# Verification checks (P4)

• Clean re-run of out/analysis.py: results.json byte-identical (cmp) -> PASS
• n_analysed == rows analytic.csv -> PASS
• n_analysed == n_balanced*23 -> PASS
• balanced+incomplete==total -> PASS
• unique (iso3,year) in analytic -> PASS
• no NaN in analytic -> PASS
• ledger chain -> PASS
• ledger start=n_input_rows, end=n_analysed -> PASS
• change == m2022-m2000 -> PASS
• recompute means from analytic -> PASS
• baseline recompute 2022 (20 countries, TUR once) -> PASS
• magnitudes 40-90 years -> PASS
• balanced_countries sorted -> PASS

Note: baseline (78.692, all 20 countries) is lower than balanced 2022 mean (81.553) because it includes the 4 low-LE countries (BGD, ETH, IND, NGA) excluded from the balanced set; direction expected.