EVALUATION / PRE-LAUNCH
A track record starts with evidence.
No audited predictions have been evaluated. Historical model scores will appear only after a reproducible forward evaluation.
What a published evaluation must contain
- Predictions archived before kickoff, with model and input-data versions.
- A declared forecast horizon and full-time outcome definition.
- Verified results, a fixed evaluation cohort and a visible match count.
- Brier score, log loss, accuracy and calibration diagnostics.
- Comparison against baselines on the same fixtures and time horizon.
- Explicit treatment of cancellations, missing data and model changes.
Why this page has no simulated scores
Interface demonstrations cannot establish predictive performance. Illustrative probabilities are excluded from future production evaluation.