About¶

Sports Models KB is a compiled knowledge base for sports betting model development, covering the statistical methods and data sources that practitioners use to build, evaluate, and deploy predictive sports models.

The KB is organized around the core modeling techniques that power most sports prediction systems: Poisson distributions for goal-scoring models, Elo and Glicko rating systems for team strength estimation, the Dixon–Coles model for football, and Bayesian inference for updating beliefs with new evidence. It also covers the Kelly criterion and its half-Kelly variant for bankroll management, and calibration techniques like reliability diagrams to check whether model probabilities are trustworthy.

On the data side, the KB documents major sources including API-Football for competition and fixture data, Football-Data.co.uk for historical odds and results, StatsBomb for event-level xG data, Pinnacle for sharp closing lines, and The Odds API and OddsJam for odds scraping. Understanding which sources are "sharp" (Pinnacle) versus "public" (most bookmakers) is essential for building models that can actually identify edge.

The validation section emphasizes walk-forward backtesting, avoiding overfitting, and using closing line value (CLV) as the ultimate test of whether a model's predictions add genuine value beyond the market's own estimates.