| """The card and SCORING.md must not repeat the claims plan 10 found false.""" |
| from pathlib import Path |
|
|
| ROOT = Path(__file__).resolve().parents[1] |
| SCORING = (ROOT / "SCORING.md").read_text() |
| README = (ROOT / "README.md").read_text() |
| PROMPTS = (ROOT / "scrub" / "prompts.py").read_text() |
| RUN = (ROOT / "scrub" / "run_eval.py").read_text() |
|
|
|
|
| def test_scoring_corrections(): |
| for stale in ["Inconsistent Language has 13", "2,599", "has 5 gold items", "dominant failure mode", |
| "never annotated", "never emits them"]: |
| assert stale not in SCORING, stale |
| for fact in ["Inconsistent Language has 87", "41 gold items", "2,678", "no Inconsistent Language row", |
| "23 of the 44"]: |
| assert fact in SCORING, fact |
|
|
|
|
| def test_readme_disclaims_and_does_not_claim_reproduction(): |
| head = "\n".join(README.splitlines()[:20]) |
| assert "not the ContractScrub dataset" in head |
| for stale in ["reproduces", "F1 well under 0.5", "accompanying report", "unsupported on SM86"]: |
| assert stale not in README, stale |
| for fact in ["0.472", "0.365", "0.642", "Known defects of the synthetic gold", "python scrub/score.py"]: |
| assert fact in README, fact |
|
|
|
|
| def test_no_fidelity_or_report_claims_in_code(): |
| assert "exactly as the paper describes" not in PROMPTS |
| assert "report" not in RUN.lower().replace("reported", "") |
|
|
