# Contract — F1-conf-01

Question. 12-month growth of US CPI (CPIAUCSL, SA), g_t = 100*(P_t/P_{t-12} - 1), for every month 2001-01 … 2026-09, and summaries.

Estimands. g_t per month; mean over window (309 months); sub-period means 2001-01…2013-12 and 2014-01…2026-09; g_t at 2014-06, 2016-07, 2020-05, 2022-06, 2026-09; max/min g_t and months. Meaningful under: P_t taken from most recent vintage covering the month (2026-10-28 for 2014-01…2026-09, 2016-07-15 otherwise). Note: g_t for 2014-01…2014-12 mixes vintages (P_t from 2026 vintage, P_{t-12} from 2016 vintage) — this is what the rule prescribes.

Outputs. out/results.json (task_id, first_growth_month, last_month, n_months_growth, n_months_missing_growth, n_analysed, mean_growth_12m, mean_growth_12m_2001_2013, mean_growth_12m_2014_2026, growth_2014_06, growth_2016_07, growth_2020_05, growth_2022_06, growth_2026_09, max_growth_12m, max_growth_month, min_growth_12m, min_growth_month, feasibility, feasibility_reason, data_issues, n_input_rows, random_seed); out/analytic.csv (date,cpi,cpi_lag12,vintage_used,growth_12m; n_analysed rows); out/exclusions.csv; out/report.md; out/analysis.py; out/claims.json.

Information rules. Most-recent-vintage per month; no interpolation/imputation; missing g_t reported and counted; only data/ files.

Invalidity. Using the hand-assembled file instead of vintage files where it errs; imputing missing levels; using a level from an older vintage when the newer covers the month.