1# Contract (P0) — F6-conf-042**Question.** Effect of one extra training hour per employee (programme year) on productivity growth (pp, following year), simulated firm data.3**Estimands.** (1) Baseline: OLS slope of productivity_growth on training_hours with intercept (unadjusted association). (2) Target: causal effect of training_hours (pp/hour), meaningful under selection-on-observables given the only pre-programme covariate `skill_index` (measured year before programme, so a valid pre-treatment control, not a mediator), linear and homogeneous effect, no unobserved confounders.4**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 fixed order); out/analysis.py; out/claims.json; plus protocol files (state/*, exclusions.csv).5**Information rules.** Only data/firms.csv. Unadjusted model = OLS y ~ 1 + training_hours. skill_index is pre-programme and may be conditioned on; productivity_growth is the outcome only.6**Impossible/invalid if.** Missing outcome/treatment/covariate; no variation in treatment; covariate measured post-treatment; unobserved confounding (cannot be tested — stated as a limitation).7