWN_FAILING: every chamfer that ends tangent (status
1001, SmFilletNx1Corner), the two rounds together (the corner patch 0.0374 at r2 and 0.128 at r5
out of the front face, and the rounds meeting with normals 180 degrees apart where their
contact lines on the top cross), the same on the STEP-read peg (status 1001) and on the drafted
peg (its corner refused as unsupported).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Rig: fillet corners graded curvature-continuous, the default to be

Six fillet cases that pick their edges with no type, so they take the fillet's default, which the
owner has ruled is to be curvature-continuous (G2) at every corner, with circular rounds kept as
an option. Each grades every seam of the rounds and of the corner patch G2, on one valid body:

- an oblique convex corner (a block drafted 15 deg);
- a four-edge and a five-edge apex corner (pyramids lofted to a point);
- a pocket's concave corner;
- a mixed corner (a step: one concave edge, two convex);
- a corner beside a cylinder (the top corner of a half cylinder).

All six fail on SMLib 2fc375f, listed in KNOWN_FAILING: today's default is the circular round,
whose seams jump in curvature by 1/r. On the five-edge apex the corner patch also meets two
rounds at 8.45 degrees (G0).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Rig: the test-peg's front chamfer builds

With the SMLib fork's smlib-tangent-ends (a rail's end on its side edge is the rail/edge
crossing), report-peg-front-chamfer-d2 passes. Its line leaves KNOWN_FAILING, for the pin that
carries that fix.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Rig: the curvature-continuous corner cases pick the curvature fillet

The six corner cases added in 3a854104a took the fillet's default type, on the premise that the
default would become curvature-continuous. The default stays the circular round, so each case
now asks for the curvature fillet itself (type curvature: curvature-continuous sections and
corner patches) and keeps its grading: every seam of the rounds and of the corner patch G2, one
valid body. They are renamed from cov-fillet-g2-default-* to cov-fillet-g2-*.

The oblique three-edge corner (a block drafted 15 deg) and the pocket corner pass. Four fail, as
KNOWN_FAILING now says:

- the corner beside a cylinder: the curvature fillet builds its sections only between planar
  faces;
- the four- and five-edge apex corners: refused because the default setback is shorter than
  where the rounds' contact lines meet (3.90 and 3.51 mm from the apex);
- the mixed corner: SMLib sets back a corner only where its edges are all convex or all
  concave.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Rig: the test-peg's concave chamfer, alone and after the front chamfer, builds

With SMLib's tangent end as an event of the fillet's trace (blend-end-event 6d73e4f), the
concave chamfer's rail on the cylinder ends where it crosses the seam along which the front
face runs tangent into the cylinder, continued past the vertex, and its end rib on the front
face is built from there. report-peg-concave-chamfer-d2 and
report-peg-front-chamfer-d1-then-concave-chamfer-d2 pass and leave KNOWN_FAILING. The two
chamfers together still fail, now at the mixed-convexity corner where they meet: its line says
so. Rig against SMLib 0cc48bd: no area regresses (963/993 pass, 961 on a87c502), ladder 168/168,
kernel lib only its listed failures.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: constant rounds of the exact classes, and a tee's round

A plane cutting a cylinder or a cone at an angle, a sphere or a cone coaxial with the other face, a
cone and a cylinder touching one ball, two pipes touching at a point: their rounds' centre lines
are rational, so the rounds are rational surfaces of degrees fixed by the construction ((4, 2),
(2, 2) revolved, (6, 2), (12, 2)). The tee's centre line is not rational; its round is graded on
its volume. Volumes from the divergence theorem over the exact pieces, at rel 1e-6. The six that
SMLib 0cc48bd traces are listed in KNOWN_FAILING with what they come out as.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the exact-class rounds pass on SMLib's classified exact rounds

The rounds now come as the rational surfaces their classes give, closed ones as two faces split
at their seams, the touching pipes' as four: one splines entry a face, each of the class's degrees.
Their KNOWN_FAILING lines go.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* SMLib: pin 7ac037e (the new blend's tangent-end event, zero-speed G2 ends, and its exact-class rounds)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the fillets known-failing list is the 22 cases that fail on SMLib 7ac037e

Twelve listed cases build on the new blend's end event and exact classes, or were already
built on develop: they come off the list.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: #316's bottom rounds as large as the wall, or larger, shell with a step wall at the opening, not refused

Where the round meets the open bottom the cavity's wall steps down along the bottom's normal, so the
round meets it at 90 degrees and the opening is exactly the picked face. The cases are renamed and
graded on the exact volumes, and come off the shell known-failing list.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Merge fixes: every crate and test target compiles on the merged tree, but for crates/geom's hlr

world_state, view::bounds, home_camera and scene return Result on develop; new
main code takes that with ? or session_world. New fixtures carry the spline's
nurbs. ForeignCopy::of returns Result. occurrence_placement_with walks the path
with its moves, since the placement index holds pending_moves only.

The hlr port is separate and still to come: this commit does not compile as it
stands in crates/geom/src/hlr.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk

* Drawings hidden lines: read develop's B-rep records

Edges come from the record's edge list, cut to their interval (C3::of_edge), and
faces from its face list; a cylinder, cone, sphere or torus chart takes its bounds
from where its NURBS form lies on its label (S3::of_face). A rigidly placed
revolution has main's revolution chart; other placements stay refused. Moving a body
to its centre goes through Brep::moved. Loops are read against the natural normal of
the face's parameters where a chart wraps.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk

* Hidden-line fixtures: build develop's records from the fixtures' data

Fixtures keep their bounded curves and surfaces and their own edge numbers; a private
builder makes the canonical record, and the local prepare puts the edges back in the
fixture's numbering. Assertions are unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk

* Hidden lines: tests of the bounded edges, face charts and placed revolutions

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk

* drawing-hlr doc: what is drawn and refused on develop's records

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk

* hlr tests: a placed revolution is refused when its placement is not rigid

The new model has one record for main's Revolution, which was drawn, and its
PlacedRevolution, which was refused. An identity placement is now the former, so
this test's fixture takes a non-rigid one. The refusal it asserts is unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk

* Hidden lines: a traced intersection is drawn only where its spline lies on both surfaces

A leader off either surface is refused like a curve with no spline to follow. Tests: a traced rim
drawn on the circle and plane within tolerance, the same leader 1 mm off refused, and kernel bodies
with arcs through the production prepare against closed forms.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk

* Hidden lines: kernel bodies through the production prepare against closed forms

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk

* Tests from main: the world state is a Result here

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk

* A blend's pick measures close edges on their curves, not their polylines

A polyline can be a chord off its edge, more than two edges are apart near a
tangent point. A pick exactly on a ball's rim took the straight edge the rim is
tangent to, and solidcore built a chamfer of that edge with a clean check.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk
(cherry picked from commit c786db6bf4)

* Merge fixes from the first SMLib test pass

Model cache key covers the crates develop's kernel adds under doc (profile, smlib, step).
A piece taking a name also takes the feature that made it: SMLib lists a cut's pieces in another
order, and the swap left the name on a piece of the wrong feature.
A replaced analysis keeps its status slot. A command with no world picks does not place the model
first, so assembly.mass_properties answers for a flattening placement. FeatureCommon loses the
touching key the merge added; only the contracts that read it take it.
Two of main's tests listed as failing on SMLib, with why.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk

* Merge fixes from the ui test pass and the gates

Main's thread.direction keys go with develop's thread hand, which replaced them. Vietnamese for the
messages develop added while main made vi a complete catalog. CGA and 1-bit themes set develop's
comb and section hatch colours. A ribbon button with a persistent effect is pressed only while its
tool runs, so persistent_ribbon_pressed alone decides how its effect is shown; other analyses stay
pressed while on. Two of main's ui tests listed as failing on SMLib, with why.
Gates: fixed thread tables become slices, one collapsed if, rustfmt.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk

* A silhouette takes its facing from the surface's normals

A flat triangle beside the outline can tip the other way from the surface it approximates, and its
edges joined the outline. On SMLib's sphere that made 192 segments instead of the equator's 64, 100
curves and 95 profiles from the tight fit of #7989. Each corner's normal is used where it is close
to the triangle's own; a mesh without such normals (none, zero, not numbers, or averaged across
creases) keeps the triangle's normal, as before.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk

* Clippy and rustfmt on the merged tree

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MNXLXkZLQ6T7TSi4wswcmk

* Stepwise open checks every feature under the session's mode

OpenSteps::start evaluated under the default check mode, and finish kept those results under the session's mode: an each_step or off session opened in steps recomputed nothing. start now takes the mode (the app passes its session's), and a mode that changed before finish evaluates the design again under the new one.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEu4Lr3moCCisdZShKz

* Clippy without SMLib: the planar-sheet edge reader is built only with it

Its one caller is in the smlib-only sheets module, so without the feature it was dead code and
clippy -D warnings refused the kernel.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEu4Lr3moCCisdZShKz

* SMLib: pin d01d7b0 (the new blend's one Gⁿ section generator: the chord, circle and curvature sections, and the variable rounds as the union of their balls)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the section orders, chord, circle and curvature section, on planes and against a cylinder

Eight fillet cases for the blend's cross-section by its order of contact with the faces, each
graded on its analytic volume at rel 1e-7 and its exact continuity (the order, and not the next):

- a box's top back edge chamfered 3 (G0), rounded r3 (G1) and curvature-rounded 3 (G2): the
  areas cut off the corner s^2/2, (1 - pi/4) s^2 and 113/756 s^2, the last by the default
  thumbweight's definition;
- a pipe R10 dipping into a plate, its concave edge chamfered 2 (the chord between the contacts of
  the ball SMLib rolls for that setback, radius 1), rounded r2 and curvature-rounded 2, each
  section's added area by Green's theorem along the plate, the section and the pipe's arc; the
  order-2 section is built from the faces' feet of the point at distance d from both, which on
  the pipe are not linear in d;
- a boss root curvature-rounded 2, the square corner's section revolved (Pappus);
- a variable round r1 to r5 along a box edge (rel 1e-5): the balls' envelope is a circular cone,
  and the volume is the cone's sections integrated along the edge.

Three fail today and are listed: the two curvature rounds whose faces are not a square corner of
planes (SMLib's quintic blend curve, not the order-2 section: rel 6.5e-7 and 1.2e-5) and the
variable round (great circles, not the envelope's: rel 4.8e-5). The other five pass on SMLib
7ac037e.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the curvature fillets' expected bodies from the order-2 section at thumbweight 0

SMLIB_BLEND_CURVATURE's section is now the order-2 G^n section at thumbweight 0. Between two
planes square to each other it cuts A = 0.154192655384275 s^2 off the corner, its first moment
about either face 0.0291753295725360 s^3 (Green's theorem on the section): the curvature cases
built on that corner (862, 2708, every-box-edge, every-hole-top-rim, every-l-bracket,
every-block-cylinder-crease, the three-edges-at-bar-corner reports, 3604's fillet) take their
volumes from it by their own methods, the peg's front blend 23.3237146876 (its section along x
up to the wall). Between planes the section's t^3 term vanishes (their feet are linear in the
distance): the box edge case is exactly G3. The boss root is a certified fit, G3 to the plate and
G2 to the wall at Continuity Analysis's tolerances: at least G2. The boss root and the pipe on
the plate pass on SMLib's section and leave the known-failing list.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Curvature fillets: the section's default thumbweight is equal area, so the round's volumes are the circle's

On the round's rails the order-2 section takes off what the ball's circle does: (1 - pi/4) s^2 at a
square corner, first moment 0.0474559669443 s^3 (w = -0.5252123889). The kernel and engine tests
and the curvature rig cases take their volumes from it by their own methods; the pipe on the plate
from its own wedge (70.52878 degrees, w = -0.8544103493). The G2 corner patch's seam test is listed
until the corner step.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: four G2 corners wait for the corner step: their patch meets the new section's rounds G1 at one face seam

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the turn at a corner of two rounds, Mitre (the default) and Swept, at a pocket's floor and a boss's base, by exact volumes

A round's spandrel A = (1 - pi/4) r^2 and its first moment about its wall M = r^3 (5/6 - pi/4):
Mitre adds L A -+ 2 M per edge (cut at, or run on to, the corners' bisector planes), Swept turns
the section about a convex corner's edge, (pi/2) M a corner, and refuses a concave corner rounded
less than the ball. The pocket's tangent chain round its R3 corners is a torus each. Graded to
1e-7, where Mitre and Swept differ by 6.6e-6. The Swept cases are listed until solid.fillet
takes a turn.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: a junction family: oblique wedges 30 to 150 degrees rounded and chamfered (exact), bosses upright and tilted, pipe tees and a Y junction

Four are listed with their causes: the equal-pipe tee (its exact conic-pair rounds meet at the
ellipses' crossings in a corner the old corner code cannot join), the tilted bosses (1001, before the
blend rework too) and the Y junction (refused for room at its triple points).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Variable rounds: graded by the union of their balls (their envelope), not the balls' great circles

The smlib crate's laws and ladder 005, 006, 007 and 012 take the volumes the balls' envelope
removes; the law with a corner is its own test, listed as failing until the corner step.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel tests: variable rounds graded by the union of their balls, the removed volume to 1e-4 of itself

removed_by_balls integrates, in each plane square to a right-angled edge, the corner the union
of the balls leaves (by rays from the edge, the least entry over the balls), cut to the body's
section. The law with a corner through a middle point is its own test, listed as failing.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel tests: a spline radius whose balls fold over one another at its peak is refused

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: the variable round on its cone passes (the round is the balls' envelope)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Tests: three rounds of different radii at a corner listed: the corner's transition kinks with the balls' touch circles

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Ladder: the variable rounds graded to 1e-6 by the union of their balls; 012 by its law going on past the edge

At the default 1e-3 the great circles' volumes (4e-6 off) passed too. SMLib builds each within
1.6e-7. The golden takes the new volumes and areas; counts unchanged. The G2 corner test listed
as failing passes on SMLib's step 3 and leaves the list.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Reference tests: the variable rounds 77, 78 and 81 pass on SMLib d01d7b0; their known-failure marks come off

The new section builds their edges on the surfaces, so STEP export no longer finds them 4.6e-5
to 6.1e-5 off. develop's Reference tests (60393a718) report all three as passing while marked.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* STEP import: a ball's halves whose seams are off each other's quarter lines each form within one turn

A sphere split into two halves along a circle about its axis, each half covering a pole with no
seam, gets a seam per half from the first vertex of its loop to the pole. The circles between the
halves group their frames so they turn together, and when the two seams are not a whole number of
quarters apart no one turn fits both: the group kept its frames, a half's form ran past one turn,
SMLib's trim (SMLib and solidcore alike) across the form's end crossed the face, and the mesh
folded over the ball (a rotor's ball showed torn; its half had 196 times its area in triangles).

When no one turn fits the group, each spilling face is now turned on its own to start on a quarter
line. On the rotor file this also clears three other faces that were filled leniently and a
0.076 mm edge gap; regression test in step_in/tests.rs with the area and volume of each half.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEu4Lr3moCCisdZShKz
(cherry picked from commit b3188e005d427043b918267449c2336a1777e54f)

* SMLib: pin 5e8c9c1 (the new blend's corners: Ball and End/Join; exact fold and room checks with fit sizes; overflow in place; the round as its definition)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel tests: a law's dip shows its corner ball's sphere, a peak's two rounds cross (the reasons had them swapped)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Reference tests: the variable rounds 77, 78, 80 and 81 expect the union of their balls

A variable round is the envelope of the balls rolled along the edge, which takes less than a
quarter round in each section. The expectations are now that union, the law continued past the
edge's ends along its end slopes, integrated by Gauss-Legendre panels in x and t; 2, 4 and 8
panels agree to 1e-9 of the body's values, and 77's removed volume matches a brute-force union of
balls to 2.5e-10. The nearest-ball search now scans the balls the ray meets around ball x
(|x - s| < r(s) sqrt(sin 2t)) on their own, which near the faces and where 78's radius closes to 0
are narrower than the global scan's step. Regenerating keeps each case's known_failure. 79, a
constant radius, is unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Known failing: the law corners' kernel and crate tests and rig 1207 pass on SMLib 48ea616

SMLib builds a variable round's law corners as its balls' union (the corner
ball's zone at a dip, the sides' crossing at a peak): the kernel's corner
through a middle point (39.5410) and the crate's 4 -> 2 -> 4 law (39.5421) pass.
Rig 1207 (the bar's top edge rounded after the boss's arc, r0.5) builds and
grades on the same SMLib; it failed on c820346.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Kernel rig: every fillet and chamfer case rounds its tangent chain

The owner's ruling of 2026-11-01: the rig and the ladder round with tangent_chain true, how a
user's default pick behaves. The 38 fillets-area recipes and the ladder's 07-fillets/016 and 018
that passed false now pass true. The shell recipes keep theirs: a shell's tangent_chain is its
open-face chain, a different feature.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel rig: the partial-selection cases kept as -partial-chain copies, tangent chain off

Where a case's point is one edge of a tangent chain rounded alone, a copy keeps tangent_chain
false as its own case (SMLib rounds exactly the picked edges, ending at the smooth vertex); their
expectations stay as they were, to be re-derived:
- fillets/1207-fillet-box-end-edge-into-unblended-round-partial-chain;
- fillets/every-fillet-after-fillet-curvature-front-then-side-r3-partial-chain;
- fillets/1207-fillet-bar-top-edge-after-boss-arc-round-partial-chain, renamed from
  ...-no-tangent-chain: with the chain on it is exactly 1207-fillet-bar-top-edge-after-boss-arc-round,
  which the rig already has;
- ladder/07-fillets/020 and 021, the r5/r3/r7 and r5/r3/r2 corners (016 and 018) with the r3 on
  its edge alone: with the chain on, the r3 runs on along the r5 round's end arc and down the
  vertical edge the third round was to take.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel rig: the concave arc between the block's tangent sides kept with tangent chain off

fillets/every-block-cylinder-tangent-concave-arc-r2-partial-chain: the pick takes one piece of the
concave half circle, which SMLib splits; with the chain off its other pieces are left out, and the
blend is refused at the piece's end (the base gate's outcome). With the chain on the case rounds
the whole half circle. The copy keeps its expectation, to be re-derived.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel rig: two chains that now build graded by their exact volumes

With the tangent chain on, two cases round more than the edge they pick, and their old
expectations were the single edge's:
- 1207-fillet-box-end-edge-into-unblended-round: the r1 chain is the end face's top edge, the r3
  round's end arc and the back upright; 34 A + (pi/2) A (3 - c) by Pappus on the arc:
  V = 7953.139216 (it built 7953.139216, the old 7957.505 passed only inside 1e-3), 10 faces.
- 851-fillet-top-front-then-front-left-upright: the upright's chain runs up the first round's
  end arc on x = 0 and along the left top edge, so the corner becomes the R2 ball's eighth:
  V0 less the first round's spandrel to the wall (35.344727) less 27 (4 - pi) + 2 pi/3:
  V = 9530.777979 (built 9530.777987), 10 faces.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel rig: a curvature chain round a round's end refused for its size; the concave half circle builds

- every-fillet-after-fillet-curvature-front-then-side-r3: the side edge's chain runs round the
  first curvature round's end curve, whose radius of curvature has a weighted mean of 3 (its
  setback) and is unbounded at the ends, so somewhere below 3: the size-3 blend's contact line
  folds back on itself. Refused, "larger than the faces beside its edges allow", as ladder
  07-fillets/016's r7 on the r5 arc. (It was solid or refused for running on tangentially.)
- every-block-cylinder-tangent-concave-arc-r2-no-tangent-chain: the chain takes every piece of
  the half circle and ends at the convex edges it runs on into; it builds the analytic 19091.9620,
  so the outcome is solid (the refusal alternative was the chain-off one's).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Known failing: two 851 chains whose corner piece is a ball of tube-centre radius 0

851-fillet-arc-then-top-front and 851-fillet-top-front-then-upright: with the tangent chain on,
each R2 chain runs along the end curve of an R2 round on a curved face (the arc round's end on the
front face; the first round's end on the wall), where the rolling ball sits on that round's own
centre line, so the piece is the ball about (39.2538, -19, 8): the three-edge rolling-ball corner.
SMLib fails with status 1001; the same piece on a plane end arc builds (front-left upright).
Expectations unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Ladder: 016 and 018 refused, their r3 chain takes the vertical edge; 020 and 021 recorded

With the tangent chain on, 016's and 018's r3 runs on along the r5 round's end arc and down the
vertical edge, so the third pick (60, 40, 8) is 3 from both of that round's contact lines and
names no edge: refused, asking for a point along one edge. The corners they were made for are
020 and 021 (the r3 alone), whose golden entries are 016's and 018's (021 measured on f12a101).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel tests: three rounds at a box corner end on one rational patch on the ball's sphere, not an analytic sphere

SMLib's Ball corner is always the rational patch exactly on the sphere (owner ruling, blend
design §23), so the corner counts as an 'other' face beside the three cylinders and six planes.
The removed volume still checks the corner against r³ − πr³/6.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel rig: the curvature round on a pipe on a plate is listed: a seam grades G3 where exactly G2 is wanted

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel tests: the box-and-fillet wire fixture records the round's cylinder frame as SMLib now keeps it

The round's cylinder keeps the frame it was made on when it is turned (SMLib's SwapUV no longer
re-derives it): its origin sits at x = −20 on the same axis line, (x, 2, 8) along x, radius 2.
Same cylinder; one number in the record moves.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Known failing: two kernel tests' reasons as they fail on SMLib 5e8c9c1 (a partial selection running into an unblended tangent edge)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel rig: two curvature rounds get 600 s, past the 180 s default, while G2 certification is slow (tangency-labs/smlib#96)

On the CI runners the boss root and the hole rim, both size-2/size-1 curvature rounds, build
correctly but take more than 180 s: the order-2 section's span certificate and the round's
definition are evaluated station by station. The per-case limit is the one the thread cases use.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Lid ring sketch test: find the ring under the pick, not by main's kernel's name

On SMLib the lip join's exact history names the lid underside's ring a piece
of the lid's own start (F14:start#[...]), as the naming rule says for a face
that came from the feature's input; main's kernel named it after the case's
top it rests on (F2:top). The test looked the ring up by that name and so
stopped before the attachment it guards. It now takes the underside piece
under the pick point, which the cap certificate attaches on either kernel.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEu4Lr3moCCisdZShKz

* Viewport culling test: each box's triangles from its mesh, not 12

SMLib meshes a 10 mm box in 108 triangles, so the hard-coded 12 a box stopped
the test before it reached the culling. The expected counts are now each
body's span in the scene.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEu4Lr3moCCisdZShKz

* Timeline chip test: a fillet that really reduces

The R20 fillet with fit reduce on one top edge of a 10 mm box builds on SMLib
without a reduced-radius warning, so the fixture had no warned feature. It
now rounds the four vertical edges at R6, reduced to the full round (5), as
the copy-warning test beside it does, and checks the warning says so.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEu4Lr3moCCisdZShKz

* Oracle: 34-chamfer-then-fillet and 35-fillet-curved-curved pass on SMLib 5e8c9c1 (their STEP edges are on their surfaces)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Known-failing: the shell of a fully rounded box opened at one face fails on SMLib 5e8c9c1 (tangency-labs/smlib#102)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Known-failing: the action matrix's oblique cylinder cut fillet passes on SMLib 5e8c9c1; two shells still fail

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel rig: the boss root's curvature round gets the rig's 1800 s cap: on the CI runners, 8 cases at a time, its G2 certification passed 600 s (tangency-labs/smlib#96)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel rig: a case's continuity can ask for an order per seam (seams: [{order, edges, exact}])

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel rig: the pipe on a plate's curvature round meets the plate G2 and the pipe G3, 40307.8856312 (option B's order-2 jets, tangency-labs/solidcraft#10845)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Edge names: a seam names its one face once; the curve-id test reads every edge kernel-neutrally

A seam's two sides are the same face, and edge names list each face an edge
bounds once, so SMLib's extruded circle names its side's seam F2:side:c1, with
no separator. The name grammar now says one face also means a seam. The #5858
test no longer counts one '|' an edge: every part of every edge name is a
whole face name of the edge, one per face it bounds, two for the rims and one
only for the side's seam (checked to run bottom to top on the circle).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEu4Lr3moCCisdZShKz

* rustfmt on the merged tree

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEu4Lr3moCCisdZShKz

* Known failing on SMLib: two of main's new tests assume truck's box mesh and fillet torus

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEu4Lr3moCCisdZShKz

* SMLib: pin f916a8d (a round on a definition face is its definition; the G1 setback fill certified convex, setbacks too short refused naming the least; the G2 certificate halves a span no level can certify at once)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Fillet: a setback too short for its corner is refused naming the least that builds (smlib_blend_fit_size)

SMLib's setback corner takes a G1 fill only once certified convex, and refuses a setback too
short for that as SMLIB_ERR_NO_ROOM with SMLIB_FIT_SETBACK_AT_LEAST, the smallest setback that
builds. The kernel reads it (solidcraft-smlib binds smlib_blend_fit_size) and names it; the
Fillet dialog says it in plain words. Other room refusals keep their words. The r5 box corner
set back 5.5 is refused naming 5.962, which builds.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel tests: three rounds of different radii at a corner meet in SMLib's setback corner

Each round runs whole to its section square to its edge at its setback (the owner's rule, at
least 1.5 r, scaled to the least at which the G1 fill is certified convex), and the hole is the
setback split: one patch per round and the triangle in three quadrilaterals, 15 faces, G1.
Outside the box the sections span the body is the three strips', so its volume is theirs within
that box's. A later r3 round at the mitred corner makes the body all three at once make
(7928.142563938, 15 faces), as the engine and the model-cache tests have it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel tests: SMLib's own corner between curvature rounds is curvature continuous, as the G2 patch is

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel rig: the boss root's and the hole rim's curvature rounds take the default timeout again (37 s and 34 s on the box under the rig's parallel load, tangency-labs/smlib#96)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Agents: model decisions use exact geometry; a mesh is for display, export, previews and mesh bodies

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* xtask: mesh-decisions, model decisions use exact geometry; today's 176 mesh decision sites are listed and the list only shrinks

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Sheet metal: cuts and holes pass over sheets and bends by their exact boxes, never a mesh's

A primitive cut skips the sheets kernel::apart (exact bounds grown by the record tolerance)
finds clear of its tool; the bend check leaves the clear case to kernel::boolean, which answers
it from the same exact boxes with no intersection work. The mesh-box helpers are gone, and their
four lines leave xtask/mesh-decisions.txt (172 sites left).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Agents: AGENTS.md loses a stray merge-conflict marker line at its end

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Locales: Vietnamese for the setback-too-short refusal, now that vi is a complete catalog

Main made vi a complete catalog; develop's new message (ui.fillet_error_setback_short) had no vi
entry, so every_complete_catalog_carries_every_english_message failed on the merged tree.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel rig: exact volumes for 1213's bent-tube round and 4782's bar round, rel 1e-6

1213: the loft is a cubic skin through the five sections (ellipses between
them), not a torus; its volume 131601.597487935 by quadrature less the exact
rolling-ball removal 233.662461919 gives 131367.935026 (was a Pappus estimate
on a torus, 131360.148, at rel 5e-4).

4782: the section's area in closed form, 25 pi/3 - 5 sqrt(75) + 25, so
V = 630.2934873 (was a grid estimate, 630.293125).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEu4Lr3moCCisdZShKz

* Kernel rig: the boss root's and the hole rim's curvature rounds take 600 s again: under CI's parallel load they pass the 180 s default (develop 820b7b6)

They take 37 s and 34 s alone on the box, but CI runs 8 cases at once on slower runners. Their
G2 certificate cost is the open item (tangency-labs/smlib#106, moved to solidcore#2344).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* After the merge: clippy's auto-deref, and the mesh-decisions list for the merged tree

Develop's list was taken on develop alone. Main's code (tap-marking reach, cut and hole box
checks, join pieces, print faces, placed measuring, selection items, imported mesh payloads) and
the branch's own exact_edges and named_vertex use meshes too; three listed functions were renamed
by main (face_outline and plate_recesses into FaceSlice, paste into build).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEu4Lr3moCCisdZShKz

* After the merge: main's new tests on the SMLib kernel

- Cut To the far side of a curved wall (#8634): the tool stops past the first crossing of the
  face ahead of the profile; SMLib's whole bore is one face that a line crosses twice, so the
  single-crossing rule found none and the cut took nothing.
- Clearing view analyses goes through refresh_display, as main's #10687 made the others, so an
  open feature editor keeps OK and Cancel after Zebra.
- vi, a complete catalog on main, gets the setback-too-short message.
- fillets/KNOWN_FAILING: 4782's cause is now its continuity check (SMLib f916a8d builds the
  expected volume).
- Known failing, as main's tests assume truck: the spline wall's spans as one mesh patch, a
  hole touching an edge at a point leaving one body, and a copy of a copy's triangles matching
  a fresh mesh.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEu4Lr3moCCisdZShKz

* Twisted sweep: circles about a straight path are exact cylinders (solidcore#2184)

A twist about a straight path maps a profile circle centred on the path's
line onto itself, so its swept wall is exactly the cylinder. The skin
through the stations stayed within 3.9e-9 of it (6.6e-9 on the outer wall),
and a boolean with another body on that cylinder failed with status 1001
(solidcore#2184; the kernel's handling of a B-spline face lying on a
cylinder stays open there).

sweep_controlled now builds those walls exactly when the path is lines
along the profile's normal and there is no end scale or rails: a profile
whose every loop is such a circle is swept untwisted, and such holes in
another outer loop are cut as exact cylinders from the skinned sweep of
the rest. Other loops, and an axis-centred arc inside a loop of other
curves, are skinned as before; a failed cut falls back to skinning all.

Regression test in smlib_sweeps_tests.rs: #2184's tubes joined at 10, 16,
18, 20 and 30 degrees give one closed body of 11.5625 pi with cylinder
walls; a twisted hexagon with the bore keeps a cylinder bore.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdcgEtwSrdwZxqTQp5MfnU

* After the merge: a hole that misses is found by exact boxes; a face is a one-face patch; main's mesh-box helpers gone

- `hole_misses` (main's #10701, through brett's branch) decided from the targets' mesh boxes. It now uses their exact boxes (`Body::exact_bounds`). An unknown box still leaves the hole not known to miss.
- With #10303's two shortcuts and `hole_misses` off the mesh, `body_box`, `mesh_box` and `padded_box` have no caller left, and are deleted.
- `project`'s intersection link passes its one face to `meeting_edges` as a one-element patch. This branch picks one face, not #9009's mesh patch.
- xtask/mesh-decisions.txt drops the six sites that no longer use a mesh here: `apply_traced_with`, `cut_end_past_face`, `beyond_cylinder_reach`, `hole_misses`, `body_box`/`mesh_box`/`padded_box`.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* After the merge: split_move's Result reaches its callers; to_local's duplicated move check goes

`frames::split_move` returns a Result on this branch: an unplaceable model is an error, not a silent no-split.
- Move/Copy of a component (main's #9914) propagates it, as its other calls do.
- `to_local` had the solid.move check twice after the merge (brett's Option one and ours). Ours stays, after the no-picks/not-placed return.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* After the merge: main's two new tests take world_state's Result, as the rest of their files do

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Native B-rep: a ball patch is its own surface kind, read from and written to SMLib's flat kind 6

SMLib's flat arrays now carry the corner of three equal rounds as `SURFACE_BALL_PATCH` (kind 6): its net as a NURBS face's block, with its sphere's centre and radius as its definition (ids 36 and 166). The record reads it as `ball_patch {center, radius}`, a label like the analytic kinds, with the net as its `nurbs`. A plain sphere with a form won't do: the net is not a piece of SMLib's sphere NURBS, so SMLib's `IsNurbSurfaceSphere` would refuse it.

Written back, a ball patch is kind 6 with the same definition, which SMLib takes once its net is certified on that sphere. Elsewhere:
- Offsets classify it as its sphere, so its offsets stay exact and collapse at the centre.
- Face kinds count it as a sphere: fillet_three_edges_at_a_corner's corner is now `sphere`, no longer `other`.
- STEP, which has no ball-patch entity, gets the net as the face's B-spline surface, pcurves included.
- Iso curves and SSI of the label are refused; the face's form is what evaluates.

Test: a_rounded_cubes_ball_patches_go_round_with_their_spheres. The cube's eight corners are ball patches about the corners pulled in by r. Points of each patch lie on its sphere to 1e-12. The patches are kind 6 again after import, with the same definitions bit for bit, and the closed-form volume holds.

Format: pending/ball-patch-faces.rs. Docs: docs/native-brep.md and the rig README's `surface_types` (`ball_patch`).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* SMLib: pin v1.1.0-alpha (a18bbe5): the generic blend's step-4 corners, the Ball's sphere patch through the flat arrays, all-target CI

Since f916a8d:
- #120: every target again, plus the warnings fix;
- #121: the Ball's patch as flat kind 6 with its sphere, read by the previous commit;
- #122: tests gated on counts;
- #123: the Mixed corner;
- #124: the End at a smooth vertex;
- #125: four or more rounds, and open setbacks;
- #126: the equal-area thumbweight guard;
- #127: the trunk merge.

The SMLIB_PREBUILT checksums are from the release prebuilt-a18bbe5…, written by `scripts/fetch-smlib.sh --pin`.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* smlib export tests: rustfmt's line for the rounded cube's edge midpoints

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel rig: the fillet and ladder results on SMLib v1.1.0-alpha, the new failures recorded with their reasons

Fixed (off the known-failing list): the Mixed corner's peg cases and the concave arc with the front top edge (smlib #123), and the 40° boss junction (smlib #124). Eight cases in all.

New failures, recorded as they are, each with its reason. They passed on f916a8d.
- **Six cases grade a blend seam G0 where G1 (or G2) is wanted,** at the end of the round's contact seam, after smlib #124's End at a smooth vertex. Two of them also shift in volume.
- **Four cases come back unsupported** after smlib #125, which leaves setbacks to the Setback's corners only.
- **Ladder 07/012** makes 9 faces, 20 edges and 13 vertices, where its golden had 8, 18 and 12. Its golden entry is the new result, with the reason added.

Whether each change is by design or a defect is being established. A defect is fixed in solidcore, where the End and the Setback are being brought over (#2364, #2365), not in the frozen SMLib.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel ladder: 07/021 and 06/018 recorded as they are on SMLib v1.1.0-alpha, each with its reason

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel ladder: a golden case marked unstable is gated on its status, and 06/018 is marked

On SMLib v1.1.0-alpha, 06/018 makes 597 or 594 edges from the same input:
the union's post-process keeps pointers to edges it has deleted. A golden
entry's `unstable` reason makes the gate compare that case's status only
and name it in its output; recording keeps the reason, and only an edit of
the golden removes it. v1.1.0-alpha.1 fixes the cause and removes the mark.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* After the merge with main: its new code on develop's exact bounds, types and thread fields

- The sheet cut's cull (#10301) and the tapped wall's (#10227) use each
  body's exact box (`exact_box`), not its display mesh.
- model.threads replies with develop's thread fields: form, hand (right,
  left or both) and starts.
- surface.boundary's schema checks a network curve's continuity as the
  handler does, and words a list entry's error as it does.
- Main's new tests take develop's Hole fields and world_state's Result.
- mesh-decisions: contact_sweep's triangle check is now `depth` (main
  split `hits`).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Known failing: the 14 cargo tests SMLib v1.1.0-alpha fails, with what each sees; two cases that pass now

Develop's CI after the a18bbe5 pin (run 41522001173) fails 14 tests that
passed on f916a8d, and passes two listed ones. Each new line says what the
test sees on v1.1.0-alpha; v1.1.0-alpha.1 fixes the setback hole's cause.
shell_tangent_chain_that_opens_every_face_explains_how_to_keep_walls and
oracle 37-pyramid-apex-blend pass and leave their lists.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Cargo.lock: develop's profile, smlib and step crates at main's 0.30.1

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel rig: main's six new sketch-only commands are exempt from a kernel case

sketch.construction_line (and its horizontal and vertical forms) and
sketch.spline.insert_point, remove_point and reverse change only the 2D
sketch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Cuts leave the bodies their tools can't reach, by exact boxes (#10225 on exact geometry)

- The per-target cut loop skips a body whose exact box lies clear of
  every tool's, as main's version did with mesh boxes. The sheet culls
  (#10226, #10301) and this one now leave the far sheets and bodies alone:
  a_cut_leaves_bodies_it_cannot_reach_as_they_were passes and leaves the
  known-failing list.
- a_cut_still_runs_where_its_tool_touches_grazes_or_meets_a_curved_face:
  a tool 0.5 µm clear of the block is now left out (exact boxes; 0.5 µm is
  far past the bodies' tolerances); touching and 0.1 mm in still cut.
- Known failing, with reasons: main's edge lookup by a stored point
  (#11286, on the display mesh) is not taken; SMLib refuses the
  wrong-way chamfer for its tangent chain first.
- EXEMPT: main's eight new sketch-only commands.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel rig: main's new view, rendering, appearance and settings commands are exempt from a case

62 of the commands main added change no geometry: the view and render
panes, rendering lights and results, appearance tools, settings profiles,
point clouds (drawn as points, never a body), the selection, two sketch
edits, component links and History bar visibility, edit.repeat,
timeline.play, inspect.compare_picture (a picture against the outline),
and component.paste_from_design, which pastes the app's cross-window
clipboard that no command fills. Each joins EXEMPT's first section under
its family's reason.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel rig: construct.plane.axis_point gets construction cases

The plane through a box edge and its far corner, graded against its
closed-form normal (0,-2,3)/sqrt(13) through the origin, and the refusal
of a point on the axis.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel rig: assembly cases for mass properties, align, find limits and moving items between parts

- assembly.mass_properties: two boxes (2 g/cm3) in a rigid group: mass
  0.006 kg, centre (18.333,5,5), inertia diag(0.1, 1.0333, 1.0333) kg mm2
  by the parallel-axis theorem.
- occurrence.align: a block's end face onto a base's top face: a quarter
  turn about y, centre (70,20,20), box (65,15,10)-(75,25,30).
- joint.find_limits: a slider between two stops: min -20, max 10 mm.
- component.move_features, browser.reparent and
  component.move_sketches.to_parts: items moved into a turned and moved
  part keep their world place (volumes, centres and boxes by hand).

The mass case assigns a material it creates, so material.create and
material.assign leave EXEMPT. PATHS names the new command modules.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

* Kernel rig: export and print face cases in the import area

- file.export_step, export_stl, export_3mf, export_obj: a 30x20x10 box
  written and read back (the meshes converted to a solid): V 6000,
  A 2200, 6 faces, box (5,-10,0)-(35,10,10).
- file.export_urdf: a hinged arm on a grounded base: 3 links, 2 joints,
  and the base's mesh, written in metres, reads back as the
  1 x 0.5 x 0.2 box (V 0.1, A 1.6).
- body.print_face: a cylinder's flat end is accepted, its side refused.

PATHS adds the URDF writer and the print face modules.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DtL5M41G8Nua7HvuFPZkmJ

---------

Co-authored-by: jyviko <2501648+jyviko@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: brett19 <brett19@gmail.com>
2026-11-02 22:53:32 +00:00

66 KiB
Raw Blame History

Where the time goes: the profiler

Every app is built with the profiler (crates/profile, the apps' default profile feature). It records nothing until switched on: SOLIDCRAFT_PROFILE=1, the CLI's --profile, the Profile operations preference (Graphics page) or diagnostics.profile {"enabled": true}. Off, a span is one relaxed atomic load.

What it records, as nested spans on each thread:

Category Spans
smlib every SMLib C API call, timed by the C API itself (steady_clock)
kernel booleans, blends, sweeps, lofts, shell/draft/offset, healing, sewing, conversions, meshing, measuring
import the stages of a STEP or IGES import: parse, index, each solid, placement, mesh check, convert batches
feature each feature computed, by its type; doc evaluate around them
command each Session::execute, by command id
io import, export, design read and write

Self time is a span's time less the spans inside it; its total counts only its outermost run on a thread, so a command run by a command is not counted twice. Work handed to other threads shows under their own spans. The report is a table (the most self time first, with self time per category) and the individual spans as a Chrome trace-event file for chrome://tracing or ui.perfetto.dev (at most 200,000 spans; later ones still count in the table).

solidcraft-cli run part.step --profile --profile-out trace.json   # table on stderr
solidcraft-cli run part.solidcraft --profile --quiet
solidcraft-cli bench-open part.step --profile

Over the control channel or MCP: diagnostics.profile {"reset": true, "enabled": true}, do the work, then diagnostics.profile {"top": 30, "trace": "trace.json"}. A STEP or IGES import's reply carries its own profile while recording.

Design loading

Edges made by booleans: each point evaluated once (2026-10-16)

A boolean meshes both operands with truck-meshalgo, which first divides every edge into a polyline (bounded_polyline, vendored). An edge that an earlier boolean made is an intersection curve. Each of its points is found by a grid search on each of the two surfaces, 51 × 51 evaluations of a NURBS surface each, before Newton. The division evaluated both ends of every piece again, so 2 of its 4 evaluations per piece repeated a point it already had. Profiles of main (01b1b705) on the machine below put the division at 42 % of the CPU time of opening the rover, 58 % of bench-agent and 82-85 % of a drilled hub or a union of tubes.

Each piece now carries the points of its ends, and its midpoint is both the probe and an end of both halves: 2 evaluations per piece. The fallbacks for samples that do not evaluate (#1364) are kept: a midpoint that fails is NaN as a probe and the curve's approximation as an end, exactly as before. The polyline is the same, bit for bit: edge_polyline_tests compares it with the previous division (points, order and number of evaluations) on the intersection curves of a drilled cylinder and on a circle.

Release solidcraft-cli from main (73493b94) and the same with the change, 3 alternating pairs (before/after, after/before, before/after), each run a new process timed with /usr/bin/time -v, started once the 1-minute load average was below 2.5. Medians, with the range of the 3 runs. 2018 laptop: Intel Core i7-8565U (4 cores / 8 threads, 15 W), 7.4 GB, Ubuntu (Linux 7.0), on mains power:

Run Evaluation Meshes Wall time CPU time Peak RSS
bench-open, examples/acceptance/sensor_enclosure.json 3,203 → 3,089 ms 6 → 6 ms 3.2 → 3.1 s 8.7 → 8.5 s 91 → 91 MB
bench-open, a hub barrel with one radial hole cut through it 966 → 817 ms 878 → 481 ms 1.9 → 1.4 s (-24 %) 4.4 → 2.8 s (-37 %) 27 → 30 MB
bench-open, the frame of the bench-agent bicycle (one union of tilted tubes) 8,204 → 6,277 ms 1,711 → 893 ms 9.9 → 7.2 s (-28 %) 33.3 → 20.6 s (-38 %) 44 → 46 MB
bench-open, a rover (3,324 features, 1,395 bodies) 26,783 → 22,792 ms 7,960 → 6,046 ms 35.7 → 29.8 s (-17 %) 129.0 → 101.9 s (-21 %) 2,076 → 2,079 MB
bench-agent (build 33.5 → 30.2 s, -10 %) 10,553 → 8,565 ms 2,778 → 2,081 ms 47.6 → 41.1 s (-14 %) 124.0 → 91.0 s (-27 %) 1,428 → 1,432 MB

bench-agent's evaluation and meshes are those of reopening the saved design. CPU time is the steadier measure on this 15 W processor: its before and after ranges are apart for the four cases with boolean edges (the hub 3.0-4.4 s before, 2.5-2.9 s after; the rover 113.6-133.5 s, 101.0-102.5 s), while the first pair's "before" run, which started on a cool processor, overlaps the "after" wall times of the two short cases. The sensor enclosure does not change: its time goes to grid searches in face meshing, which this does not touch. An earlier series on main (8535d229), on a quieter machine, gave -38 %, -35 %, -18 % and -16 % of wall time for the same four cases. The CPU time saved is close to half the division's share in the profile. The mesh phase gains as well, so these display meshes also call the reference mesher somewhere (where is not traced here; tess::edges, the display plans' own edge division, already evaluated each point once). In the rover, the second shell (which names every body's faces, so meshes them all) goes from 7.8 s to 5.7 s.

No model changed. solidcraft-cli eval of the five designs (the bicycle as bench-agent saved it) is the same with both binaries apart from the per-feature milliseconds: bodies, volumes, areas, centres, bounding boxes, topology counts and warnings. bench-open meshes the same triangle counts, and the bicycles both binaries save differ only in the document's name.

STEP assemblies: a repeated part is meshed once (2026-10-14)

The reader already converts each B-rep item once and keeps it for the instances that list it again. Every instance still became a body of its own geometry (it has to: features cut and join it where it is), and each of those bodies made its own display mesh: tessellation, seam and tangent flags, the most expensive step of opening a large assembly. A board with 60 of each resistor meshed the same resistor 60 times.

A later instance of an item now shows its first instance's display mesh, placed by the difference of the two placements: the same triangles, faces and edge flags, moved. The mesh is made once, by whichever instance is drawn first. Only a rotation and translation is shared this way (anything else meshes itself), and anything that changes what a body shows (split lines; any operation, which makes a new body) drops the link. Measurements, booleans and export read the instance's own geometry as before, so solidcraft-cli eval of an assembly is unchanged.

Two generated assemblies (SolidCraft designs with placed copies of components, exported as STEP AP214 assemblies), solidcraft-cli bench-open, release build, shared 16-core Linux box:

file bodies mesh phase before after CPU before after peak RSS before after
4 parts, 500 instances (bolts with a torus ring, brackets) 901 2,215 ms 139 ms 28.4 s 4.1 s 918 MB 749 MB
10 filleted parts with 8 pins, 600 instances 5,401 1,028 ms 85 ms 16.7 s 6.0 s 1,095 MB 783 MB

Both import to the same triangle counts (4,404,412 and 4,316,412). Each instance's display plan (the GPU viewport draws bodies from plans) is still made per instance; sharing those the same way is the next step.

Designs built by agents: work per command that grew with the design (2026-10-14)

A design an agent builds over MCP is made of many small commands: a cylinder per spoke, a box per chain plate, one part built at the origin and its component moved into place. Each command did work over the whole design, so the 1,000th command cost far more than the first and a 1,300-body design took minutes to build and to show. solidcraft-cli bench-agent builds such a design (a generated bicycle: about 1,260 commands, 1,280 bodies, 1,250 features, wheels built at the origin and moved), timing each command and then what a window needs to draw its result (the world state and its display meshes), and finally opens the saved design like bench-open. The script is generated, so it needs no fixture; --scale changes the counts and --script-out writes the commands for replaying elsewhere (engine.execute one by one over the control channel reproduces the desktop case).

What grew with the design, per command:

  • Moved components were placed again after every change. world_state transformed every body of a component whose occurrence is moved and gave it a new, empty display mesh, so every command meshed all those bodies again (the two wheels: 150 bodies and two tori, 130-190 ms per command). It now keeps the bodies it placed and reuses each one whose definition (same body, same display cell) and placement are unchanged.
  • A free dN name searched the design once per number (next_d_name), quadratic in the number of named inputs, for every new feature input.
  • Every expression evaluated every parameter (Document::eval), several times per feature command (each coordinate, each size). Parameters are now evaluated only when the expression names one.
  • The parameter table's used_by searched all parameters for each one (quadratic); it is one pass now. The desktop also stops gathering expression completions (eleven parameter tables) on every change while no text box has focus.

solidcraft-cli bench-agent, release build, 16-core Linux shared with other builds (load average 15-20), one run each:

per command, by design size before: command before: display after: command after: display
200 commands (227 bodies) 6.0 ms 129 ms 1.8 ms 2.7 ms
600 commands (621 bodies) 19.9 ms 169 ms 4.9 ms 2.4 ms
1,000 commands (1,021 bodies) 45.1 ms 187 ms 9.8 ms 4.5 ms
1,200 commands (1,221 bodies) 56.8 ms 194 ms 12.5 ms 5.6 ms
whole build (1,260 commands) 241 s 16.3 s

The command column still grows (the timeline's fingerprints and parameter values are walked once per command), but slowly. solidcraft-cli eval output for the design corpus, the examples, the regression scripts, the generated bicycle and a 3,300-feature agent-built design is unchanged apart from the timings.

The viewport rebuilt its whole model scene after every change and uploaded all of it again (every body's triangles and edges, 60 MB for the bicycle where bodies are built on the CPU). It now builds the scene in blocks of a few bodies, cut where a body's own key says so that adding or removing a body changes only its block, and keeps each block's bytes and GPU buffers while the block comes back unchanged. The bodies' boxes (fit, picking, the ground shadow) are kept per mesh the same way. Built command by command over the control channel (engine.execute, then ui.inspect) in the desktop app under Xvfb with software GL:

desktop, bicycle built over the control channel main with #1148 and this
UI work per change at ~1,000 bodies (frame_ms) 1,374 ms 18 ms
model scene build at ~1,000 bodies (build_ms) 41 ms 1.4 ms
whole build 1,409 s 210 s

Most of what is left there is software rendering of 770k triangles per frame.

STEP import: shells and item solids without copies (2026-10-13)

A file that lists one product definition many times (a bolt used 70 times, a panel in several assemblies) is imported item by item, and every item's B-rep was deep copied twice before it was handed on: once by the caller of Solid::try_new, which clones the shells so that the failure path still has them, and once by the reader's own cache, which kept the solid for a later item that might list the same definition.

read_item now runs those four checks (non-empty, connected, closed, no singular vertices) on borrowed shells, so a shell that passes is moved into the solid instead of cloned. The reader keeps each prepared solid in an Arc, and an item counts as once when one representation lists it and exactly one product-definition walk reaches it: that item's solid is moved out of the cache with Arc::try_unwrap instead of copied, because no later item can ask for it. The walk follows MAPPED_ITEM targets too, so an item reached through a mapping in another product's node is still converted once - a second conversion would repeat that item's warnings. A file whose walk exceeds the work budget (MAX_CLOSURE_WORK, 65 536 nodes) falls back wholesale to the copying path, as does any file with component occurrences, which walks a definition again per instance.

Nothing observable changed. Fifteen fixtures (primitives, a void, a fillet, a revolve, colours, centimetre units, 70 duplicates of one definition, a broken file with mappings, a two-instance assembly and an expanded one; 99 bodies) give the same import dump - name, path, closed flag, file faces, face origin, topology and merged counts, bounding box, volume, area, centroid, paint, warnings and tree - apart from the per-feature milliseconds, and all fifteen STEP in/out round trips are byte-identical.

Two generated flat STEP files, timed like the table above (same binary pair, no other work running); four alternating runs each, two for the 31 MB file:

File Baseline peak RSS Patched peak RSS Import
2500 boxes, 9.5 MB 150.1 MB 145.9 MB 1.81 s → 1.74 s
2500 filleted boxes, 31 MB 428 MB 433 MB 70.3 s → 71.0 s

The first row's ranges do not overlap (150.0-150.6 MB against 145.0-146.0 MB), and the saving is the shell copy that is no longer made for every item. The second file's runs land between 422 MB and 444 MB either way, so two runs each do not resolve it: there the conversion of the geometry dominates the per-item copy. An item that keeps every conversion - an assembly, or a file over the work budget - is unchanged.

Display tessellation plans (#303, 2026-10-12)

Body::display_mesh meshes from a TessPlan (crates/kernel/src/tess): every edge is sampled once and both faces use those nodes exactly (watertight by construction); planar faces are a constrained Delaunay triangulation of their edge nodes alone; faces bounded by their parameter box are structured grids zipped to the edge nodes; other four-cornered faces are Coons grids; only genuinely trimmed curved faces (holes, more than four corners) use spade's constrained Delaunay with sag and Ruppert refinement. A body the planner cannot handle is meshed by truck-meshalgo as before, which also still meshes for modelling (booleans, measure, provenance). Evaluating a plan (surface points and normals) runs on the CPU (TessPlan::evaluate) or in a compute shader (solidcraft-gpu).

Release build, Apple M4 Max, medians of five. 400-hole plate (200 × 200 × 5 mm, 806 faces), cargo test --release -p solidcraft-gpu bench_plate -- --ignored --nocapture:

Tolerance Reference mesher Plan CPU evaluation GPU evaluation Triangles (reference / plan)
0.3 mm 23.3 ms 11.9 ms 1.1 ms 1.1 ms 20,812 / 20,812
0.03 mm 54.2 ms 31.6 ms 2.5 ms 1.9 ms 76,812 / 52,812
0.006 mm 118.9 ms 65.1 ms 5.3 ms 3.1 ms 302,412 / 104,012

Planning (edge sampling, parameter projection, triangulation) dominates; evaluation is a few milliseconds either way, so GPU evaluation saves little until planning is cheaper. The GPU timing includes upload and read-back.

Parallel STEP body preparation (2026-10-12)

Profiling an original 48-cylinder STEP fixture before changing the importer measured 3.17 ms parsing, 6.06 ms reconstructing B-reps and 45.58 ms preparing/orienting bodies. The five complete imports had a median of 59.21 ms. The expensive orientation mesh, rather than the text parser, was the first parallelization target for #287.

Desktop import now prepares consecutive independent solids in a representation with up to eight workers. Each batch contains at most 64 solids. Parsing, B-rep reconstruction, assembly traversal, first-use unit resolution, naming and result collection remain ordered. Mapped items end a batch. Fewer than four solids use the serial path; wasm always uses it. Thread creation failure falls back to serial work, and kernel/worker failures remain per-body errors reported by the importer.

Windows x64, AMD Ryzen AI Max+ 395 (16 cores / 32 logical processors), optimized ci profile, no concurrent build: three fresh test processes per fixture, five alternating serial/parallel pairs in each process. These are complete uncached kernel imports, including parsing and body preparation; fixture generation/export and later desktop display meshing are outside the timer. Both paths use the same input text and code.

Original generated fixture Bodies Serial median 8 workers median Reduction
Cylinders, radius 5 mm, height 12 mm 48 57.54 ms 17.38 ms 69.8%
Cylinders, radius 5 mm, height 12 mm 512 896.31 ms 313.06 ms 65.1%
Tori, major radius 20 mm, minor radius 4 mm 48 269.05 ms 51.90 ms 80.7%
Boxes, 10 × 15 × 12 mm 512 247.97 ms 90.15 ms 63.6%
Boxes, 10 × 15 × 12 mm 4 1.49 ms 1.02 ms 31.5%
One cylinder (serial fallback) 1 1.22 ms 1.29 ms serial fallback

There are 15 samples per cell. The single-cylinder modes run identical serial code; their 0.07 ms median difference is within the overlapping 1.11–1.57 / 1.13–1.52 ms ranges. The 512-cylinder runs ranged from 567.40–983.61 ms serial and 197.72–345.51 ms parallel; scheduling/thermal variation is material. These synthetic results do not establish a speedup for every assembly or complete desktop Open. In particular, parts containing only one solid are not parallelized across assembly nodes. Cross-part scheduling, parallel B-rep reconstruction and later display meshes remain open.

Reproduce after compiling, without other builds running:

cargo test -p solidcraft-kernel --profile ci benchmark_independent_step_solids --no-run
SOLIDCRAFT_STEP_BENCH_COUNT=512 SOLIDCRAFT_STEP_BENCH_KIND=cylinder \
  cargo test -p solidcraft-kernel --profile ci benchmark_independent_step_solids \
  -- --ignored --nocapture --test-threads=1

Use torus or box for the other fixtures and repeat in three fresh processes. SOLIDCRAFT_STEP_BENCH_FIXTURE optionally writes the original STEP for native inspection. No timing threshold runs in CI. Regressions compare serial and threaded imports with different worker counts: exact display meshes, measures, body/face source identities, names, colours/opacity, units, rotated assembly placements, duplicates and partial failures across the batch boundary. The public analytic/FreeCAD reference suite also checks STEP export/import; its current results are reported in docs/oracle.md.

Design storage: container and binary payloads (#303, phase one)

Design containers avoid formatting/parsing imported mesh coordinates as JSON numbers. The metadata serializer replaces embedded fields before creating JSON values, without cloning the full document or duplicating its schema. Binary coordinates retain f64 precision, indices retain u32, and SHA-256 deduplicates identical payloads. The reader checks archive budgets, CRCs, hashes, exact lengths, finite coordinates and index bounds.

Reproduce the storage measurements with an optimized ci IO test executable:

cargo test -p solidcraft-io --profile ci benchmark_ -- --ignored --nocapture --test-threads=1

The benchmarks generate an original 100,000-vertex/100,000-triangle mesh and a STEP containing 512 disjoint 10 mm boxes. Fixtures are written under CARGO_TARGET_DIR (or the crate's target directory). Each compares nine encoding/decoding samples; atomic-save samples include encoding, temporary-file flush, replacement and .bak creation and alternate JSON/container ordering. Documents are compared after decoding. These are storage measurements, separate from timeline evaluation or first-frame latency.

On Windows x64, based on upstream e6e0a7e before STEP parallelism (#312), three consecutive optimized runs without concurrent compilation gave the following medians of the three nine-sample medians (milliseconds, fast mode):

Fixture / operation JSON Container
Mesh size, bytes 15,139,115 3,600,828
Mesh encode 23.336 6.094
Mesh decode 57.175 2.962
Mesh atomic save 52.924 14.856
STEP size, bytes 1,998,793 1,950,056
STEP encode 1.507 1.473
STEP decode 1.479 1.318
STEP atomic save 7.213 7.087

Three alternating fresh-process solidcraft-cli bench-open runs of each STEP encoding returned the same 512 bodies and 6,144 triangles. Median read-plus-parse was 2.274 ms for JSON and 2.151 ms for the container; median evaluation was 218.559/236.915 ms and display meshing 203.160/197.790 ms respectively. Geometry dominates and these variable results do not establish an end-to-end STEP opening improvement. Fresh processes do not imply a cold operating-system file cache, and these are not desktop first-frame times.

Fast mode stores STEP and mesh payloads directly. Explicit "compression":"compact" compresses them at Deflate level one; it is a space/CPU tradeoff. In the initial 512-box STEP experiment, compression reduced 1,998,793 JSON bytes to 578,659 package bytes but increased median encoding from 1.68 to 5.02 ms and decoding from 1.76 to 4.18 ms. That regression is why compression is not the default. PNG/JPEG are stored directly in either policy; small parametric metadata uses Deflate level one.

Plain JSON remains a supported Save As/export format; .json saves keep JSON and scripts may explicitly select either format. The CLI detects packages before attempting JSON parsing. Desktop/browser recovery and browser downloads also use the container. See the versioned container format for layout and bounds.

This phase does not restore evaluated B-reps, skip STEP import, or make cold geometry rebuilds instant. A matching cache key is not enough to trust unvalidated geometry. The native B-rep increment described below supplies provisional display, not editable session restoration. It is separate from #287's import parallelism.

Opt-in mesh preview measurements (#303, phase one)

The phase-one desktop implementation can show an experimental saved picture while the file worker rebuilds geometry. It requires an explicit preview: true engine Save/Save As call. Ordinary saves omit display meshes to avoid the measured overhead; these measurements predate native B-rep caching. The cache contains only already warm evaluated display meshes, capped at 8 MiB, 1,024 bodies and 100,000 triangles. Larger or cold models save without it. Cached geometry is never used for editing, measurements or picking. Source/document fingerprints and payload validation reject stale or malformed caches; cancellation and document/revision checks prevent a background result from replacing newer work.

The cached picture is prepared on the file worker from the same file snapshot as the normal Open command. It adds work before reconstruction, rather than bypassing it, so measure preview readiness and full editing readiness separately. Cached display also adds file size and save work; the storage-only results above exclude that optional data.

For an engine-level save benchmark, first generate the original 512-box STEP fixture with the ignored IO benchmark, then run the warmed-session benchmark:

cargo test --workspace --profile ci benchmark_step_storage -- --ignored --nocapture --test-threads=1
cargo test --workspace --profile ci benchmark_warm_design_save_with_preview -- --ignored --nocapture --test-threads=1

Set CARGO_TARGET_DIR to the workspace's absolute target directory in both commands. The engine benchmark alternates nine JSON/container atomic saves, includes the optional display cache, and verifies both preview decoding and authoritative document identity. Three runs on Windows, with no concurrent build or test suite, measured median atomic JSON saves of 7.679/7.610/7.630 ms and container-with-preview saves of 10.319/10.633/10.541 ms. Preview decoding took 0.964/0.961/1.130 ms; the container was 2,510,490 bytes. This optional preview costs save time on this STEP fixture. An actual desktop sensor-enclosure opening on e8a8e0f prepared its saved picture in 24.320 ms; geometry reconstruction, display warmup and thumbnail preparation took 8,913.212 ms. The same preview experiment is now explicitly opt-in; full geometry readiness still requires reconstruction. For actual desktop opening phases, launch with SOLIDCRAFT_PROFILE_OPEN=1: ready_ms reports saved-picture preparation, while the normal phase log reports reconstruction, display warmup and thumbnail completion. Neither log measures the exact GPU first frame.

Native cache opening (#303)

The portable native B-rep cache is opt-in with cache: true on warm container saves, without duplicating display meshes. Ordinary saves and browser downloads default to cache: false: the measurements below do not justify imposing the additional save time and file size on every design. Desktop Open reads metadata and optional geometry through a seekable file, produces a bounded provisional picture, then resolves original sources and reconstructs the complete session before editing. This improves the opportunity for early display, not the cost of authoritative feature evaluation. An absent, invalid or incompatible cache uses the original reconstruction. Browser OPFS/downloads share the format; browser opening still rebuilds synchronously.

The reproducible engine benchmarks separate atomic saving, provisional mesh readiness, reconstruction and final display-mesh readiness:

cargo test -p solidcraft-io --profile ci benchmark_step_storage -- --ignored --nocapture --test-threads=1
cargo test -p solidcraft-engine --profile ci benchmark_native_cache_save -- --ignored --nocapture --test-threads=1
cargo test -p solidcraft-engine --profile ci benchmark_native_cache_open -- --ignored --nocapture --test-threads=1

Use an absolute CARGO_TARGET_DIR. The save benchmark defaults to the generated 512-box STEP document and accepts SOLIDCRAFT_BENCH_FILE for another public design. It alternates nine saves of each encoding, including flush/replacement/backup, and verifies document identity and native-cache presence. It never times initial fixture evaluation as save work. Open runs once per process; SOLIDCRAFT_BENCH_ENCODING selects 0=JSON, 1=container without cache, 2=native cache, 3=compact native cache. Changing only cache/key.json's application version provides a cache-miss control. Fresh processes do not imply a cold operating-system file cache. The engine callback time is provisional mesh readiness, not a rendered desktop frame; use the desktop profile log and actual screenshots to check the saved picture and completed model.

Measured 2026-10-13 on Windows 11, Ryzen AI MAX+ 395 (16 cores/32 threads), ci profile, with the neutral codec revision 75381fa, envelope SCBREP03, producer solidcraft-truck-adapter/2, and opening integration revision 3243422 on upstream 2880525 (0.6.0). These replace the earlier prototype measurements. They compare encodings on the same code and do not attribute upstream modelling improvements to the cache. Runs started after local builds and the complete CI suite finished; ordinary desktop activity remained. Save values are medians of nine alternating atomic saves. Open values are medians of five fresh processes per encoding, alternating forward/reverse order. Memory is each process's peak working set from Windows GetProcessMemoryInfo, sampled every 10 ms, then the median across those five processes; it excludes a desktop window/GPU.

The original STEP fixture is a 32 by 16 array of 10 mm cubes, 12 mm apart, exported by SolidCraft: 512 bodies and 6,144 final triangles. Its JSON embeds about 2 MB of STEP. Every encoding reopens without feature errors and with those same counts.

STEP encoding Bytes Atomic save, ms
JSON 1,998,826 12.616
Container, no cache 1,950,084 10.516
Native cache, fast 4,694,035 47.755
Native cache, compact 837,067 44.015
STEP encoding Provisional mesh, ms Reconstructed, ms Final meshes, ms Peak working set, MiB
JSON none 175.842 748.911 40.102
Container, no cache none 175.885 722.306 40.504
Native cache, fast 62.099 228.794 747.976 40.883
Native cache, compact 65.981 235.403 762.120 40.918
Native cache miss none 169.039 647.617 40.453

The public examples/acceptance/sensor_enclosure.json script, saved as a JSON design, has 21 features, two bodies and 4,364 final triangles on this revision.

Sensor enclosure encoding Bytes Atomic save, ms
JSON 52,281 5.243
Container, no cache 5,545 4.969
Native cache, fast 379,095 12.646
Native cache, compact 72,603 14.111
Sensor enclosure encoding Provisional mesh, ms Reconstructed, ms Final meshes, ms Peak working set, MiB
JSON none 3,979.063 3,984.593 70.438
Container, no cache none 4,038.127 4,042.006 71.789
Native cache, fast 241.541 4,269.124 4,274.272 70.465
Native cache, compact 236.739 4,271.254 4,275.159 70.148
Native cache miss none 4,136.517 4,140.903 69.941

Native display is available earlier, at the cost of encoding, file size and work before reconstruction. It does not make the full design editable earlier, and these results do not establish a memory reduction. Final-mesh readiness varied from 611 to 856 ms for STEP and 3.84 to 4.60 s for the enclosure; small differences in full-opening medians should not be read as modelling speedups. Compact storage reduces this STEP example's size and atomic-save time, but costs slightly more save time on the enclosure. The default cache: false retains the phase-one save path; compression: "compact" is an explicit space/CPU choice. Larger or unsupported geometry can exceed the documented budgets and reopen without a provisional image.

A separate desktop check used a fresh 0.6.0 application process for each enclosure encoding, with no local compilation or test suite running. The native cache's provisional picture was prepared at 429 ms and captured visibly during opening; the completed model was adopted in 6.48 s. JSON had no provisional picture and completed in 6.08 s. These are single visual checks, not repeated timing medians; the worker reported 6.10/5.66 s respectively, and external observations include UI scheduling and screenshot requests. During native opening, 94 control-channel inspections had a median round trip of 20.8 ms and a maximum of 106.5 ms (JSON: 95 inspections, 20.3/36.9 ms). The provisional picture is explicitly labelled as being checked before editing; the final frame shows the rebuilt Case and Lid and their editable feature timeline. This demonstrates early feedback, not faster completion of feature evaluation or a worst-case UI-latency guarantee.

Analytic extrusion source preparation

New-body extrusion certificates now prepare the sketch's bounded analytic segments once per certification pass instead of allocating segments and resolving their points for each generated cap boundary and side face. Saved designs run this same code during cold rebuilds. Exact containment tolerances, ambiguous-source rejection and the certification work budget are unchanged. Prepared segments borrow an immutable evaluated sketch and are not retained across edits.

The reproducible microbenchmark is:

cargo test -p solidcraft-doc --profile ci benchmark_prepared_extrude_sources -- --ignored --nocapture

Three Windows x64 optimized ci runs, during other compilation, compared the previous segment reconstruction loop and the prepared matcher in the same process. For 1,000 analytic lines and 1,000 generated face segments, median time fell from 45.90 ms to 12.30 ms (73% less, about 3.7x faster). Preparation is included and every returned source identity is compared. This measures certification only, not complete extrusion, file-open latency or Boolean operations; it does not establish an end-to-end loading speedup. Remaining cold geometry rebuild and import costs are tracked in #255; certification work is #254.

Cold geometry rebuilds after 0.3.1

Three alternating cold-process comparisons on the same Windows machine and optimized ci profile, with no concurrent build, compare the unchanged 0.3.1 source with the straight-line projection and exact surface-refinement reuse changes:

Saved design Before, median evaluation After Reduction
Public sensor enclosure, 21 features, 52 KB 11.96 s 8.85 s 26.0%
Local Pi case assembly, 25 features, 2,990 bodies 14.75 s 11.57 s 21.6%

These are geometry rebuild times, not complete desktop opening times. The public fixture can be generated with solidcraft-cli run examples/acceptance/sensor_enclosure.json --save sensor.sc --quiet, then measured with solidcraft-cli bench-open sensor.sc. Each benchmark starts a new process; this does not measure a warm timeline-cache hit. The local assembly is private and is not part of the repository.

Straight segments now use guarded algebraic closest-point projection instead of sampling 65 points and running Newton iterations. Other curve types retain their existing behavior. Curved-face refinement reuses derivative and midpoint evaluations with exact finite input keys and bounded, local caches. It keeps the original tolerance, seam and singularity checks, refinement passes, and triangle limits.

Identity assembly placements retain the definition body's shared display-mesh cache. Changed placements continue to build separate meshes in their own coordinates. Face-name preparation remains eager during placement: deferring it merely shifted expensive work to thumbnails and the first face query, so that approach was excluded.

Both saved fixtures retain exactly the same body names, face/edge/vertex/shell counts, bounds, volumes, and surface areas. Focused tests compare cached and uncached refinement coordinates and normals bit for bit on periodic, singular and transformed surfaces.

For desktop phase profiling, set SOLIDCRAFT_PROFILE_OPEN=1 before launching the app. Interactive Open reports geometry, placement, tessellation, thumbnail and total worker times to stderr without logging the source path. Thumbnails were about 0.1 seconds of the slow assembly's open and were not the main bottleneck. Further cold Boolean and import costs remain; these changes do not make all documents open instantly.

Prepared assembly geometry and selection

Prepared world geometry is now keyed by the document, evaluated model and exact placement context instead of the session revision. Selecting a face or edge, changing visibility, or moving the view no longer discards already prepared assembly bodies and meshes. Geometry edits, pending occurrence moves, authoring context changes, undo/redo and timeline rollback still invalidate the cache. Scratch sessions share prepared geometry while their inputs match, and adopting a computed session retains its warm cache.

On the private 2,990-body Pi assembly, a matched Linux desktop control probe reduced an edge-selection round trip from 9.24 s to 0.57 s. Initial opening stayed essentially unchanged (33.54 s to 33.32 s); this fixes a separate selection stall, not cold geometry evaluation. Three fresh Windows desktop runs with the corrected cache measured a median 15.87 s to ready: 13.04 s geometry, 2.20 s placement, 0.53 s meshing and 0.07 s thumbnails. These used the optimized ci profile and actual Vulkan rendering, with isolated settings and caches. Linux and Windows timings are separate measurements and must not be compared as a speedup.

solidcraft-cli bench-open design.sc reports JSON lines for file reading, JSON decoding and migrations (including input-name repair), timeline evaluation, individual features, and world placement/display meshing. --parse-only stops before geometry evaluation. The parse timer includes document validation and repair, not just the JSON parser.

Windows x64, optimized ci profile, measured during other builds:

Design Parse/repair Evaluate Display meshes
19-feature enclosure, 34 KB 1.2 ms 14.9 s 6 ms
25-feature assembly, 15.2 MB, 2,990 bodies 21 ms 12.5 s 2.1 s
Generated 1,000 independent boxes, 185 KB 5 ms 0.82 s 0.15 s
Generated 5,000 boxes in five grids, 10 features 0.45 ms 17.6 s 0.81 s

The enclosures spend most of evaluation in joins and cuts, not parsing. Designs save feature definitions; opening them rebuilds their geometry. Those times remain after the repair fix.

A separate generated timeline with missing input names reproduced minutes in the reader:

Timeline (geometry rolled back) Parse/repair before After
2,000 boxes, 6,000 unnamed inputs 78.1 s 10–20 ms
5,000 boxes, 15,000 unnamed inputs interrupted after 131 s, still repairing 26–46 ms

Previously, each missing input restarted the d1, d2, … search, scanning the document for every candidate. This approaches cubic work as unnamed inputs accumulate. Loading now reserves names once and assigns missing names with an advancing cursor. Existing names, names on later features, parameter collisions and reuse of surplus names retain the previous behavior. Regression tests compare the old and bulk repairs and load a 5,000-feature unnamed timeline.

Viewport performance

Measured with the control channel (ui.inspect frame_ms: CPU time of one UI frame, which includes picking under the cursor) on the sample plate patterned 20 × 20 (400 bodies, about 600k triangles), 1600 × 1000 window under Xvfb, release build. Hover: 60 cursor moves across the model; orbit: a right-drag.

hover median hover p90 orbit median
before 18.7 ms 37.5 ms 21.0 ms
after 1.9 ms 2.4 ms 2.6 ms
after, Browser hidden 0.4 ms — 0.4 ms

What changed:

  • Picking culls by bounding box. Per-body boxes are cached until the model changes. The face raycast skips bodies whose box the ray misses or enters behind the nearest hit so far. Edge and vertex tests skip bodies whose box, projected on screen, is not under the cursor.
  • Picks are reused while the cursor, camera and model are unchanged, and no picking runs while the view moves (orbit, pan, zoom).
  • The model extent is cached (it sets the view's depth range and was recomputed from every mesh several times per frame).

With 49 bodies the frame was already 2.8 ms (hover) before these changes. The remaining frame time with the Browser shown is mostly the Browser listing every body row. GPU ID-buffer picking was not needed at this size.

Browser

Same model and method (400 bodies, Bodies folder open, 60 hover moves over the viewport):

median p90
Browser shown, before 1.95 ms 2.71 ms
Browser shown, after 0.72 ms 1.19 ms
Browser hidden 0.30 ms 0.72 ms

What changed: rows scrolled out of view only reserve their height. They register no widgets and paint nothing (no text layout, icons or eye/fold buttons), so the cost follows the rows on screen rather than the rows in the tree. Scrolling, Find in Browser and drag and drop still see every row's place.

Recompute performance

solidcraft-cli bench-edit <script> <param> "(<old>) * 1.02" builds a design from its script (timing the build), then times a parameter edit, setting the parameter back to its old expression, undo and redo. Each cell is before → after the evaluation cache, with the number of features recomputed in parentheses. Release build, on a machine shared with other build jobs (load average about 45; the times vary by ±50 % run to run, so compare the recompute counts and the zeros). "Early" edits a parameter of the first sketch or feature, "late" one of the last features.

Design Edit Build Edit Set back Undo Redo
sample plate, 20 × 20 pattern early (width) 7587 ms (10) → 1764 ms (10) 7969 ms (10) → 1691 ms (10) 7557 ms (10) → 0 ms (0) 6534 ms (0) → 0 ms (0) 4570 ms (0) → 0 ms (0)
sample plate, 20 × 20 pattern late (d9) 7870 ms (10) → 1699 ms (10) 3944 ms (1) → 812 ms (1) 4039 ms (1) → 0 ms (0) 1464 ms (0) → 0 ms (0) 1398 ms (0) → 0 ms (0)
oracle 58 phone stand early (d1) 163 ms (6) → 140 ms (6) 228 ms (5) → 140 ms (5) 171 ms (5) → 0 ms (0) 158 ms (0) → 0 ms (0) 130 ms (0) → 0 ms (0)
oracle 58 phone stand late (d4) 144 ms (6) → 195 ms (6) 118 ms (1) → 118 ms (1) 145 ms (1) → 0 ms (0) 149 ms (0) → 0 ms (0) 187 ms (0) → 0 ms (0)
oracle 59 enclosure base late (d8) 11582 ms (11) → 11179 ms (11) 5772 ms (1) → 5742 ms (1) 5606 ms (1) → 0 ms (0) 6023 ms (0) → 0 ms (0) 6119 ms (0) → 0 ms (0)
oracle 60 pipe flange early (d1) 8137 ms (7) → 8348 ms (7) 7569 ms (6) → 10143 ms (6) 8354 ms (6) → 0 ms (0) 8607 ms (0) → 0 ms (0) 10034 ms (0) → 0 ms (0)
oracle 60 pipe flange late (d6) 7661 ms (7) → 11990 ms (7) 67 ms (1) → 361 ms (1) 69 ms (1) → 0 ms (0) 84 ms (0) → 0 ms (0) 66 ms (0) → 0 ms (0)
oracle 61 shaft keyway early (d1) 1723 ms (6) → 3016 ms (6) 1785 ms (5) → 2745 ms (5) 1760 ms (5) → 2323 ms (5) 1667 ms (0) → 0 ms (0) 1719 ms (0) → 0 ms (0)
oracle 61 shaft keyway late (d4) 1905 ms (6) → 2247 ms (6) 1734 ms (1) → 2101 ms (1) 1682 ms (1) → 0 ms (0) 1719 ms (0) → 0 ms (0) 1876 ms (0) → 0 ms (0)
oracle 62 heat sink early (d1) 382 ms (6) → 643 ms (6) 362 ms (5) → 697 ms (5) 233 ms (5) → 568 ms (5) 245 ms (0) → 0 ms (0) 360 ms (0) → 0 ms (0)
oracle 62 heat sink late (d5) 479 ms (6) → 688 ms (6) 270 ms (1) → 657 ms (1) 321 ms (1) → 0 ms (0) 413 ms (0) → 0 ms (0) 283 ms (0) → 0 ms (0)
oracle 63 hinge knuckle early (d1) 1305 ms (8) → 1671 ms (8) 1329 ms (7) → 2397 ms (7) 1971 ms (7) → 0 ms (0) 1705 ms (0) → 0 ms (0) 1321 ms (0) → 0 ms (0)
oracle 63 hinge knuckle late (d4) 1430 ms (8) → 2072 ms (8) 1387 ms (1) → 2073 ms (1) 1480 ms (1) → 0 ms (0) 1192 ms (0) → 0 ms (0) 1035 ms (0) → 0 ms (0)

What changed: every feature evaluation is cached by a key made of the feature's own inputs (definition and parameter values), the document context it may read (construction planes, sheet metal and plastic rules, components) and a digest of the model it starts from, in which each body is identified by the evaluation that produced it (bodies a feature passes through unchanged keep their identity). The same feature on the same inputs is never evaluated twice: setting a value back, undo, redo, a preview followed by the same command, and the repeated evaluations while a script builds the design all reuse earlier results. The cache is shared by a model and its copies (previews, scratch sessions), and keeps the last 4096 results within 64 MiB.

An edit still recomputes the edited feature and every later feature whose starting model changed; in these parts every later feature works on the edited body, so early edits recompute almost everything, as they must. cached_evaluation_matches_from_scratch checks that cached and from-scratch evaluation give the same models (bodies, volumes and areas to 1e-6, topology, errors) through edits, suppression, roll-back, undo and redo. Setting a sketch dimension back can still recompute (oracle 61, 62): solving the sketch moves its points by rounding noise, so its inputs are not exactly the old ones. Oracle 59's first parameter edit fails (a recipe expression the bench's * 1.02 can't scale) and is left out.

Prism pairs, body measurements and the evaluation cache budget (2026-10-13)

A design whose features repeat one shape (a hole pattern, a plate cut by many tools, a part built by one growing body) pays several per-feature costs that a rebuild was paying again and again.

A body's size is measured once, not once per call. Every kernel tolerance is a fraction of the body's bounding-box diagonal, and Body::size() walked every vertex of the solid to work it out. Body::new already reads those vertices to orient the body, so it keeps the diagonal it measured there; a body built by hand, or a mesh body, still measures once.

The same volume is not measured twice. A boolean measures both operands' volumes to bound its result, and the body it returns becomes the next feature's operand, so a timeline of n cuts meshed every intermediate body twice. Body::volume_at(tol) keeps the last measurement for the solid, shared by its clones (they share the solid), and the bounds ask at the same tolerance every time.

A prism pair is decided in 2D before either operand is meshed. The general boolean builds a refined mesh of both operands before it can decide anything, and that mesh is most of its cost: a pattern, or a sketch with several profiles, rebuilds the whole target's mesh once per tool. A plate and a tool through it are prisms along one direction, which prism::prism_boolean decides exactly from their 2D sections - and that route ran after the meshes. A cut or an intersection now asks it first and returns without building them. A pair the route cannot answer takes the old path unchanged, meshes and all; a union still asks after the sampled shortcut. One answer is sent back to that shortcut on purpose: when the 2D route returns one of its operands (a tool that misses its target, or covers it whole), the sampled path returns that very body, so an untouched body keeps its identity in the timeline (ops::prism_fast_path_tests).

A placement reads its orientation from its matrix. Placing a body (a pattern instance, an aligned copy) used to mesh the whole copy to decide which way its faces point and to count the faces its coarse mesh misses. A rigid placement leaves both as they were; a mirrored one (a negative determinant) turns the solid inside out. Both are now read from the matrix, so only the geometry is transformed.

The evaluation cache is bounded by the bytes it keeps. It held the last 4096 results, and a result is as big as the model it holds, so a long editing session kept every superseded state: nothing dropped the entries nothing referred to any more. It now also holds at most 64 MiB, oldest first, and always keeps the newest result, so a single evaluation is never thrown away in the same breath.

Release solidcraft-cli, eval <design> --quiet, wall time of the whole process and peak RSS from /usr/bin/time -v, alternating the baseline and patched binaries, median of 12 runs each (8 for the 400-hole plate and the vent plate), no other build or test running (Fedora Linux, AMD Ryzen 7 5700X, 8 cores / 16 threads, 32 GB):

Design Baseline Patched Change Baseline peak RSS Patched peak RSS
400 × 400 × 6 mm box, one sketch with 100 circles cut through it 1.24 s 0.83 s -33% 151 MB 119 MB
The same plate with 400 circles in the sketch 5.67 s 4.04 s -29% 556 MB 481 MB
Vent plate, a 10 × 8 pattern of one cut 0.10 s 0.08 s -20% 22 MB 22 MB
14 × 14 pattern of a filleted box (196 copies) 6.70 s 6.52 s -2.6% 67 MB 67 MB
60 sketches, 60 extrudes into one growing body 8.80 s 8.88 s +0.9% 225 MB 225 MB
examples/acceptance/sensor_enclosure.json 3.81 s 3.30 s -13% 198 MB 196 MB

The hole plates are what the prism route is for: one boolean whose tool is a prism with 100 or 400 sections, in a sketch with that many circles, so the tool's mesh alone is what the route saves. A variant of the patched tree with the route asked in the old order - after the operands' meshes - timed the 100-hole plate at 1.27 s and 157 MB (range 1.22-1.30 s), the baseline's own numbers (1.25 s and 156 MB over the same runs, range 1.21-1.33 s). The two plate rows, and the peak they lose, are the prism route.

The last three designs are the other side: their booleans are not that case. The 196-copy design and the 60-sketch design are within 2.6% and 0.9% with overlapping ranges. The sensor enclosure's medians differ by 13%, but its runs land anywhere between 3.0 s and 5.0 s and the two ranges overlap (3.01-4.97 s against 3.00-4.89 s), so its row is not a result either way and is left as measured. The body measurements, the placement and the cache budget are not what these rows measure: they stop work the corpus repeats too rarely to see (a hidden vertex walk, an operand meshed twice for one number, an orientation mesh per placement, a cache with no byte bound).

No model changed. The evaluation dump of 21 designs (10 to 400 circles in a plate, 16 × 16 to 31 × 31 box grids with and without fillets, 2500-body grids, assemblies of 40 to 500 bodies, 30 and 60 sketches, a 196-instance pattern, the bundled sample and the sensor enclosure) is the same model before and after - bodies, volumes, areas, centres, bounding boxes, topology counts and warnings - apart from the per-feature milliseconds. The kernel's test suite passes, including four new tests for the prism route and its identity rule.

Reproduce after compiling a release build, without other builds running: time target/release/solidcraft-cli eval <design> with /usr/bin/time -v, alternating the baseline and patched binaries, and compare the middle of the runs. The corpus (20 generated designs: hole plates, box grids, assemblies, many-sketch parts) and the measuring scripts are in https://github.com/sahenjp/solidcraft-perf under perf/.

Desktop loading and recent thumbnails (2026-10-10)

The Arduino Uno enclosure (19 features, 34 KB) and Raspberry Pi 5 case (24 features, 61 KB) each decode in about 1 ms. A release build spends roughly 12.1 s and 8.2 s, respectively, evaluating their geometry on this Windows machine. Changing the document encoding does not eliminate these boolean operations.

Interactive Open, recent-file clicks, drag/drop and STEP/mesh insertion now execute on a worker, including preparation of the world-space bodies and display meshes. The existing design stays available until the result is ready; a failed or cancelled job cannot replace it. Insertion preserves the file, saved state, earlier undo history and view settings. Timeline copies share their lazy mesh cell, and adopting a prepared session keeps its warm world-space cache instead of transforming and tessellating its components again on the UI. A STEP body's missing-face check reuses its orientation tessellation when the tolerances match and its orientation was not reversed.

Live desktop tests (ci profile, with other builds/tests running) opened the Arduino and Pi case in 15.2 s and 10.5 s. During those jobs, median ui.inspect round trips were 20.5 ms and 8.3 ms; the slowest were 33.7 ms and 28.6 ms. This fixes the window freeze, not the remaining cost of cold geometry evaluation.

Recent thumbnails are 240 × 150 PNG files in SolidCraft/thumbnails under the platform cache directory (LOCALAPPDATA on Windows). Their keys include the source path, file size and modification time. Open and Save render from the model already in memory. Cached recents decode the PNG without reading or rebuilding a document, and run before cold recents and sample thumbnails. Missing or invalid pictures fall back to a background rebuild. No thumbnail or base64 image is added to the design JSON. A regression test verifies the source bytes remain unchanged, a cached picture works even without a parseable source file, and changing the source invalidates the cache.

solidcraft-cli bench-open also accepts STEP/IGES. For exchange files its parse phase includes constructing and checking the imported B-rep, not merely decoding STEP syntax.

The live test loaded both cached recent cards in 21 ms and inserted the 2 MB lightweight Pi STEP (27 bodies) into the case in 279 ms; undo restored the original two bodies. The full 77 MB board STEP (2689 solid definitions, 45,689 faces) still had not completed after 300 s. Its window continued handling inspection and Cancel, and cancelling preserved the case. Complex STEP construction remains an outstanding kernel performance problem.

Sketch solver and drags (2026-10-11)

Unknowns the constraints pin down one after another from the constants (each one's own rows of full, well-conditioned rank, reading nothing but constants and unknowns pinned before it) are eliminated exactly: values the constructive seed placed and whose constraints hold there stay constants while Levenberg–Marquardt solves the rest, and the rank analysis (degrees of freedom, determined geometry) runs on the remainder only. A component tied to no constant takes one point as a constant first (its translation). All the held solves of one live drag frame share a 1M decomposition work budget (about 16 ms; a solve's own, 20M, is about 0.3 s), and a drag's steps are a quarter of the smallest held size instead of an eighth, except once a seed ran out of budget: the rest of that drag goes in eighths, as before, and continuation keeps the branch the seed no longer places.

cargo run --release -p solidcraft-sketch --example solve_bench, macOS arm64, load average 45–75 (best of 5):

Grid (origin corner fixed, rows horizontal, columns vertical, every edge a distance) Start Before After
10 × 10 at solution 0.36 ms 0.21 ms
10 × 10 ±1 mm 0.34 ms 0.21 ms
15 × 15 at solution 1.41 ms 0.41 ms
15 × 15 ±1 mm 1.39 ms 0.40 ms
20 × 20 at solution 4.06 ms 0.72 ms
20 × 20 ±1 mm 4.44 ms 0.73 ms

The seed constructs every grid point, so LM gets no variable (a test counts them). Before, the time was the dense rank analysis of the whole grid. A 100-rectangle strip with nothing fixed, solved whole from a moved corner, still costs 20–27 ms: nothing is constructed there and LM iterates on every point with dense normal equations (a sparse solve would be the next step).

Drag frames, sketch.drag + world_state() per frame (60 frames of 0.5 mm, best of 5; the "before" column is one run of the same bench on the previous code):

Sketch Points Scripted ms/frame before → after Live ms/frame before → after drag_end ms before → after
100 rectangles 401 2.77 → 2.37 1.13 → 1.02 1.44 → 1.37
100 rectangles, rolled back under an extrude 401 2.49 → 2.33 1.13 → 0.97 1.42 → 1.28
100-rectangle connected strip 401 100.9 → 2.32 1.32 → 1.20 53.6 → 1.41
100-rectangle strip, rolled back under an extrude 401 203.8 → 2.23 1.31 → 1.18 54.0 → 1.42
rigid truss of 40 triangles, pinned at one end, dragged at the other 42 — → 6.8 — → 6.3 — → 0.59

The strip floats (nothing fixed), so its full solves (scripted frames, drag_end) ran the dense rank analysis on 800 variables; with one corner taken as a constant all of it is eliminated. A live frame's held solve was already cheap: the seed places the strip and LM returns at once. Of a live strip frame, the seed reads the constraints in about 0.15 ms, so a per-gesture cache of that reading is not worth its invalidation (its signs depend on the reference positions, which change every step).

Portable thumbnail save cost (#303)

Portable thumbnail generation requires explicit cache: true. Ordinary saves and browser downloads default to cache: false, so they incur none of the thumbnail work or added bytes measured below.

Measured 2026-10-13 on the same Windows machine and ci profile as above, but with SolidCraft 0.7.0 on both sides: d3d215b before the thumbnail change and thumbnail revision e180cdc after it. These measurements precede the integration of upstream 9f88537; later geometry changes are not included in these timings. A preserved executable of the baseline and the new executable ran the existing benchmark_native_cache_save on the same STEP and sensor documents. Three paired batches alternated before/after order; each batch performed nine alternating atomic saves per encoding. The values below are medians of the three batch medians, not a percentile of all individual saves. No compilation, test suite or desktop model was running during these measurements.

512-cube STEP encoding Before PNG (ms) After PNG (ms) Container bytes before → after
Plain JSON 10.334 9.676 1,998,826 → 1,998,826
Container, cache: false 8.850 9.013 1,950,084 → 1,950,084
Native cache, fast 45.323 53.473 4,694,035 → 4,720,385
Native cache, compact 42.739 48.643 837,067 → 863,417

The PNG itself is 25,916 bytes; its key and ZIP records bring the increase to 26,350 bytes. Fast-cache batch medians ranged from 41.585–47.779 ms before and 53.212–56.556 ms after. Compact ranged from 36.672–43.040 ms before and 47.688–51.724 ms after. The approximately 8 ms fast-save increase is a measured cost on this example, not a general latency bound. JSON bytes and document/source blob bytes were compared across both implementations and remained identical.

The sensor enclosure deliberately receives no portable PNG: its projected triangle boxes total 4,293,457 supersampled pixels, above the four-million work budget. The native fast container remains 379,099 bytes. Its save batch medians are 14.714 ms before (12.187–14.956) and 13.728 ms after (13.439–14.648); these vary within the observed run ranges and do not establish a speedup. Home retains its existing thumbnail-generation fallback for this case. A small triangle count alone does not guarantee eligibility, because large overlapping faces cost more.

Actual desktop check: copy the STEP container to a new public temporary path, start SolidCraft 0.7.0 with that path in Recent and an empty local thumbnail cache, and show Home without opening the design. The portable picture was ready in one observed 47.6 ms check; no local PNG was created and the active document remained empty. The screenshot shows the transferred file's picture while built-in sample pictures are still loading. This is one functional visual check, not a first-frame benchmark. The separate counting-reader regression proves source-blob deferral.

Extrude drag preview delivery

On macOS with Metal, editing a 15 mm radius cylinder's 10 mm extrusion and dragging its arrow upward by 150 screen points (120 input steps, Ctrl held) exposed preview starvation. A completed worker result was discarded whenever the pointer had already changed the requested distance. The arrow also drew the previous frame's distance.

Main's tool-only preview already delivers intermediate Join/Cut/Intersect geometry. New Body extrusions (including feature edits) and Offset Face use full-result previews and still discard completed distances during a moving drag. Those paths now accept successful intermediate distances from the same setup. Selections, operation, extent mode, feature identity, revision and colours remain part of the setup identity. Errors still apply only to the exact requested inputs, and the final distance is computed normally. Tool previews and background Apply retain their existing paths. The arrow and value-box anchor use the current frame's pointer value.

Release-build control-channel checks against upstream 4dc63810 with the arrow fix alone, then with the distance fallback, used the same 150-point, 120-step drags. The cylinder used the setup above; planar-face Press Pull (Join) and Offset Face used a 30 × 30 × 10 mm box. The samples captured 25–27 distinct requested distances per drag:

Preview updates before reaching the final distance Main + arrow fix With distance fallback
Edited New Body extrusion 0 24
Planar-face Press Pull (Join) 22 24
Offset Face 0 23

Join already streams on main; the small count difference reflects sampling, not a change to that path. Applying the fixed New Body extrusion produced the requested 19.585 mm height; applying the 11.672 mm face offset produced a 21.672 mm tall box.

These are sampled preview deliveries (ui.inspect.preview.ms changes), not every rendered frame or a geometry-kernel speed benchmark. With main's preview code unchanged, the New Body, edited New Body and Offset Face drag regressions fail while Join passes. With the fallback, all six distance-preview regressions pass, covering intermediate geometry, exact final convergence, unchanged document/undo, obsolete errors and setup changes. A separate regression checks the arrow's position on each held input frame.

Placement-only model reuse

Model::evaluate_shared(&Arc<Document>) retains the document that produced a model and a weak reference to that exact model state. A subsequent shared document can keep the model without entering Model::evaluate when its geometry inputs are unchanged and all its features have explicitly supported local inputs. Session opening and refresh use this entry point; drive, reach, occurrence movement, scratch sessions and Undo/Redo share its decision. Joint solving, motion traces, view revision and world placement still follow the normal refresh path.

Eligibility is computed during ordinary evaluation with its existing feature-owner index. The initial supported set is new-body boxes/cylinders/spheres/tori, embedded STEP/mesh imports, sketches on explicit local planes without links or 3D wires, and new-body extrudes of their own component's sketches without To Object, From Object or Through All. Other features, attachments, failed/skipped/stale results, primed evaluation and no_cache retain ordinary evaluation. The borrowed-document evaluate invalidates the retained proof. A saved model establishes a shared-input proof on its first ordinary refresh; this does not change saved model formats. Renaming results and adopting solved sketches also invalidate it: result labels and fingerprints can change while the geometry state stays identical.

Every non-placement document field and every feature field is compared explicitly. Immutable import arrays/text are accepted by retained Arc identity; replacing an allocation conservatively evaluates again, including an equal-content replacement. New fields of the document, feature, mesh or supported feature variants require a decision in this comparison/eligibility policy. There is still a linear comparison of lightweight feature definitions and result freshness on a new document Arc. This is not a constant-time geometry revision system; the avoided work is the evaluator, parameter/context preparation, fingerprints and imported-payload traversal.

Regression tests in the existing placement-cache and evaluation-cost suites check zero timeline evaluations, real world-space vertices/normals/topology against fresh evaluation, actual joint poses, retained snapshots, scratch and Undo/Redo. Direct geometry/parameter changes, mesh payload and metadata replacements, foreign model results, live sketches and unknown features invalidate or retain the fallback. The cross-component cut control is a 10 x 10 x 4 mm plate and a 2 x 2 x 6 mm tool: after a 5 mm slider move it still evaluates to 392 mm3, undo to 384 mm3, redo to 392 mm3.

Release measurements (2026-10-25)

Windows x64, Intel Core i9-14900K, Rust 1.99.0, release profile (opt-level 3, thin LTO, 16 codegen units). Baseline is main ac53fc1e6fab85a50bdc410178cc75133183393f; the changed build adds only this placement-reuse implementation. Each new process warms canonical meshes and eight operations, then runs 64 alternating +10/-10 degree drives/reaches. Three pairs alternate which build runs first. Numbers below are medians of the three per-run command/total medians, in milliseconds. Total includes the command, world placement and canonical display/reference mesh preparation; it excludes UI scene packing, GPU/presentation and input-to-photon latency.

To separate additional benefit from #5480, a second comparison includes that same PR's 4720ac3c3f296bfc1d957c0363c656bce8d845cd implementation in both builds, preserving main's feature-JSON and body-measurement caches when integrating it locally:

Case, with #5480 in both builds Command before Command after Total before Total after
One moving component, 256 box bodies 0.8599 0.5457 2.7114 2.3632
SO-101 drive 0.3829 0.1363 1.4217 1.1349
SO-101 reach (the drag command) 0.4024 0.1401 1.4474 1.1314
Cross-component cut, fallback control 0.0335 0.0339 0.0842 0.0866

That is 36.5% less command time / 12.8% less total for 256 bodies, and 65.2% / 21.8% for SO-101 reach in this integration. The fallback control has no intended speedup; its small absolute variation is retained in the table. Against main alone, 256 bodies measure 0.8666 -> 0.5524 ms command and 2.7039 -> 2.3913 ms total. SO-101 reach measures 2.8656 -> 0.1443 ms command and 4.0597 -> 1.1411 ms total; that larger difference overlaps #5480 and is not additive to it.

Every pair agrees on all 64 command responses and occurrence matrices, the final world-space position/normal/topology hash, body/triangle counts and Undo/Redo. SO-101 retains its public 13 bodies / 398,884 triangles; its input SHA-256 is a335456ea4fba3729d41f901374575c785a14d2c4e6bf3bb7081d64062f3f7cb.

A separate counting build uses #6465's recorder from 969c15c2, with only the model-evaluation and feature-fingerprint hooks wired. It records 1 evaluation / 256 fingerprints -> 0 / 0 for 256 boxes, and 1 / 1 -> 0 / 0 for SO-101 drive and reach. The cut control stays at 1 / 3. The captures are complete for those two instrumented units; no claim is made about the other recorder units. The control explicitly disables reuse while retaining the ordinary feature cache. These counting-only hooks/switch are absent from the published source and timing builds.

Reproduction code, exact source/executable hashes and per-operation release samples cover both baseline contexts and the independent counting probe.