# Contract (P0) — F6-conf-04
**Question.** Effect of one extra training hour per employee (programme year) on next-year productivity growth (pp per hour), simulated firm data.
**Estimand.** Causal, linear, homogeneous coefficient of `training_hours` on `productivity_growth` under selection-on-observables given pre-programme `skill_index` (the only pre-treatment covariate). Meaningful only if no unobserved confounders, skill_index is pre-treatment (stated: measured year before), linearity/homogeneity hold, and positivity/overlap in hours across skill levels.
**Baseline.** OLS of growth on training_hours + intercept (naive slope, 95% CI, RMSE sqrt(mean e^2)).
**Outputs.** out/results.json (naive_estimate, naive_ci_low/high, baseline_metric, model_metric, effect_estimate, effect_se, effect_ci_low/high, skill_index_coefficient, conclusion, task_id, feasibility, feasibility_reason, data_issues, n_input_rows, n_analysed, random_seed); out/figure1.png; out/table1.csv (model, estimate, se, ci_low, ci_high); out/report.md (6 sections); out/analysis.py; out/claims.json; plus state ledger files and out/exclusions.csv.
**Information rules.** Use only data/firms.csv; no imputation; no other variables. skill_index is pre-programme, so adjusting for it is allowed (not a post-treatment control).
**Invalidating conditions.** Missing/unusable treatment or outcome; skill_index post-treatment; no overlap; impossible values silently used.