Tools › Brier score calculator
Scores your forecasts in your browser; nothing is sent anywhere · Kalshi benchmark from the accuracy study, re-collected weekly
The Brier score is the average of (forecast minus outcome) squared, with the outcome counted as 1 if it happened and 0 if not: 0 is perfect, always saying 50% scores 0.250, and lower is better. Paste your forecasts below, one per line, to get your score, your calibration by 10-point band and a chart. For comparison, Kalshi's own prices a week before close scored 0.223 on markets priced between 20% and 80% (0.109 on all 1,766) in Kalshi View's accuracy study.
Forecasts read: 10
Your Brier score: 0.179
Always saying 50%: 0.250; always saying the base rate: -
Improvement on always 50%: -
Lower is better. Lines that cannot be read are skipped.
| Forecast band | Forecasts | Average forecast | Happened |
|---|
Each dot is a band, at its average forecast and how often it happened; dots on the dashed line are perfectly calibrated. Bigger dots hold more forecasts.
In Kalshi View's accuracy study (1,766 settled markets, collected September 25, 2026), Kalshi's price a week before close scored 0.109 overall and 0.223 on the 503 markets priced between 20% and 80%. The overall number looks better than the second because most markets were long shots that did not happen, which are easy to score well on; compare your own forecasts with the 20% to 80% column unless yours are mostly long shots too.
| Price taken | Markets | Brier score | Brier, priced 20% to 80% (markets) |
|---|---|---|---|
| 30 days before close | 1,094 | 0.120 | 0.229 (358) |
| 7 days before close | 1,094 | 0.088 | 0.211 (290) |
| 1 day before close | 1,094 | 0.063 | 0.206 (240) |
Same markets at all three points in time, so the rows compare like with like. Data: accuracy.csv (CC BY 4.0).
Brier score = the average over all forecasts of (p minus o) squared, where p is your probability and o is 1 if the event happened and 0 if it did not. A forecast of 70% on something that happened adds (0.7 minus 1) squared = 0.09; the same forecast on something that did not happen adds 0.49. Always saying 50% adds 0.25 every time, which is the usual baseline. Always saying the base rate (the share of your questions that came true) is a tougher baseline when most questions go one way.
To score a strategy of trading at Kalshi prices, download the study data (accuracy.csv) and paste the columns price_7d_before_close and resolved_yes, or use your own trade log.
How accurate is Kalshi? · payout calculator · fee calculator · all tools
Below 0.250 beats always saying 50%. What counts as good depends on how hard the questions are: forecasts full of near-certain calls score low easily, so compare scores on similar questions. Kalshi's own prices a week before close scored 0.223 on markets priced between 20% and 80% and 0.109 overall.
For each forecast, subtract the outcome (1 or 0) from the probability, square it, then average over all forecasts. The 10 example forecasts on this page score 0.179.
The Brier score is one number for overall accuracy. Calibration asks whether things you called 70% happened about 70% of the time; the table and chart show it band by band. A forecaster can be well calibrated and still score worse than one who commits to sharper probabilities.
Yes. Numbers above 1 are read as percent or cents, so 62, 62% and 0.62 are the same forecast. The outcome can be yes/no, 1/0 or won/lost.