Review of 2026-09-24T0519Z-h183-passtd-premise

Owner lane: Not recorded

What held

The mechanism reversal (G-TDANC), the anchored control (0.989 / 1.000 on the served path), the in-memory reverted mechanism, the shuffled-label control, the disclosure that M is a closing team total and not a prop price, the 17-interval family, and that F is far worse than the market team total (M-F -0.0755 to -0.0964 in every window). The point estimate for M-MF is right; the football-only line is at best a sliver on top of the closing total.

What did not hold

'does not beat a constant at the team grain' and 'loses to one per QB' as headline statements about the line's information content. They are two-season, raw-line results; on the extension the line beats the constant at both grains once its scale is fitted, and the raw QB-grain loss is a calibration defect (over-dispersed), not an absence of signal. 'Carries no measurable information without it' is not supported as worded: it carries a small measurable one (team grain, 2021-25), and its size is set by how many seasons are stacked.

Finding and verdict

WEAKENED

Next action

(1) Addendum, not a rewrite: the constant-comparison sentences stand only for 2024-25. (2) The route to a usable pass-TD number is scale: shrink the QB-grain line (optimal slope about r*sd_y/sd_line, i.e. well under 1) or take the scoring environment from a football source, which the record's follow_up already names; a shrink fitted on seasons < S is a shadow candidate and needs its own frozen protocol. (3) Prospective: score the 2026 pass_tds captures (285) when H-179 settles rather than a closing-total proxy.