Methodology
Scoring version score-v1. Updated November 1, 2026.
What a call means
A logged position implies a probability for the outcome you took. Buying Yes at a price of 0.62 implies at least 62% for Yes; buying No at 0.62 implies at most 38% for Yes. You may also state your own probability when you log the call. When you do, the stated number is what we score for Brier and calibration; the price is still used for P&L.
Brier score
For a binary market the Brier score is (p − o)², where p is your probability for Yes and o is 1 if Yes happened and 0 if not. It runs from 0 (perfect) to 1 (certain and wrong). We report the mean over your resolved calls. Categorical markets are scored as the sum of squared errors across outcomes divided by two, so the scale matches. A coin-flip forecaster scores 0.25.
Time-averaged Brier
If you add to, reduce or restate a call before the market closes, each probability counts for the time it stood. This stops a last-minute change from erasing an early bad call. The simple "final call" Brier is shown too.
Skill against the market
We score the market price at the same moments your calls stood and compute 1 − (your Brier ÷ market Brier). Above zero means your calls carried information the price did not yet reflect. Below zero means you would have done better by taking the market's own probability.
Calibration
Resolved calls are grouped into ten probability bins. For each bin we plot the average stated probability against the share that actually happened. The expected calibration error is the weighted average gap between the two. Well-calibrated 70% calls happen about 70% of the time.
P&L and ROI
Each contract pays 1 if your side happens and 0 if not. A Yes bought at 0.62 returns +0.38 per contract when it happens and −0.62 when it does not; a No bought at 0.62 (that is, Yes sold) mirrors this. Fees you enter are subtracted. ROI is total P&L divided by total amount risked. Voided markets refund the stake and are excluded from every score.
Ranking and minimums
A profile is ranked only with at least 10 resolved calls in the period. Below that it shows "unranked" with its numbers. Periods are 30 days, 90 days, 365 days and all time. The primary ranking key is skill against the market; ties go to the profile with more resolved calls.
Resolutions
Resolutions follow the source exchange. If an exchange later corrects a resolution, we re-score. Markets the exchange voids are removed from scores. Where you log a result before the exchange publishes it, the call is marked "claimed" and does not count until confirmed by public data.
What we do not do
We do not infer positions from your exchange account or wallet; you log every call yourself. We do not score open calls. We do not show open calls on public profiles. We do not mix play-money markets into money P&L.
Changelog
- score-v1 (November 1, 2026): initial method as described above.