KUNKAFA FREE FORECAST TOOLS
Probability Calibration Calculator by Kunkafa
Compare forecast probabilities with recorded outcomes. Calculate a binary Brier score and inspect a reliability diagram using your own data. Everything runs locally in your browser.
Forecast reliability results
- Forecast pairs
- 10
- Binary Brier score
- 0.1450
| Probability range | Count | Mean forecast | Observed rate |
|---|---|---|---|
| 0–20% | 1Small sample | 10.0% | 0.0% |
| 20–40% | 2Small sample | 25.0% | 50.0% |
| 40–60% | 2Small sample | 45.0% | 50.0% |
| 60–80% | 2Small sample | 65.0% | 50.0% |
| 80–100% | 3Small sample | 90.0% | 100.0% |
Lower Brier scores mean lower average squared probability error on this dataset. A low score alone does not prove calibration, profitable trades or out-of-sample skill.
What does probability calibration mean?
If forecasts assigned about a 70% chance to an event, that event should occur about 70% of the time across many comparable forecasts. It does not mean every individual 70% forecast will succeed.
Each point groups forecasts into a fixed probability range and compares their mean probability with their observed event frequency. Points near the diagonal suggest agreement within those buckets. Empty buckets are omitted from the plot; small buckets can move substantially with one extra outcome.
How the binary Brier score is calculated
For each row, square the difference between the probability and the binary outcome, then average those squared differences. The binary score here ranges from 0 (perfect probabilities for these outcomes) to 1 (confidently wrong for every outcome). A 50% probability contributes 0.25 regardless of whether the event occurs.
Brier score measures more than calibration. Event frequency, discrimination and the dataset affect it. Compare models on the same events and dates, against an appropriate baseline, rather than comparing unrelated scores.
Use comparable, completed forecasts
Choose one precise event definition and duration. Record the probability before the outcome, include failures and wait until each outcome can be resolved. Overlapping market forecasts may not be independent. This tool does not validate your labels, remove selection bias or calculate confidence intervals.
Kunkafa publishes market forecasts as predicted changes, direction and probabilities. Use this calculator to understand probability evidence; use Kunkafa’s published forecasting results for the site’s own evidence and definitions.
Kunkafa forecasting methodology · Kunkafa risk/reward calculator