Does ordering start/sit by the conditional line plus a post-inactives swap beat ordering by p_play x line (the served E[pts])?

Owner lane: allocation

Question

Does ordering start/sit by the conditional line plus a post-inactives swap beat ordering by p_play x line (the served E[pts])?

Prediction

Registered on the H-223 row (filed 2026-09-15T1209Z) and restated in prereg/2026-09-23T1044Z-h223-swap-vs-discount.md: eligible pairs 8-15% of close pairs; on eligible SWAP-DISCOUNT regret -0.10 to -0.30 and accuracy +1 to +3pp, both excluding zero; pooled regret gain under 0.03; consensus margin over shipped on eligible shrinks by a third or more; NIL if picks differ on fewer than 3% of close pairs. REFUTED if the eligible regret interval covers zero or favours DISCOUNT.

Finding and verdict

TESTED -- REFUTED, WRONG SIGN. Ordering by the conditional line plus a post-list swap costs +0.162 regret per eligible decision (Bonf-2 [+0.014,+0.299]) and does not help accuracy (-0.7pp, spans 0). On this room and this model the served p_play x line is the better start/sit number. That does NOT show the discount is right in principle: the likelier reason is that the conditional line runs high at low p (QB2), and the discount cancels it. The swap step itself carries information (PERM control), but reaches 0.6% of decisions. Consensus beats both, and its margin grows under SWAP.

Reasoning

H-091 found p_play's value lives in no MAE endpoint; H-223 asked if it lives in the decision. It does, in the direction shipped already uses.

Next action

(1) The separating run above -- fair-rescaled c under the same SWAP rule, same pairs -- is H-223's own follow-up, not a new row; it needs its own prereg. (2) The floor on the served line hides nearly every scratched man from the decision set; a floor on c would admit them, and that is where a swap can pay. Not run; the family above records it as NOT RUN. (3) Consensus margin on designated pairs 0.48 vs 0.42 on all close: the consensus ranks designated men better than shipped does.