27 Sep 2026 · 10:06 AM MTUpdated…
Authored research record

Research output

← All research outputs

calibrationFinding recorded

The served 80% band (intervals.shipped_band on SHIPPED_CURVE) was fitted on the UNBLENDED walk-forward centre and is served around the 40%-FantasyPros-blended centre. Does that centre mismatch cost the band calibration, and would fitting the band on the blended centre fix it?

Owner lane: calibration

Question

The served 80% band (intervals.shipped_band on SHIPPED_CURVE) was fitted on the UNBLENDED walk-forward centre and is served around the 40%-FantasyPros-blended centre. Does that centre mismatch cost the band calibration, and would fitting the band on the blended centre fix it?

Prediction

Frozen in the runner docstring before any run: nil on the primary, |d cover80| < 0.01 and |d pinball| < 1 percent; any real gap in the staleness control, not the centre. Refuted if |d pinball| >= 1 percent with CI excluding it.

Finding and verdict

TESTED. The centre mismatch H-184 names is real in code and costs the band nothing: serving SHIPPED_CURVE around the blended centre IMPROVES quantile loss by [-0.0417, -0.0516, -0.0315] pts (2025, held out) and [-0.0487, -0.0652, -0.0345] (2024), and refitting the curve on the blended centre adds nothing ([0.0013, -0.0053, 0.0076]). What remains is the curve's own shape: on 2025 it covers 0.869 against 0.80, the lower tail at 0.052 strict / 0.373 weak (zero atom), the upper at 0.079 strict against 0.10 -- the ceiling sits slightly high, and that is the next question, not the centre.

Reasoning

Settles H-184's band clause before any prop is priced off the board centre (RANKED #40 asked for exactly that). Its ledger clause is stale at HEAD; its P(play) clause is H-180's.

Next action

(1) The ceiling's strict exceedance is 0.079/0.081 against 0.10 in two seasons, U and B centres alike -- a ceiling-only refit question for this lane. (2) Grade the served RB/WR/TE SIM band, which replaced SHIPPED_CURVE on 09-23, from a pre-kickoff filing once week 3+ settles: it has no history to walk forward. (3) Weeks 1-2 of the live ledger grade the retired constant band; any scorecard/AAR coverage number off them should say so.

Source provenance and publication scope

Owned research record: research/scientist/experiments/2026-09-24T0222Z-h184-band-centre.json

This public reading view includes authored question, finding, review, reasoning and next action fields. Raw measurements, commands, logs and local paths are withheld. The record ID clock is not proof of completion or deployment.