chore(physics): evidence matrix + validator check 9 (Task 7)

[neuron7xLab]

Summary

Closes Task 7 of the Physics-Invariant Rocketization Protocol — final task in the 7-task chain (T1 PR #426 → T2 #427 → T3 #428 → T4 #429 → T5 #430 → T6 #431 → T7).

• tools/physics_evidence_matrix.py — deterministic markdown table generator over .claude/physics/INVARIANTS.yaml. Per-invariant row: tier · priority · source · unit test · integration test · runtime status · on-disk ✓/✗/—. Byte-identical re-runs.
• .claude/physics/INVARIANTS.yaml — added integration_test: + runtime_evaluable: yes to the three ANCHORED runtime invariants exercised by the substrate-gate-chain suite (BEKENSTEIN-COGNITIVE, ARROW-OF-TIME, ANCHORED-SUBSTRATE-GATE). Also added the missing provenance: ANCHORED to bekenstein_cognitive — it was treated as ANCHORED everywhere but never explicitly tier-labeled (caught by check 9).
• .claude/physics/validate_tests.py — new check 9: for every invariant with provenance==ANCHORED and runtime_evaluable∈{yes,true}, the validator confirms integration_test: is declared AND resolves on disk. Same shape as checks 7/8.
• docs/physics/evidence_matrix.md — generated artefact (95 lines), 87 invariants tabulated.

Reproduced evidence

$ python .claude/physics/validate_tests.py --self-check
9. Evidence matrix OK: every ANCHORED runtime invariant has
   a resolving integration_test path
✅ Self-check PASSED  (exit 0)

# corrupt one integration_test path → DOES_NOT_EXIST.py
$ python .claude/physics/validate_tests.py --self-check
9. FAIL: 1 ANCHORED runtime invariants miss integration evidence:
   ["INV-BEKENSTEIN-COGNITIVE.integration_test = '...DOES_NOT_EXIST.py'
    does not resolve"]
❌ Self-check FAILED  (exit 1)

Determinism: two consecutive python tools/physics_evidence_matrix.py runs → diff clean.

Test plan

• python .claude/physics/validate_tests.py --self-check → 9/9 PASS, exit 0
• Negative injection (corrupt path) → check 9 FAIL, exit 1
• Deterministic regen: byte-identical output across runs
• ruff format --check clean
• ruff check clean
• black --check clean
• mypy --strict tools/physics_evidence_matrix.py → 0 issues
• mypy --strict .claude/physics/validate_tests.py → 0 issues

Closure

CLOSED. Validator now fails on intentional broken evidence; matrix regenerable and not editable by hand.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 7fbc36bb09

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Require integration_test to point to a file

The new check 9 treats any existing filesystem path as valid integration evidence, so a directory like tests/integration would pass even though no concrete test file is referenced. Because --self-check is enforced in the physics gate, this can silently bypass the intended guarantee that each ANCHORED runtime invariant names a specific integration test artifact. Validate with is_file() (after stripping any ::symbol suffix) to avoid false positives.

Useful? React with 👍 / 👎.