Does the prop path's one-band-per-quantity conformal calibration (props.conformal_cal: 2024 rows, weeks 1-18 pooled) hide a weeks-1-3 regime -- higher rushing and QB throwing centres and a wider residual band -- that mis-prices early-week P(over)? And, since Sean's first stake (2026-10-10) is 2026 WEEK 5, not weeks 1-3: do the early rows distort the band used in weeks 4-18?

Owner lane: decision

Question

Does the prop path's one-band-per-quantity conformal calibration (props.conformal_cal: 2024 rows, weeks 1-18 pooled) hide a weeks-1-3 regime -- higher rushing and QB throwing centres and a wider residual band -- that mis-prices early-week P(over)? And, since Sean's first stake (2026-10-10) is 2026 WEEK 5, not weeks 1-3: do the early rows distort the band used in weeks 4-18?

Prediction

Frozen in prereg/2026-09-24T0222Z-h185-prop-band-early-weeks.md before any outcome was read. P1 QB throwing-yards centre HIGHER in weeks 1-3 than 4-18 by +2 to +12 yd; carries/rush_yards differences inside +/-0.4 / +/-4 with intervals spanning 0; receiving inside +/-0.2 / +/-3. P2 weeks-1-3 80% coverage below weeks 4-18 by 0.00-0.04 on every quantity. P3 arm A early bias on QB throwing yards +0.01..+0.07; rushing within +/-0.04 of the later bias. P4 the week-bucket Mondrian arm falls back for QB throwing yards; its Brier gain sits inside the shuffled-label null for >= 3 of 5 quantities; late-bucket A-W inside +/-0.01. My prediction REFUTED if a rushing E1 contrast exceeds +0.4 carries / +4 yd with the Bonferroni-15 interval clear of 0, or the QB throwing-yards E1 is <= 0 with the interval clear of 0.

Finding and verdict

TESTED -- H-185 REFUTED AS FILED, REFINED. Weeks 1-3 are not where the prop band is widest (coverage early-minus-late -0.017 to +0.040, all spanning 0; QB residuals narrower early) and not where the QB throwing centre runs highest (-9.6 yd, lower early in 4/5 folds). One piece survives at 95% but not at the frozen family level: the rush_yards centre runs +1.04 yd higher in weeks 1-3 (5/5 folds, week-relabel p 0.03), consistent with early offences throwing more and running less in our own team rows. The claim's premise that weeks 1-3 are the betting weeks is stale: the first stake (2026-10-10) is week 5. The version the data carries: the pooled band over-prices OVERs in EVERY week bucket (+0.018 to +0.068 at a line at our own number, all five quantities clear of 0 at 99.67% in weeks 4-18), and week-banding does not touch it.

Reasoning

Kept the drawn row: it sits on the props path this lane owns, H-181's production harness made it one run, and the premise had two separable parts (location vs width) that a single contrast would have confounded. Extended to five folds because three early weeks per season gives 6 blocks at H-181's two folds.

Next action

(a) The all-week OVER over-pricing is the money-relevant defect for week-5+ bets and belongs with H-180(a): a calibration on the pairs the prop path prices, WITH the H-041 level correction -- re-measured on P(over) with real book lines once W-003 lands. (b) Early-week rush_yards over-projection (+1.0 yd) as a frozen family-of-one on 2016-2020 folds (held out from this run) would confirm or kill it; the mechanism to test is TRAIN_MIN_WEEK=4 hiding the early pass lean (H-070's territory, mechanical lane). (c) One calibration season cannot band QB throwing yards per bucket or carries-early at all; any Mondrian calibration on this path needs two seasons.