AI Performance
Our accuracy is computed from real results after matches end, fully transparent.
Accuracy curve
Rolling hit rate (10-pick window) across the latest settled results — each point answers: of the last ten picks up to that moment, how many hit?
How the model works
From raw match signals to a calibrated probability, scroll through the pipeline stages and watch the data flow between stations.
Match results, goals for/against, home & away records and head-to-head, every signal enters the line raw.
1Recency weighting: newer games count more. Attack/defence rates are derived relative to the league's own scoring environment.
2A full Poisson matrix over every possible scoreline, with the Dixon-Coles low-score correction (0-0, 1-1), the professional standard.
3A strength index weighted by form, home edge and ranking, it reads what goals alone can't see.
4The two models are blended, then the matrix is re-balanced onto the final probabilities, every market comes from one consistent distribution.
5Confidence is capped by the evidence, and only the elite get published: a pick below the quality bar stays a draft.
6