13,575 completed games · 2015-10-07 through 2026-10-28 · Updated 2026-10-29 UTC
End-of-period win probabilities. Both Raw and Model adjust for rest days. Raw = historical counting data (rest_30s lookup). Model = XGBoost trained on same inputs.
ℹ️ Rest days → model: 0 days = B2B, 1 day = one-day rest, 2+ days = rested. Model uses XGBoost (non-linear, captures B2B jump vs diminishing returns).
| Score Diff (Home) | Raw % (Home) | Model % (Home) | Away Raw % | Visual (Raw) |
|---|
Raw: historical win frequency, at least 30 matching games; broader samples used when needed.
Model: daily XGBoost fit on one state per game per 30-second interval; chronological validation before publication.
Both columns reflect rest-day adjustments. Rest=0 = B2B (played yesterday or same day).