SHIPPED_CURVE (the fantasy floor/ceiling) is fitted on TARGET-week walk-forward residuals and served around a projection made from the board's served row, whose player-history block trails one game at HEAD even after CP-H49. Does the served row break the band's coverage, and does refitting the curve on served-row residuals repair it?

Owner lane: serving

Question

SHIPPED_CURVE (the fantasy floor/ceiling) is fitted on TARGET-week walk-forward residuals and served around a projection made from the board's served row, whose player-history block trails one game at HEAD even after CP-H49. Does the served row break the band's coverage, and does refitting the curve on served-row residuals repair it?

Prediction

Frozen in the runner docstring before the run, refined from H-222's (which assumed the pre-CP-H49 row): SERVED ceiling exceedance +0.3 to +1.5pp over TARGET pooled, below-floor +0.2 to +1.0pp, pinball90 worse with a Bonferroni-2 interval excluding zero; mean width within 0.2 pts; WR/TE move most, QB least. REFUTED if the pooled ceiling-exceedance interval contains 0.

Finding and verdict

TESTED. The board's floor/ceiling is fitted on a row it does not serve, and it costs the CEILING, not the floor: on the HEAD served row (after CP-H49), SHIPPED_CURVE's ceiling is exceeded 14.3% of the time against 8.3% on the target row every band gate grades -- +5.92pp [Bonf-2 +5.51, +6.34], pinball90 +0.220 [+0.204, +0.237], 20,066 player-weeks 2023-25, stable in every season (+5.7 to +6.1pp) and on a walk-forward refit curve (+6.19pp). Below-floor +0.02pp [-0.34, +0.38] and width +0.03 pts: nil. QB -0.20pp (nil); RB +5.3, WR +6.1, TE +8.8pp. Refitting the curve on served-row residuals recovers only a third (control FAILED as registered: +3.87pp remains, band +0.50 pts wider). The exceedance lives on the rows whose last game changed their role.

Reasoning

The band is a served number a person reads (floor/ceiling on the board, lo80/hi80 in aar_forecasts) and every gate that grades it reads wf_std's target row, so this 6pp is invisible to all of them. Settles H-222 in its narrowed HEAD form.

Next action

(1) The fix belongs in the SERVER, not the curve: serve the row the band was fitted on (a Wednesday row whose windows include the last played game -- CP-H49 generalised past shares to fp_l3/fp_season/pbp/form). Re-run this runner's S arm on that row; its bar is S/ship within 0.5pp of T/ship. (2) Grade aar_forecasts hi80 exceedance on 2026 once weeks 3-4 settle -- the prospective check of this 6pp, QB-only on the imminent week because RB/WR/TE now come from sim_world. (3) Rest-of-season weeks serve the same stale row k weeks out: this is a lower bound there.