For the 16 player_pbp_l4 windows (15 pooled rates in pbpplayer.L4_POOLED + onfield_l4), does an 8-lag geometric pool (half-life 2, D-209) (E1) estimate the player's own next-row rate better than the shipped 4-row pool, and (E2/E3) carry anything the shipped line does not already contain?

Owner lane: serving

Question

For the 16 player_pbp_l4 windows (15 pooled rates in pbpplayer.L4_POOLED + onfield_l4), does an 8-lag geometric pool (half-life 2, D-209) (E1) estimate the player's own next-row rate better than the shipped 4-row pool, and (E2/E3) carry anything the shipped line does not already contain?

Prediction

Registered in protocol.json (sha 422b523d) at 12:17Z before any computation: (a) EWMA beats flat model-free on >= 10 of 16, every gain < 5% of flat error; (b) every |corr(x, shipped residual)| < 0.03 on all 48 cells, none outside its Bonferroni-48 permutation band; (c) no E3 walk-forward delta below its permutation 2.5th percentile after Bonferroni-16, i.e. nil; so no end-to-end refit warranted. REFUTED if any window has E2 outside its Bonf-48 band AND E3 below its Bonf-16 permutation band.

Finding and verdict

MEASURED NOTHING end to end on 16 of H-021's windows; the model-free EWMA gain is real (16/16) and does not reach the line

Reasoning

Screens 16 of H-021's remaining 33 windows in six seconds, without a refit, on the exact replica baseline. None warrants the end-to-end arm: E2 (shown to detect corr 0.05) sees at most 0.018; E3 (valid on 9) never beats shipped.

Next action

H-021 has 17 windows unscreened: target_share_season, carry_share_season, snap_pct_season (season-to-date; alternative is a cross-season EWMA, a different composition), td_oe_l8, hurry_delta_l8 (tdoe/finishing), own_scramble_rate_l4 and the 11 team windows (need MATCHED_BOTH_BALL, since the opponent copy changes). Reuse this runner shape: E2 only, since E3 cannot see through heavy-tailed x without per-position trimming. The E3 blindness on off-position rows is a harness lesson: a per-position OLS slope on a passer rate fitted over receivers is dominated by trick plays.