How the Models Work

Each sport uses a different statistical framework matched to its scoring characteristics:

MLBNegative Binomial

Run scoring has overdispersion (variance/mean ~2.1). The negative binomial distribution handles this by decoupling mean and variance via a dispersion parameter.

NHLPoisson

Goal scoring is well-modeled by independent Poisson processes. Each team's lambda is computed from 5v5 xG, goaltending, special teams, and situational factors.

EPLDixon-Coles

Modified Poisson framework with a tau correction for dependency in low-scoring outcomes. Adjusts draw probability by 2-3pp, the highest-Sharpe edge in football betting.

All models follow the same pipeline pattern: fetch data, build team/player ratings, apply contextual adjustments (home advantage, rest, weather, etc.), run the scoring engine to produce a full probability distribution, then derive fair odds for every market.

The parameter values on this page are served live from the model configuration and refresh periodically; when a weight or threshold changes, this page reflects it automatically.