Does serving the league-mean opponent block (what PM.latest_rows hands the model) instead of the real opponent's trailing block cost accuracy on the target row, and is the real opponent's block worth anything over a random one?

Owner lane: b-judgement

Question

Does serving the league-mean opponent block (what PM.latest_rows hands the model) instead of the real opponent's trailing block cost accuracy on the target row, and is the real opponent's block worth anything over a random one?

Prediction

From the H-213 row (2026-09-15), restated unchanged in prereg/2026-09-23T1400Z-h213-served-opp-block.md before the run: NIL. (b)-(a) within +/-0.01 pts/wk pooled and within +/-0.03 on WRs in the extreme quintiles of opp_points_allowed_l3; (c)-(a) better by under 0.02. REFUTED if (b)-(a) has a Bonferroni-2 interval clear of +0.01.

Finding and verdict

MEASURED NOTHING (NIL, prediction held). Serving the league-mean opponent costs -0.0034 [-0.0104, +0.0037] pts/wk; the real opponent's block is worth no more than a random opponent's (-0.00003 [-0.0076, +0.0072]) or a hindsight season mean (+0.0000).

Reasoning

The served-row opponent substitution (FINDINGS H-074 row: 'the cost of each substitution needs a target-row walk-forward') is now priced for the OPP block: zero, because the block itself is inert on accuracy. Priority consequence: refilling the served opponent block is not worth a build item; a defence object that is not inert is (H-214/H-220/H-230).

Next action

Do not re-open the served-OPP substitution. Opponent value must come from a different object: H-214 (faced-defence columns, running separately), H-230 (coverage allocation), or a schedule-adjusted points-allowed (literature below). The QB hindsight c-a +0.014 [+0.001,+0.028] is one of 8 unregistered per-position cells and is not a finding.