| 1 |
Input SHA-256 equal to data/README.md and bundle.json |
pass (all 7 files) |
| 2 |
Clean re-run (results.json deleted, regenerated) byte-identical |
pass (cmp identical) |
| 3 |
Body title line == header Title for every extract (assert in script) |
pass |
| 4 |
Entry arXiv number appears in Bibliographic record (assert) |
pass (5/5) |
| 5 |
unverifiable_references == ids failing >=1 check; n_verified = 5 - len(unverifiable) = 4 |
pass |
| 6 |
n_analysed (5) = n_raw (5) - excluded (0) per exclusions.csv |
pass |
| 7 |
n_input_rows equals wc -l (README 17, sources 134/58/107/66/91) and 5 JSON items |
pass |
| 8 |
Arithmetic: R1 exp(-0.431)-1 = -35.0% consistent with printed % change; R4 -0.268/0.366 = -0.7322 ≈ -0.732; R5 -0.0962/0.261 = -0.3686 vs printed -0.3685 (rounding) |
pass (rounding noted) |
| 9 |
Baseline (title/year only) flags R3 only; full check also flags only R3 (finding fails too) |
consistent |
| 10 |
Year rule: R4 entry 2025 = v1 year (2025) though extract is v3 (2026) -> consistent per rule |
pass |
| 11 |
No CI/SE keys in results to test positivity (task has none) |
n/a |