# Contract — T1-forecast (P0)
**Question.** In real time (using only data published at each of 10 month-end origins 2025-10-31 … 2026-07-31), how accurately can US retail & food services sales YoY growth g_m = 100·ln(L_m/L_{m-12}) be forecast h=1 and h=3 months beyond last(V) (last month in the RSAFS vintage V), and does my own model beat the naive (no-change) forecast when errors are measured against the first-release YoY (`data/first_release.csv`)?

**Estimands** (h ∈ {1,3}, 10 origins each): (a) RMSE/MAE of naive, seasonal naive, AR(1) (OLS on 1993-02..last(V), iterated) and my model vs first release; (b) 1 − RMSE_model/RMSE_naive; (c) DM stat (d_t = e²_naive − e²_model, Bartlett LRV with lag h−1, no HLN correction, N(0,1) two-sided p); (d) mean |g_current − yoy_first_release| over the 10 h=1 targets. Meaningful under: forecasts genuinely real-time; first-release actuals are a consistent target; with n=10 the DM asymptotics are only indicative.

**Outputs.** out/results.json (keys per TASK.md table: task_id, model_name, model_description, n_origins, first_origin, last_origin, rmse/mae × {naive,snaive,ar1,model} × {h1,h3}, rel_improvement_rmse_h1/h3, dm_stat/pvalue h1/h3, conclusion_code_h1/h3, mean_abs_revision_h1_targets, feasibility, feasibility_reason, data_issues, n_input_rows, n_analysed, random_seed); out/forecasts.csv (80 rows); out/table1.csv; out/figure1.png; out/report.md (6 sections); out/analysis.py; out/claims.json; out/exclusions.csv; out/state/*.

**Availability rules.** At origin V: only RSAFS vintage V, PAYEMS vintage V, UMCSENT months ≤ month(V). RSAFS_current only for (d). Targets = last(V)+h, actual from first_release.csv. vintage_used ≤ origin < target.

**Invalidity.** Using RSAFS_current or later vintages in forecasts; target month missing from first_release.csv; model selection on evaluation errors.