How our model probabilities held up against what actually happened. Every resolved prediction counts, the misses as much as the hits. How the Brier score works.
No predictions have resolved yet.
When we said a given probability, how often did it happen? A well-calibrated model has similar numbers in both columns.
| We said | Predictions | Average forecast | Happened |
|---|---|---|---|
| 0-20% | 0 | n/a | n/a |
| 20-40% | 0 | n/a | n/a |
| 40-60% | 0 | n/a | n/a |
| 60-80% | 0 | n/a | n/a |
| 80-100% | 0 | n/a | n/a |
Nothing has resolved yet. Results appear here as markets settle.
Updated nightly. Only markets where the model formed a confident view are graded; low-confidence estimates are not included. Each market is graded on the first confident forecast we recorded for it.