The Council
The internal model carries the largest single vote, at thirty per cent. It ingests pitcher matchup, bullpen fatigue, park factor, weather, and lineup xwOBA. A causal expert engine, weighted around twenty-two per cent, applies pythagorean log-five with park and umpire adjustments. Market consensus, near eighteen per cent, listens to the de-vigged market. Several public model sources round out the rest, with weights tuned by historical calibration.
The Vote
Each engine produces a probability for the home side. A gradient-boosted meta-model takes those nine numbers plus context features — park, weather, umpire, day of week, pitcher rest — and outputs a final consensus probability, a confidence score, and a variance estimate.
The Tier Verdict
Four tiers emerge: V.STRONG (rare), STRONG, LEAN, and TOSS-UP. A call must clear three independent thresholds to be published: engine agreement, edge over consensus, and a minimum confidence floor. Anything below TOSS-UP is held back. Silence is also a verdict.
The Public Record
Every call written to the ledger is final — timestamped before first pitch, never deleted, never edited. Wins and losses are graded the following morning. The Track Record page is not a brochure; it is the brochure.