docs: redefine phase 1 rebaseline steps

2026-05-04 12:15:07 +09:00
parent a385235e14
commit 357b4bd893
20 changed files with 1013 additions and 18 deletions
@@ -0,0 +1,61 @@
+# Sprint Contract: Stored Reference Regression
+
+## Objective
+Run automated stored-reference displacement regression against accepted Phase 1-compatible S4 cases, starting with the `quad_02` compatibility path chosen in step 1.
+
+## Required Reading
+- /AGENTS.md
+- /docs/VERIFICATION_PLAN.md
+- /docs/ABAQUS_INPUT_SUBSET.md
+- /docs/RESULTS_SCHEMA.md
+- /docs/MITC4_FORMULATION.md
+- /references/
+
+## Scope
+- Wire accepted Phase 1-compatible reference inputs into automated tests.
+- Compare FESA `U` against `*_displacements.csv` using documented tolerances.
+- Keep original unsupported Abaqus files as provenance and negative/compatibility tests.
+- Record pass/fail comparison details.
+
+## Allowed Files
+- `references/`
+- `tests/`
+- `include/`
+- `src/` only for minimal fixes needed by reference regression
+- `docs/VERIFICATION_PLAN.md`
+- `PLAN.md`
+- `PROGRESS.md`
+
+## Explicit Non-Goals
+- Do not require Abaqus execution.
+- Do not accept `S4R`, `Part/Assembly/Instance`, `*Density`, or `NLGEOM=YES` through the parser without a prior parser contract.
+- Do not mark Phase 1 complete with only one reference case unless `PLAN.md` records the remaining PRD gap.
+
+## Tests To Write First
+- Reference manifest or fixture discovery tests.
+- `quad_02` accepted derivative input regression test if normalized in step 1.
+- Original `quad_02.inp` unsupported-provenance test if parser scope still excludes `Part/Assembly/Instance`.
+- CSV displacement comparison test with absolute and relative tolerances.
+
+## Reference Artifacts
+- `references/quad_02.inp`
+- `references/quad_02_displacements.csv`
+- Any normalized Phase 1-compatible derivative files from step 1.
+
+## Acceptance Commands
+```bash
+python scripts/validate_workspace.py
+```
+
+## Evaluator Checklist
+- Regression uses stored artifacts only.
+- Original reference provenance is preserved.
+- Comparison is node-id-based and reports errors clearly.
+- Remaining PRD requirement for three stored references is recorded if not satisfied.
+
+## Handoff Requirements
+- Record reference pass/fail, tolerances, and residual risks in `PROGRESS.md`.
+- Update `PLAN.md` with additional reference cases still needed.
+
+## Do Not
+- Do not tune solver formulas to one CSV without element-level tests remaining green.