M4.1b — Calibration tests in CI

[Ferrari-72]

Parent

Spawned from #8 (M4.1 tracking issue).

Objective

Add the verifier's known-answer tests (fixture certificates from #2, novelty/judgment tests from #5) to the CI workflow so that any future change that weakens the checker fails the build automatically.

Definition of done

• .github/workflows/publish.yml (or a separate ci.yml) runs make verify-calibration and pytest -m judgment on every push
• A deliberate breakage of verify.py causes CI to fail
• make test-unit still passes

Dependencies

Depends on #2, #5, #7.

[Ferrari-72]

Completed — commit f4721fc

What was done

.github/workflows/ci.yml (new file)

Triggers on every push to master and every PR. Two steps after checkout + pip install:

1. pytest -v -m "not integration" — all unit tests including judgment tests (robust equality, novelty filter, HTML structure, CI workflow structure, docs)
2. python -m benchmark.verify --calibrate — the 3 canonical fixture certificates must all get the expected verdict

If either step fails, the CI job fails. Any future change that weakens verify.py (making it accept a false alarm or reject a real bug) will break calibration and block the PR.

test_ci_workflow.py — 4 new tests (TestCIWorkflow)

Test	Verifies
test_ci_workflow_file_exists	.github/workflows/ci.yml present
test_ci_triggers_on_push_and_pr	triggers on push or pull_request
test_ci_runs_unit_tests	pytest invoked
test_ci_runs_verify_calibration	verify --calibrate or verify-calibration invoked

Test results

145 passed, 2 skipped   (make test-unit)