# Verification checks (P4)

All PASS (run 2026-10-30):

• Clean re-run (deleted results.json, analytic.csv, exclusions.csv): results.json byte-identical; analytic.csv identical md5.
• analytic.csv rows (308) == n_analysed.
• n_months_growth + n_months_missing_growth == 309 window months.
• n_months_growth_2001_2013 (156) + n_months_growth_2014_2026 (152) == 308.
• Exclusion ledger: n_before − n_removed = n_after each row; chained; final n_after = 308; 957 − 649 = 308.
• Hand recomputation from raw vintage files: growth_2022_06 (new/new), growth_2014_06 (new/old vintage mix), min_growth_12m 2009-07 (old/old) match to 1e-9.
• Weighted-mean identity: (156·mean_2001_2013 + 152·mean_2014_2026)/308 == mean_growth_12m.
• min ≤ mean ≤ max; all cpi, lag, growth finite.
• vintage_used == 2026-10-28 exactly for dates ≥ 2014-01 (latest-vintage rule).
• 2025-10 absent from analytic.csv (no imputation).
• Magnitude: overall mean 2.57% close to baseline 2001–2013 mean 2.36%; peak 8.98% in 2022-06 and trough −1.96% in 2009-07 are plausible for US CPI.
• No CIs/SEs: estimands are deterministic functions of published data; no sampling uncertainty applies.