# Contract (P0)

Question. Real-time out-of-sample accuracy of 1- and 2-quarter-ahead forecasts of US real GDP growth (A191RL1Q225SBEA, % q/q SAAR) at 8 forecast origins (vintage dates 2024-01-31 … 2025-12-31), naive vs one simple time-series model, scored against first-release values. Estimands. RMSE and MAE per horizon h∈{1,2} for naive (last value in the origin's vintage) and the model; rel. improvement 1−RMSE_model/RMSE_naive per h; mean |current − first release| over the 8 h=1 target quarters. Meaningful under: each origin uses only its own vintage file; first release = advance estimate as at month-end; 8 origins → very small sample, errors serially correlated across h=2 overlaps. Outputs. out/forecasts.csv (32 rows; origin_date,target_date,horizon,forecast,actual,model,vintage_date,train_last_date,train_last_value,n_train); out/results.json (task_id, model_name, model_description, n_origins, first_origin, last_origin, rmse/mae naive & model h1/h2, rel_improvement_rmse_h1/h2, mean_abs_revision_h1_targets, feasibility, feasibility_reason, data_issues, n_input_rows, n_analysed, random_seed); out/report.md (Question; Data and exclusions; Method; Results; Uncertainty; Limitations and open issues); out/analysis.py; out/claims.json; plus state files and out/exclusions.csv. Information rules. At origin d use only vintage file dated d (all values as in it). current_vintage.csv only for revision statistic. Actuals from first_release.csv. Horizons relative to last quarter in the vintage file. Invalidity. Using current vintage or later vintages for estimation; missing first-release actuals for targets; model choice tuned on evaluation outcomes.