What does serving coach_pos_tgt = coach_pos_car = 0.0 on every row (what the board will serve once served rows are 2026 rows and rel_coach still stops at 2025) cost against the shipped values, and does a carry-forward of each coach's last prior-season value recover it?

Owner lane: allocation

Question

What does serving coach_pos_tgt = coach_pos_car = 0.0 on every row (what the board will serve once served rows are 2026 rows and rel_coach still stops at 2025) cost against the shipped values, and does a carry-forward of each coach's last prior-season value recover it?

Prediction

Frozen 2026-09-23T1636Z in prereg/2026-09-23T1636Z-h067-coach-zero-price.md before the run: E1 zero-shipped POSITIVE, +0.02 to +0.10 pts/wk, 99% interval excluding zero; largest per-position cost on RB; QB within +/-0.02; shuffle costs less than zero; lag within +/-0.02 of shipped; mean signed move under zero NEGATIVE on RB/WR/TE. REFUTED if E1's 99% interval contains zero or zero-shipped < +0.005. (The filer's H-067 row predicted +0.01 to +0.04, largest on RB.)

Finding and verdict

TESTED. The post-flip coach pair costs +0.016 pts/player-week pooled [99% +0.004, +0.028], concentrated on RB (+0.029) with QB unresolved (+0.032, spans zero); it lowers the within-week rank correlation from 0.7206 to 0.7188. Carry-forward (lag) recovers all but +0.003 (interval spans zero), so the cheapest writer is a coach-keyed carry-forward, not a new materialisation.

Reasoning

H-239 found rel_coach goes dark at the first board after week 3 (~09-29); this is the price of letting it.

Next action

A live writer for rel_coach (or a coach-keyed carry-forward at serve) is a build item that needs a separate semantic review and PROMOTION receipt; this record is EXECUTION only. Re-measure on the first post-flip board by diffing served coach_pos_* against 0.