H1 — Performance report

For each drift scenario (no_drift, gradual, seasonal, abrupt, severe_abrupt, catastrophic) we compare the MAS policy against each baseline (Static ROP, Periodic Forecasting) on two outcomes: stockout rate (service failure) and total cost (the sum of holding, ordering, and stockout penalties). Each cell pools 10 seed replications.

Significance is tested with Mann–Whitney U as the primary (non-parametric, our seed distributions aren't normal), Welch's t as a parametric cross-check, and Cohen's d for effect size. Bolded rows in the tables below indicate Mann–Whitney p < 0.05.