1# Contract — F1-conf-022**Question.** How much did mean life expectancy at birth (WDI SP.DYN.LE00.IN) change between 2000 and 2022 across a *fixed* set of countries?3**Estimands.** Balanced panel = countries (identified by iso3) with a recorded, non-missing value in every year 2000–2022. For each year: unweighted mean over the balanced set, each country once. Main quantities: mean_2000_balanced, mean_2022_balanced, their difference, plus the composition of the balanced set. Meaningful under: values are correctly recorded; one observation per iso3-year; unweighted (not population-weighted) country mean; inference describes these specific countries only.4**Baseline.** mean_2022_all_available: unweighted mean over all distinct iso3 with a recorded 2022 value (descriptive only, not the estimand).5**Outputs.** out/results.json (task_id, n_countries_total, n_countries_balanced, n_countries_incomplete, balanced_countries, n_analysed, mean_2000_balanced, mean_2022_balanced, change_2000_2022_balanced, mean_2022_all_available, feasibility, feasibility_reason, data_issues, n_input_rows, random_seed); out/analytic.csv (iso3,country_name,year,life_expectancy; n_analysed rows, unique iso3-year); out/exclusions.csv (chained ledger); out/report.md (6 sections); out/analysis.py; out/claims.json.6**Information rules.** Use only the panel CSV; no imputation/interpolation/lookup; empty cell = missing; country = iso3, max one obs per year.7**Invalidity.** Conflicting duplicate values for the same iso3-year (no rule to choose) would make affected countries undeterminable; zero balanced countries would make the estimand infeasible; counting a duplicated country twice or imputing gaps would invalidate results.8