Forecast overview
The strongest candidate for every outcome, with its evidence attached.
Best models by outcome
Click any row for the full model recordRelative quality
Best finalist in each outcomeRecent projections
Candidate model explorer
Compare finalists, then inspect validation confidence, features, and weighting.
Candidate comparison
Lower error or log loss is betterGame predictions
Select a scheduled matchup, optionally add a custom line, and use the saved best model for each outcome.
Today's games
Schedule
Saved runs
Opening a result never retrains modelsFeedback and calibration
Your calls update a transparent calibration layer; completed source games still drive full model retraining.
Point-call accuracy by metric
Each game and participant counts once; ladder levels do not add observations.Threshold probability calibration
All configured +25, +50, and other probability markersRecorded outcomes
Data and models
Sync schedules and results, resolve completed calls, refit changed models, and refresh upcoming projections.
Update model lab
Completed seasons already on disk are reused. The active season is synced, completed calls are graded from real results, changed models refit, and saved upcoming calls are recalculated.