Each engine reads the world through a different lens. A gradient-boosted meta-model weighs them by context and outputs a single probability, a confidence, and a variance.
Our in-house gradient model. Builds a per-pitcher xERA, blends bullpen fatigue, applies a park-specific run factor, and overlays the weather call — wind, temperature, humidity.
Log-5 of pythagorean win expectancies with a 3.5% home advantage, then signed adjustments for starter ERA gap, park-run inflation, and umpire historical favor.
Median of major sportsbooks, stripped of vig. Treats the market not as a source of truth but as a mirror of public conviction — valuable noise.
Daily-updated projections from the most respected public sabermetric system. Strong on long-run player performance.
ESPN’s Football Power Index family adapted to MLB. Conservative, slow-moving, useful as a baseline check.
Ten-thousand-trial Monte Carlo simulations per game with lineups, pitchers, and bullpen states.
Aggregate of SportsLine’s in-house quant projections. Tracks closer to the market than the rest.
Continuous Elo with starting-pitcher adjustments. Pure team-strength signal.
When public ticket count diverges from money handle, the market is telling us where sharp opinion sits. We listen, but quietly.
The nine engine probabilities are passed, alongside game context, into a gradient-boosted meta-model. It outputs a calibrated probability, a confidence score, and a variance estimate. Retrained quarterly on three full MLB seasons.