# Verification checks (P4)
| # | check | outcome |
|---|---|---|
| 1 | Clean re-run of out/analysis.py (results.json deleted first) reproduces byte-identical results.json | PASS |
| 2 | n_supported + n_not_supported + n_not_assessable = 6 = n_analysed | PASS (2+3+1) |
| 3 | verdicts, locators, reasons all keyed exactly A1..A6 | PASS |
| 4 | Input SHA-256 equal to README/bundle values | PASS |
| 5 | n_input_rows equal to `wc -l` (README 14, sources 134/58) and JSON item count (6) | PASS |
| 6 | A4 95% CI [-0.896, 0.084] contains estimate -0.406; also includes 0 under ±2.0·SE [-0.906, 0.094] | PASS |
| 7 | A1 percent conversion 100(exp(-0.431)-1) = -35.01% equals printed -35.0% | PASS |
| 8 | A3 percent conversion -24.65% agrees with printed -24.7% (not -28.3%) | PASS |
| 9 | A6 1.99/17.3 = 11.50% agrees with text 11.5% | PASS |
| 10 | Baseline (literal number match) vs final verdicts: baseline flags A3, A4 only; it misses A2 (sign flip, "0.063" matches literally) and A5 (no number) — expected limitation of baseline | NOTED |
| 11 | No information outside data/ used; script reads only data/ | PASS |
| 12 | A5: regex search for men/women/male/female/gender/sex in DCR extract returns no lines | PASS |