Is the anchor's pooled 2016-onward rush-TD slope conditionally calibrated on held-out seasons across the implied-total range, and does a trailing-window slope, the only fix a forecaster could have had at the time, buy anything on the anchor's own mean-minimising metric?

Owner lane: calibration

Question

Is the anchor's pooled 2016-onward rush-TD slope conditionally calibrated on held-out seasons across the implied-total range, and does a trailing-window slope, the only fix a forecaster could have had at the time, buy anything on the anchor's own mean-minimising metric?

Prediction

Frozen in the runner docstring (sha b6886aca) at 2026-09-26T0118Z before any held-out scoring: rush calibration slope of the production anchor > 1.2 on 2023-25; every W-arm dMSE interval spans zero at Bonf-12 (nil); W1/W2 points slightly negative on rush 2024-25; throwing nil.

Finding and verdict

TESTED. H-161's premise STRENGTHENED: the rush-TD slope step is not one season, it is 2023, 2024 and 2025 each ~2 SE above 2016-22, and held out the pooled production anchor is conditionally MISCALIBRATED on rush TDs: calibration slope 1.31 [95% 1.08, 1.55] on 2023-25 (Bonf-8 lower edge 1.00), high-minus-low implied-tercile bias +0.132 TD per team-game [+0.031, +0.235]. A trailing 1-season slope removes it (calibration slope 1.03, gap +0.046). On the anchor's own mean-minimising metric the fix is MEASURED NOTHING: W1 -0.0028 [Bonf-12 -0.0108, +0.0040] on 2024-25 against an MSE of 0.884, and this season's own slope in hindsight is no larger (-0.0029). Throwing TDs are calibrated (0.97-1.03) and every window is nil. So the defect is conditional bias of about 0.07 TD per team-game at each implied-total extreme, real in its sign and in three seasons, too small to resolve on team-game squared error; its consumer is anything priced on a tail of the implied-total range (anytime-TD, RB TD props at high totals), not the average projection.

Reasoning

A slope gap of 0.02 TD per implied point at an implied-total SD near 4.5 moves the mean by ~0.09 at one SD, against a team-game rush-TD variance of 0.90: ~0.3% of MSE at best, which is what the oracle shows.

Next action

(1) Re-score the frozen P-vs-W1 calibration slope on 2026 at week 17 (WAITING, not a new row). (2) Anytime-TD / rush-TD prop calibration at the top implied tercile is where this bias would be priced; that is a props-lane question and is noted, not claimed.