Every settled read, scored the way forecasters are scored — calibration and Brier, not highlight reels. Nothing is cherry-picked; this page aggregates the full settled history.
Predicted probability vs. how often the outcome actually happened, per 10% bucket. Points on the dashed line are perfectly calibrated; bubble size is sample count.
Brier score: 0 = perfect, lower is better. Δ = model − market; negative means the model was the closer forecaster.
| Sport | Reads | Market Brier | Model Brier | Δ |
|---|
How this is computed. One score per settled (match × market × outcome) read, with two-way markets deduplicated to a single canonical side so complements aren't double-counted. "Market" is the de-vigged bookmaker probability at the best available price; "Model" is the engine's own calibrated probability where it had a view. Both are scored against the actual result — the same data, the same matches, head to head.
We publish this because honest forecasting requires it: on most slices the market is at least as accurate as the model, and you can verify that here. A read is where the model diverges from the market — a model view, not a claim to beat the market. How market reads work →
For informational purposes only. Not financial or wagering advice. Statistical model outputs do not guarantee profit. Participate responsibly — only stake what you can afford to lose. BeGambleAware.org