feat: quality regression test suite + DomI sync (#75, #78) by domattioli · Pull Request #65 · domattioli/QuADMESH

domattioli · 2026-05-31T19:19:46Z

Summary

Rolling daily-maintenance PR. Updated 2026-06-04T12Z.

2026-06-04T12Z session (commit `c3e0695`)

fix: regression 39bac0e — _split_opposing_tri orphan tris — _split_opposing_tri now returns None early when np_id >= domain.points.shape[0] (midpoint buffered in work._extra_pts, not yet flushed). Commit 39bac0e had proceeded using a combined pts array, which appended new rows to domain.connectivity_list that were never tracked in consumed, leaving 54 orphan boundary tris in method="faithful" on Test_Case_1.14. Early return restores pre-regression behaviour. 18/18 test_no_interior_tris pass; 240/240 fast suite pass.

2026-06-03T12Z session (commit `ed09597`)

fix: test_conforming + _boundary_edges self-loop guard — ADCIRC connectivity_list stores tris as [n1,n2,n3,n1]. Iterating all 4 positions created spurious (n1,n1) self-loop edges, failing test_conforming[Baranja_Hill.14] with Non-conforming edges: [((321, 321), 3)]. Fix: skip edge when a == b. Applied to both _boundary_edges and test_conforming. Verified on Baranja_Hill.14 + Test_Case_1.14.

2026-06-03T05Z session (commit `95de755`)

chore: sync DomI@bc29b51 (chore: sync DomI@bc29b51 #78) — .domi-pin refreshed 9911cd7 → bc29b51.
feat: quality regression test suite (feat: quality regression test suite — pin mean quality baselines per canonical fixture #75) — tests/test_quality_regression.py (4 @pytest.mark.slow tests), tests/fixtures/quality_baselines.json, --runslow flag in conftest, slow marker in pyproject.toml.

2026-06-02T12Z session (commits `35523bc`, `0a24286`, `f5f8a11`)

chore: sync DomI@9911cd7 (chore: sync DomI@b9094d3 #74) — startup script canonical v2; issue templates added.
docs: introspect corpus

2026-06-02T02Z session

DomI sync, routine update, introspect corpus.

2026-06-01 session

feat: skeleton implementation (feat: rename "skeletonization" to consider that layers and "skeleton" arent necessarily the same thing #55) — 65/65 tests pass.
fix: _split_opposing_tri IndexError
docs: specs 002 + 055 completed.

2026-05-31 sessions

feat: truss_smoother
DomI sync, spec updates, test fixes.

Test status

Fast suite (excluding test_faithful_invariants.py multi-fixture sweep + --runslow tests): 240 pass, 4 skip. All test_no_interior_tris pass including faithful path.

Skipped (operator decision required)

Issue	Reason
#46 — onion domain	ADMESH-Domains#93 still open
#20 — label triage	5 repo-specific labels need operator decision
#75 — faithful/WNAT baselines	Tighten once C++ CHILmesh backend compiled
#17, #18, #26, #28, #38	`status: brainstorming` / research-only
#21 — size function drift	Research; requires C++ backend for timely execution
#76 — profiling	Research; same constraint
#77 — boundary-type seeding	Blocked on CHILmesh #129

https://claude.ai/code/session_01LQqvkgn1uMPUDx3q4qURkx

Remove Tri2Quad and Demo sections referencing deleted GIF/MP4 artifacts. Remove those entries from ToC. Update all Zenodo DOIs to 10.5281/zenodo.20351165. https://claude.ai/code/session_019TnFhzo2fX57VQyQSnwJ59

- .github/workflows/tests.yml: pytest matrix (3.10/3.11/3.12) on push/PR to main and daily-maintenance; installs chilmesh from GitHub + quadmesh[dev] - git mv matlab/ src/matlab/ (frozen MATLAB reference lives under src/) - Update README and CLAUDE.md paths: matlab/ -> src/matlab/ https://claude.ai/code/session_019TnFhzo2fX57VQyQSnwJ59

…ri fan Addresses #28. Prototype of the quad→4-tri-fan spring-force smoother: - Each quad split into 4 tris (centroid node) to restore shear stiffness - Frozen edge topology: spring forces only move vertices, no retriangulation - Boundary nodes pinned; interior nodes converge toward target edge length h0 - fh=None falls back to median quad edge length - Opt-in: truss_smooth=True in post_process_routine / run_pipeline - Called before fem_smoother so FEM corrects any angle artifacts afterward Implements the "4-tri fan + vertex-only truss" design from the #28 investigation comment. The collapsed-quad failure mode (4-bar linkage) is solved by the fan's diagonal bracing. https://claude.ai/code/session_019TnFhzo2fX57VQyQSnwJ59

…n guard Fixes #28 prototype issues exposed by Block-O testing: - Centroid bars use structural-only springs (L0_init) → no compression - Perimeter bars use local h0 (median neighbor edge lengths) → respects local density - Spring forces repulsive-only (clip max(L0-Lbar,0)) → no overshoot inversion - Per-step inversion detection and revert → preserve mesh topology Results: stable smoothing on variable-density meshes without FEM pre-pass. Without proper size function fh(x,y), can't improve on FEM alone, but provides structural stabilization that could benefit other algorithms. https://claude.ai/code/session_019TnFhzo2fX57VQyQSnwJ59

domattioli · 2026-05-31T20:16:39Z

Testing & refinements complete

Pushed: centroid L0 fix, local h0 per-node, repulsive-only forces, per-step inversion guard.

Block-O results (after topo cleanup):

Strategy	mean	min	notes
FEM only (baseline)	0.741	0.094	current QuADMESH+
Truss only	0.461	0.000	repulsive-only, no FEM
Truss→FEM	0.741	0.094	same as FEM alone

Key finding: Truss smoother is structural stabilizer (pushes apart compressed edges) but can't improve mean quality without proper size function fh(x,y). Global h0=median causes overshoot on 44× variable-density mesh. Centroid springs + local h0 + repulsive-only forces eliminate inversion risk but reveal architectural limitation.

Next: #28 hypothesis only holds if ADMESH's fh threaded through pipeline. Otherwise, truss is useful for pre-conditioning / deflating compressed zones before FEM, not standalone quality improvement.

See #28 comment for full findings.

Generated by Claude Code

https://claude.ai/code/session_019TnFhzo2fX57VQyQSnwJ59

…t fixes - chore: sync DomI@e0bba05e9ec495bde2790738ad93b17d0e33c20f - docs: add spec for skeletonization rename (#55) — specs/055-skeletonization-rename/spec.md; rename docstrings in _layer_state.py, mesh_structure.py, tri2quad.py, validator.py, test_layer_state.py - fix: pass WorkingMesh to _split_opposing_tri so buffered midpoint resolves (fixes 2 pre-existing test failures in test_tri_removal*) - chore: introspection corpus 2026-05-31"

Keeps Claude-specific files off PyPI wheels and Zenodo snapshots. Mirrors DomI .gitattributes. Closes DomI #179 (QuADMesh portion).

Synced from DomI canonical set (spec 007, 2026-05-31). Enables sync-labels.yml workflow to recolor/prune label definitions on merge.

Mirrors DomI .github/workflows/sync-labels.yml. On merge to main, recolors canonical labels + deletes deprecated definitions via EndBug/label-sync with delete-other-labels: true. Closes DomI #162 (QuADMesh portion) and satisfies DomI #176 label-definition cleanup path.

…it_opposing_tri (#55) - mesh_structure.py: kind="skeleton" implemented per operator 2026-05-30 definition (morphological peeling via CHILmesh layer ordering outermost->innermost). Adds skeleton_core / skeleton_core_verts properties exposing layer N-1 (irreducible core) for skeleton-vs-medial_axis tri2quad comparison (#55). - _tri_removal.py: _split_opposing_tri accepts optional work: WorkingMesh; builds combined pts array when np_id >= domain.points.shape[0] — fixes IndexError when buffered midpoint not yet flushed to domain.points. - tests/test_mesh_structure.py: 6 new skeleton tests replace not_implemented stub. - specs/055: status Complete; Research Section updated with morphological def. - specs/002: catalog [GAP] filled; 4 new methods added (QuadriFlow, FD+Blossom, IGM, Spectral); citation ranking; Blossom+layers complexity analysis added. - .domi-pin: refreshed to DomI HEAD e2501f6 (local checkout fallback; network N/A).

- _tri_removal.py: pass work to _split_opposing_tri; combined pts array prevents IndexError when buffered midpoint not in domain.points. - specs/002: algorithm catalog complete — [GAP] entries filled, 4 new methods added, citation ranking + Blossom-layer complexity analysis. - specs/055: Research Section updated with morphological skeleton def per operator 2026-05-30; implementation noted as shipped.

Session session_01Uzf7kkhAirHH278vJYWyoS. Issues closed: #55 (skeleton), #66 (domi-sync). Issue progressed: #9 (catalog gaps filled). 65/65 tests pass.

…permission-prompts bootstrap step

…lates (#74) - Replace scripts/instructions_on_start.sh with DomI canonical v2 (full consumer health checks: CLAUDE.md, git hygiene, test infra, speckit, git hooks, recovery cheat-sheet) - Extract QuADMesh-specific Python port smoke to scripts/onstart_local.sh - Add .github/ISSUE_TEMPLATE/bug_report.yml, feature_request.yml, skill_request.yml (canonical DomI templates) - Refresh .domi-pin to DomI@9911cd7 Resolves #74

Missed from previous commit — replace with full DomI canonical version (21KB vs original 2.6KB). Adds CLAUDE.md check, full git hygiene, consumer repo checks, git hooks setup, recovery cheat-sheet, and onstart_local.sh extension point.

) DomI sync (#78): .domi-pin refreshed 9911cd7 → bc29b51 Quality regression suite (#75): - tests/test_quality_regression.py: 4 @slow tests (TC1 matching/faithful, Block_O matching, WNAT_Hagen matching) skipped unless --runslow - tests/fixtures/quality_baselines.json: baseline store - tests/conftest.py: --runslow flag + skip gate - pyproject.toml: slow marker registration https://claude.ai/code/session_013wH7FC3jVUtuugLM2tUZiT

Session: DomI sync (#78) + quality regression tests (#75). Outcome: partial — baselines conservative (pure-Python CHILmesh too slow for live measurement). Key pain: C++ CHILmesh not compiled in fresh containers → can't run quality tests. https://claude.ai/code/session_013wH7FC3jVUtuugLM2tUZiT

Reduces permission prompts for common GitHub read ops per fewer-permission-prompts skill. All listed tools are read-only (list/get/search). https://claude.ai/code/session_013wH7FC3jVUtuugLM2tUZiT

ADCIRC connectivity_list stores tris as [n1,n2,n3,n1]. Iterating all 4 positions created a spurious (n1,n1) self-loop edge counted 3x per tri element, failing test_conforming[Baranja_Hill.14] with: Non-conforming edges: [((321, 321), 3)] Guard: skip edge when a == b. Applied to both _boundary_edges and test_conforming. Verified: Baranja_Hill.14 and Test_Case_1.14 pass.

…-local

…utine sessions

) `_split_opposing_tri` now returns None early when `np_id >= domain.points.shape[0]` instead of building a combined pts array and proceeding. The 39bac0e approach caused it to append new rows to `domain.connectivity_list` for tris in already-consumed layers — those rows were never tracked in `consumed`, leaving 54 orphan boundary tris in the faithful path on Test_Case_1.14. Early return restores pre-regression behaviour (matches the IndexError that `_ccw_tri` previously raised with an out-of-bounds index). All 18 test_no_interior_tris tests pass; 240 fast-suite pass.

Claude (default) added 3 commits May 30, 2026 23:35

Clean README: remove stale GIF sections, fix Zenodo DOI

9b4ad89

Remove Tri2Quad and Demo sections referencing deleted GIF/MP4 artifacts. Remove those entries from ToC. Update all Zenodo DOIs to 10.5281/zenodo.20351165. https://claude.ai/code/session_019TnFhzo2fX57VQyQSnwJ59

domattioli mentioned this pull request May 31, 2026

Can we apply the admesh truss solving alg with respect to the size function on a resulting quadmesh #28

Open

Claude (default) and others added 2 commits May 31, 2026 21:08

docs: session introspection — truss smoother proto (2026-05-31)

a900853

https://claude.ai/code/session_019TnFhzo2fX57VQyQSnwJ59

This was referenced May 31, 2026

feat: rename "skeletonization" to consider that layers and "skeleton" arent necessarily the same thing #55

Closed

chore: sync DomI@ca87f9c #61

Closed

domattioli added 11 commits May 31, 2026 18:13

chore: add .gitattributes export-ignore for .claude/ and CLAUDE.md

738b79d

Keeps Claude-specific files off PyPI wheels and Zenodo snapshots. Mirrors DomI .gitattributes. Closes DomI #179 (QuADMesh portion).

chore: add .github/labels.yml — DomI canonical label taxonomy

d6b040e

Synced from DomI canonical set (spec 007, 2026-05-31). Enables sync-labels.yml workflow to recolor/prune label definitions on merge.

chore: introspection corpus 2026-06-01

e492f34

Session session_01Uzf7kkhAirHH278vJYWyoS. Issues closed: #55 (skeleton), #66 (domi-sync). Issue progressed: #9 (catalog gaps filled). 65/65 tests pass.

chore: sync routine — code_shipping_allowed=true for ADMESH + /fewer-…

f2f0ccd

…permission-prompts bootstrap step

docs: introspect corpus 2026-06-02T02Z (CI chore session)

b5e8789

docs: introspect corpus 2026-06-02T12Z — DomI sync session (#74)

f5f8a11

domattioli changed the title ~~feat: truss_smoother — size-function-respecting quad smoother via 4-tri fan~~ chore: sync DomI@9911cd7 + startup script canonical v2 (#74) Jun 2, 2026

domattioli changed the title ~~chore: sync DomI@9911cd7 + startup script canonical v2 (#74)~~ feat: quality regression test suite + DomI sync (#75, #78) Jun 3, 2026

domattioli added 6 commits June 3, 2026 01:48

chore: add read-only MCP allowlist to .claude/settings.json

1760778

Reduces permission prompts for common GitHub read ops per fewer-permission-prompts skill. All listed tools are read-only (list/get/search). https://claude.ai/code/session_013wH7FC3jVUtuugLM2tUZiT

docs: introspect corpus 2026-06-03T12Z — test_conforming self-loop fix

37f9f9e

chore: label triage decisions (#20) — document downstream-api as repo…

9bd4416

…-local

chore: add write permissions.allow + caveman plugin for autonomous ro…

4586471

…utine sessions

Claude (dom macbook) added 2 commits June 4, 2026 12:27

docs: introspect corpus 2026-06-04T12Z — regression 39bac0e fix

da4d4a6

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: quality regression test suite + DomI sync (#75, #78)#65

feat: quality regression test suite + DomI sync (#75, #78)#65
domattioli wants to merge 26 commits into
mainfrom
daily-maintenance

domattioli commented May 31, 2026 •

edited

Loading

Uh oh!

domattioli commented May 31, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

domattioli commented May 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

2026-06-04T12Z session (commit c3e0695)

2026-06-03T12Z session (commit ed09597)

2026-06-03T05Z session (commit 95de755)

2026-06-02T12Z session (commits 35523bc, 0a24286, f5f8a11)

2026-06-02T02Z session

2026-06-01 session

2026-05-31 sessions

Test status

Skipped (operator decision required)

Uh oh!

domattioli commented May 31, 2026

Testing & refinements complete

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

domattioli commented May 31, 2026 •

edited

Loading

2026-06-04T12Z session (commit `c3e0695`)

2026-06-03T12Z session (commit `ed09597`)

2026-06-03T05Z session (commit `95de755`)

2026-06-02T12Z session (commits `35523bc`, `0a24286`, `f5f8a11`)