Performance Methodology

Every number on our performance page is computed from the same source of truth — settled predictions — with no cherry-picking.

Definitions

Prediction — a single model selection (market, odds, model probability, confidence) logged before kickoff with a model version.
Settled — a prediction graded won or lost against the final result. Only settled predictions count.
Accuracy — settled won ÷ settled, shown with sample size. Accuracy is not profitability.
Model probability — the engine's estimated probability (confidence %).
Implied probability — 1 ÷ decimal odds, as a %.
Model edge — model probability − implied probability, in percentage points.
Calibration — do outcomes at each confidence band match the predicted rate?
Brier score — mean squared error of probability vs outcome (lower is better).
Log loss — penalises confident wrong predictions (lower is better).
Closing Line Value (CLV) — prediction odds vs closing odds; positive CLV signals long-term edge.
Strategy simulation — a historical, hypothetical flat-stake simulation, clearly labelled; past simulated performance does not guarantee future results.

Verifiable aggregates are available via our public performance API at /api/public/performance.