# Contract (P0) — F3-conf-04
**Question.** Over 40 expanding-window origins (2015Q4..2025Q3), does a chosen time-series model beat the seasonal-naive forecast of Canadian quarterly NSA CPI inflation (400·Δln of quarterly-average CPI) at h=1,2,4, and is the h=4 squared-error loss differential significant with a HAC (lag ≥3) standard error?
**Estimands.** RMSE/MAE of snaive and model per h; 1−RMSE_model/RMSE_snaive per h; d̄ = mean(e²_snaive − e²_model) at h=4, HAC SE, DM stat, two-sided p-value. Meaningful under covariance stationarity of the loss differential (DM assumption); with 40 origins and overlapping errors, inference is approximate (small-sample).
**Outputs.** out/forecasts.csv (240 rows: origin_date,target_date,horizon,forecast,actual,model); out/results.json (keys listed in TASK.md); out/report.md (Question; Data and exclusions; Method; Results; Uncertainty; Limitations and open issues); out/analysis.py; out/claims.json; plus state files.
**Information rules.** Forecast at origin t uses only quarters ≤ t (estimation, lag selection all within window). Target = infl of t+h. Snaive = infl of t+h−4. HAC lag ≥ 3, report it.
**Invalidity.** Using data after t; missing target quarters (last target 2026Q3 exists); non-reproducible output.