Start with the denominator
Read the product, date range and number of settled selections beside a record. A small sample can move sharply, and a high win rate at short odds does not establish positive value.
The record page shows the results available from the current source. Missing measurements must not be read as zero or as evidence that a model has passed evaluation.
Probability and calibration
Calibration asks whether events assigned similar probabilities happen at a similar rate over a sufficiently large sample. Brier score, ranked probability score and log loss assess different aspects of a probabilistic forecast.
Those measures are useful only with the dataset, timing, exclusions, baselines and uncertainty made explicit. This page does not claim a current held-out accuracy target has been achieved.
Prices and closing lines
A closing-line comparison needs independently captured, correctly matched prices with valid timestamps and provider identity. A later capture time alone does not establish that the underlying quote changed.
Positive historical results or a single closing-line figure cannot guarantee future returns. Read each measurement with its eligibility rules and limitations.