Does the width of the served floor/ceiling band order two men with close projected means (top tail, bust, spread), and does an underdog rule on it win synthetic head-to-heads?

Owner lane: forecast

Question

Does the width of the served floor/ceiling band order two men with close projected means (top tail, bust, spread), and does an underdog rule on it win synthetic head-to-heads?

Prediction

E1 +1 to +3 pp, E3 +0.2 to +0.8 pts, E2 positive, E0 <= 0, E4 99% interval contains 0; H-207 FAILS its bar on E4. Frozen in experiments/2026-09-26T0817Z-h207-band-ordering.prereg.json (commit 50920b4c).

Finding and verdict

H-207 FAILS ITS BAR, and the band carries no SPREAD-ordering skill: the width is a location signal. E1 (top tail) +3.44 pp [99% +1.86, +5.04], outside the shuffle null; E4 (underdog rule, synthetic head-to-head) -0.085 pp [-0.174, +0.008], inside the shuffle range and nil. Among close RB/WR/TE pairs 2023-25 (30,069, played=1 SELECTED) the wider-band man busts LESS (-4.47 pp) and his absolute error is no larger (+0.14 pts, spans 0). Width tracks (sim_mean - line) at r=0.92. Even a planted directional hindsight width moves the head-to-head only +0.24 pp: in this design a single-slot choice among close means is worth well under a point of win rate, so H-207's statistic 2 was a weak instrument whatever the band. The curve band (QB, and the fallback) is a function of mean and p_play by construction.

Reasoning

The spread question needs a band whose width is not dominated by a second centre. The follow-up question the data forced is H-292.

Next action

H-292 (filed and run this session). A spread question proper needs a band whose centre equals the line: re-centre the sim band on the line (band - sim_mean + line) and re-ask E1-E3; and a head-to-head graded on real fp_league_matchup lineups rather than synthetic ones.