Skip to content

Drive linear shell-to-solid submodels and correct Q4 patch recovery - #69

Merged
audunarn merged 3 commits into
mainfrom
codex/submodel-integration-20261110
Nov 11, 2026
Merged

audunarn merged 3 commits into
mainfrom
codex/submodel-integration-20261110

Conversation

@audunarn

@audunarn audunarn commented Nov 10, 2026 •

Copy link
Copy Markdown
Owner

Coordinated submodelling review, 11 November 2026

Current review head adb0d6ac775d5c266a083e4ffa1bd86b6e7b6ea2 shares the assessed runtime source tree 6236852eac9e6e7a89061231a9f493bcd9281815. ANYfem pins the mechanics here and anysolver-results independently at e78cccd3a76feb1793a38568f59d6d68f8224cb5. The new shell-driven solid and physical Q4 recovery checks are integrated in the hosted matrix. Hosted CI also includes failures in separately owned contact, replay and qualification branches; the retained scientific criteria and failures remain in effect. Full hosted matrix.

Application acceptance record · Exact sibling revisions · Application draft

Final hosted CI

At adb0d6ac775d5c266a083e4ffa1bd86b6e7b6ea2, 39 checks passed and 13 failed. The Tests matrix finished with 36 passes and 13 failures. All three Contact Candidate jobs passed. The dedicated one-way solid/Q4 recovery step passed 57 tests. Two branch-specific test-contract defects were fixed on this head. Tests · Dedicated feature step · Contact Candidate.

Ten matrix jobs reported a named failure. Seven also failed on the observed base. Of the other three, the Ubuntu Sestra fixture emits LF where its unchanged byte validator requires CRLF, an archived S3 activation names an earlier Q4 source blob, and the curved multifacet contact restart refuses its stored checkpoint. The contact code is unchanged and the precise cause of that checkpoint mismatch remains open. Base matrix · Sestra fixture · Activation binding · Contact checkpoint.

Three Windows portable shards reached the shared 2,400-second limit without emitting a failing node. None included the four added or modified test modules. The portable runner redistributed existing modules by weight. Reconstruction of the 3.12 shard's order and completed test count put the same existing contact campaign next in both a timed-out base worker and the head worker. The logs have no live stack to confirm it was the cause. 3.13 shard 4/1 · 3.11 shard 0/1 · 3.12 shard 2/0.

Earlier review record

Review scope

Adds the one-way linear-static Q4 and S3 V2D displacement trace for a straight Tet4 child, including every P2 coefficient. The solver returns stresses and strains at the actual quadrature points, signed coefficient reactions, cut and skin forces, and moments. The existing two-way coupling disposition remains in force.

The A8 regional/cut comparison exposed a separate model-level stress-recovery defect. The corrected Q4 reader pairs stresses with their declared natural coordinates and gathers displacements in connectivity order. A cyclic permutation of 224 physical cells now reproduces both hole sensors to roundoff.

Evidence

The local focused suite passed 131 tests with ANYsolid at 06f3276d9207376316502aca651c3455db984104 and scikit-fem 12.0.2. Independent tilted V2D and disconnected P1/P2 checks cover signed offsets, driver orientation and a forged tangential locator. The corrected panel regression keeps the 7.5% bound and both refinement trends, and checks interpolation at common physical locations. The original frozen burn-in records and their historical outcomes remain distinct. Formal shell and solid assessments are tracked in the coordinated ANYfem review.

Companions are audunarn/ANYgeometry#34 and audunarn/ANYmesh#15.

@audunarn
audunarn marked this pull request as ready for review November 11, 2026 05:25
@audunarn
audunarn merged commit e0c5f32 into main Nov 11, 2026
39 of 52 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant