# Contract (F2-conf-01)

• Question: average effect of a school-assigned programme on student test scores, with design-valid uncertainty; decide at 5% (two-sided) whether τ≠0.
• Estimand τ: difference in student-weighted mean score, treated minus control schools = OLS slope of score on treated (with intercept). Meaningful as a causal effect if school assignment is as-good-as-random (simulated design); inference assumes schools independent, students within school correlated.
• Uncertainty: treatment varies only between schools → cluster-robust SE at school level (main); iid OLS SE as naïve baseline only.
• Outputs: out/results.json (task_id, tau_hat, se_iid, se_cluster, p_value, tau_ci_low, tau_ci_high, effect_significant_5pct, icc_hat, n_schools, n_treated_schools, n_analysed, feasibility, feasibility_reason, data_issues, n_input_rows, random_seed), out/report.md (6 sections), out/analysis.py, out/claims.json (incl. 5% claim).
• Information rule: only data/schools.csv.
• Invalidity: treatment varying within school, too few clusters (or only one arm), missing/duplicate records not handled, using iid SE for the decision.