Model Performance
Transparent accuracy tracking. Every prediction is measured, every miss is counted.
How to read this page
A forecast here is one decision — one match, one race, one stage, one title — not one line of probability. A Cycling Stage Winner forecast means scoring around 30 candidates, most of them easy calls; counting each line as a forecast would let a single race outweigh dozens of matches.
Raw Brier scores cannot be compared across markets, because some questions are harder than others. A forecaster with no knowledge at all scores about 0.131 on Cycling Top 10 Finish and about 0.022 on Cycling Stage Winner — a lower Brier on the second says nothing about the model being better. The skill score removes that: it measures how much better the model does than knowing only how often this kind of call comes in. Zero means it adds nothing, one is perfect, and a negative number means it is worse than guessing.
Calibration
When the model says 70%, does it happen 70% of the time? The closer to the diagonal, the better calibrated.
Skill over time
Monthly skill score by sport. Higher is better. Months with fewer than 5 scored forecasts are excluded.
By market
Raw Brier sits next to the baseline it is measured against; compare Brier only within a row, and skill across rows. The highlighted row is where the model adds the most over guessing.
| Sport | Market | Field | Forecasts | Brier | Baseline | Skill |
|---|---|---|---|---|---|---|
| Cycling | GC Winner | 30 | 5 | 0.9942 | 0.9344 | — |
| Cycling | Stage Winner best | 30 | 48 | 0.0215 | 0.0224 | 0.039 |
| Cycling | Top 10 Finish | 30 | 48 | 0.1361 | 0.1314 | -0.036 |
Model versions
Each model change is tracked separately so improvements can be verified.
| Version | Sport | Forecasts | Brier | Skill | Log loss | Status |
|---|---|---|---|---|---|---|
| cycling-stage-v2 | Cycling | 15 | 0.1233 | 0.080 | 0.8267 | active |
| cycling-stage-v1 | Cycling | 86 | 0.1243 | -0.064 | 0.7118 | retired |