Evaluation desk · recorded evidence
Research workspace
Football model performance
Recorded probabilities, settled outcomes, and market snapshots—shown with the sample size needed to tell evidence from noise.
Evaluation window
180 days
Current production models only. Retired experiments excluded.
Graded picks
327
1X2 hit rate
59.6%
Brier score
0.522
Lower is better
ROI · selected sport
N/R
Sport-scoped ROI is not reported.
Edge stays a context signal until it is earned
Model-versus-market edge is informational until a sport has independently demonstrated calibrated probabilities. It is not presented as a recommendation before then.
Validation
No validated edge yet
Probability quality
Calibration curve
Probability bands compared with their observed outcomes. The diagonal is perfect calibration. Current ECE: 6.4pp.
Reading the record
Picks are snapshotted before kickoff and graded against final results. Calibration matters more than a hot streak; flat 1-unit ROI is a model-evaluation convention, not gambling guidance.