Accuracy
Every game locked in the Oracle writes down a probability before anyone plays, and that line is never touched again. This page holds them against what actually happened.
Hit rate
52%
Predictions scored: 29
How often the favourite won. On its own it misleads: an Oracle doing its job pushes games towards 50/50, and the hit rate falls as the site gets better.
Mean error
0.265
Above the line — worse than always saying 50/50.
Mean squared error of the probabilities (Brier). Zero is perfect; 0.250 is what you get by always saying 50/50. Above that line, predicting is doing worse than not predicting.
Coin flips called
0
Predictions of exactly 50/50. They claim nothing, so they do not count towards the hit rate — and in an Oracle that works, this number grows.
Calibration curve
Each dot is a group of predictions. If the system is calibrated, the dots land on the diagonal.
| Band | Predictions scored | Predicted | Actual |
|---|---|---|---|
| 50–60% | 17 | 54% | 53% |
| 60–70% | 8 | 63% | 50% |
| 70–80% | 3 | 75% | 33% |
| 80–90% | 1 | 88% | 100% |
| 90–100% | 0 | — | no games |
What the handicap did
Two comparisons over the same uneven-team games. Less error is better.
The proposed handicap, against no handicap at all
No difference worth reading.
10 games
The proposed handicap, against what actually reached the table
The handicap played has never differed from the one proposed. With no difference there is nothing to compare.
What manual adjustments did
An adjustment is an admin saying the number is wrong. The results either bear that out or they do not — and this is what says which.
No prediction has been moved by a manual adjustment. Without that there is nothing to compare.
0 adjustments marked active.