Accuracy
How the model's predictions have performed against final scores, across every published game.
Accuracy over time
What log loss means
Log loss grades the probabilities, not just the picks. Every prediction is scored on the probability placed on the team that actually won: a right call costs a little, a wrong one costs more, and a confidently wrong one costs the most. Lower is better: 0.693 is pure coin flips and the high .60s is where solid MLB models land. A useful model lives below the coin-flip line.
The home-team baseline
Accuracy by confidence
Confidence is the win probability the model gave the team it picked, whichever side that was. Accuracy is the share of those picks that were right, and games is how many graded games fell in that range.
By team
Tap or click a team below to see how the model has done, and how confident it's been, in that team's games.
Not available yet. Check back soon.