# Contract — F6-conf-01

Question. Does the weekday carry predictive content for the daily change in the 10y Treasury CMT yield (FRED DGS10), 2000–2019? Estimands. (i) Joint equality of weekday conditional means of Δy_t (bp) — classical one-way equal-variance ANOVA F over all changes (F(4,n−5)); (ii) Monday contrast = mean(Δy|Mon) − mean(Δy|Tue–Fri), all changes, with 95% CI (choice: Welch two-sample t interval); (iii) out-of-sample RMSE (2010–2019 changes) of weekday-mean forecast (means from changes dated ≤2009-12-31) vs zero-change forecast. Assumptions: changes approximately independent across days for the classical tests (serial dependence/heteroskedasticity is a threat); ANOVA assumes equal variance (Welch ANOVA as robustness only). Info rules. Drop empty DGS10 rows first; Δy between consecutive available obs, dated at t, weekday of t; first obs has no change. Train = changes dated ≤2009-12-31 (including the change on the first 2010 day? no: that change is dated in 2010 so it's test, its y_{t-1} from 2009 is legitimately known). Weekday means for forecasting from train only. Only data/DGS10.csv. Outputs. out/results.json (keys: n_missing_rows, n_train, n_test, anova_f_stat, anova_p_value, effect_estimate, effect_ci_low, effect_ci_high, baseline_metric, model_metric, conclusion, task_id, feasibility, feasibility_reason, data_issues, n_input_rows, n_analysed, random_seed), out/figure1.png, out/table1.csv (weekday,n,mean_bp,se_bp), out/report.md (6 sections), out/analysis.py, out/claims.json. Conclusion rule. ANOVA p≥0.05 → no_detectable_effect; else sign from Monday CI (positive/negative; if CI covers zero despite ANOVA rejection → no_detectable_effect, documented). Invalidity. Missing/unparseable data making changes uncomputable → premise_invalid; using test-period info in weekday means → invalid.