H2 — Adaptability report
The original H2 was: after an abrupt demand shock, MAS will restore the 95% service level faster than Periodic Forecasting. Empirically, the hypothesis is vacuous in this experimental setup — both adaptive policies maintain post-drift service level above 99% in every scenario (the EOQ-derived order quantity absorbs Periodic's between-refit stale window). Time-to-recovery is 0 for both, so the original H2 cannot distinguish the policies.
The substantive finding has moved to the cost dimension — both policies achieve the same service level, but at very different prices. The manuscript's Discussion §5.2 reframes H2 as H2′ (Cost-efficiency under drift): at equivalent post-drift SL (≥ 99%), MAS will incur at least 30% less total cost than Periodic Forecasting, with Mann–Whitney p < 0.05.
Why the original H2 is empirically vacuous
Post-drift mean service level for every (policy × scenario) cell. Both mas and periodic_forecasting sit above the 99% line in every column, so a 95% target is never breached and time-to-recovery is identically zero.
| Scenario | MAS | Periodic | Static ROP | Verdict |
|---|---|---|---|---|
| Loading… | ||||
H2′ supporting data — total cost: MAS vs Periodic
Per-scenario mean total cost across N = 10 seeds. Δ vs Periodic is the relative change ((MAS − Periodic) / Periodic). Mann–Whitney p is the primary significance test; a value < 0.05 means the cost reduction is statistically distinguishable from chance under the null of equal medians.
| Scenario | MAS cost | Periodic cost | Δ vs Periodic | MW p | Welch p | Cohen's d | Threshold (−30%) |
|---|---|---|---|---|---|---|---|
| Loading… | |||||||