How the model works

The engine runs a 7-model ensemble per half-inning: Poisson (10.9%), ZIP (27.3%), Markov 24-state chain (43.6%), and MAPRE (9.1%) form the base layer, blended with three meta-models — Logistic Stack, NN Interaction, and Hierarchical Bayes (combined 9.1%). Dynamic Bayesian shrinkage, handedness-adjusted lineup splits, vector wind modeling, and monotonic spline calibration are applied before final blending.

P(NRFI) = clamp(0.76 × calibrated_ensemble + 0.24 × 0.614, 0.02, 0.98)

Value bets are identified when the model's implied probability exceeds the bookmaker's by ≥3%. Kelly Criterion (25% fractional) sizes the recommended wager.