Verification checks (P4)
- Clean re-run: deleted out/results.json, re-ran analysis.py; byte-identical to previous output (cmp) — PASS.
- n_verified (4) + len(unverifiable_references) (1) = n_analysed (5) = number of references — PASS.
- reference_checks has exactly R1–R5 — PASS.
- n_input_rows equal
wc -lcounts (17, 134, 58, 107, 66, 91) and 5 JSON items — PASS. - SHA-256 of all extracts equal README values — PASS.
- Arithmetic cross-checks: R4 -0.268/0.366 = -0.732 (= printed); R5 -0.0962/0.261 = -0.3686 vs printed -0.3685 (difference due to rounded mean 0.261) — PASS; R1 % change -35.0% printed directly.
- Date ordering: all submission dates (2025-05-07 … 2026-09-23) precede retrieval 2026-10-30 — PASS.
- Baseline vs final: header-only baseline flags [R3]; full checks flag [R3] (finding check adds no new flags but confirms R3 fails all three) — consistent.
- Internal source inconsistencies (R1 SE 0.210 vs 0.218; 608 vs 640 controls; R4 duplicated SE) reported in data_issues, not ignored. No CI/SE positivity identities apply (no estimates produced).