# Checks (P4)

- Reproducibility: deleted out/results.json, re-ran out/analysis.py, `cmp` with previous copy: IDENTICAL -> PASS
- all required keys present: PASS
- 5 references R1-R5: PASS
- n_verified + len(unverifiable) == n_analysed: PASS
- unverifiable == refs failing any check: PASS
- each locator names source id: PASS
- n_input_rows matches wc -l (17,5,134,58,107,66,91): PASS
- random_seed null, feasibility complete: PASS
- baseline (metadata-only) subset of final: PASS
- Arithmetic identities verified in script: exp(-0.431)-1 = -35.0% (R1); -0.268/0.366 = -0.732 (R4); -0.0962/0.261 = -0.3686 vs printed -0.3685 (rounding, R5); 0.431/0.210 and 0.431/0.218 both > 1.96 (R1) -> PASS
- Year rule: only Bibliographic record dates used (HTML rendering dates ignored) -> PASS
- No CIs/SEs estimated by us (none to check for positivity); all numbers are printed source values.