Does 23:17Z's under-shrinkage headline (-season_td target share | position correlates + with the shipped targets residual at weeks 2-4) hold on seasons no H-021 record scored (2017-18, fresh production refits), is it specific to f or a property of the line, where does it extend, is it a split or level error, and does it survive the board's MAE metric?

Owner lane: allocation

Question

Does 23:17Z's under-shrinkage headline (-season_td target share | position correlates + with the shipped targets residual at weeks 2-4) hold on seasons no H-021 record scored (2017-18, fresh production refits), is it specific to f or a property of the line, where does it extend, is it a split or level error, and does it survive the board's MAE metric?

Prediction

Frozen in protocol.json (sha d94f1b2a) before any fit: P0 positive outside Bonf-8 on NEW in [+0.06,+0.15]; S1 survives the line control; S2 >= P0; S3 positive <= half; S4 nil; S5 positive; S6 positive; S7 within >= 70%. Amendment 1 (before analyze): R1 k in [1,3] games WR/TE; R2 >= P0. Amendments 2-4 post-hoc, each with its own prediction in its file.

Finding and verdict

REFINED, and 23:17Z's headline REFUTED out of selection. P0 on the unscored seasons 2017-18 is +0.011 [-0.051,+0.076], inside Bonf-8 (refutation clause met). The season-share lead is not under-shrinkage: it is the mean line's over-dispersion (slope 0.83 at every week, not an early-season defect) netted against an opposite and larger f-given-line under-read. Both are MEAN properties. Scored in the board's units (MAE, median fit walk-forward), f adds nothing (9+: -0.004 [-0.010,+0.002]); the only MAE gain is recalibrating the mean line to a median (-0.04 targets/row), which is D-199's known gap and already fixed at the edge. H-021 end to end: still MEASURED NOTHING.

Reasoning

Two lanes stacked a mechanism (early-season, under-shrinkage) on a lead selected on 2019-25 and measured by Pearson against a mean line, with linear bounds multiplying a Pearson rho by an MAE. The next arm would have shrunk f harder, which (b) says is the wrong direction.

Next action

(1) Frozen, prospective: 2026 weeks 2-4 ISSUED lines after 2026-10-06: P0 predicted inside band (net of opposing parts), S1 negative. (2) Structure: which features carry the mean line's over-dispersion given f is under-read (the l3/l3_est terms are candidates), on the mean endpoint only if a mean consumer is named. (3) Do not arm a harder shrinkage of target_share_season: (b) says the line needs more f, not less.