Track Record
A permanent, growing archive of how real prediction markets resolved — scored against the price each market was genuinely showing before it closed, not the settled price that already knew the answer. Neither Polymarket nor Kalshi keeps any of this once a market closes.
Forecast Accuracy — Scored Before the Market Closed
Every figure here is scored against the price the market was genuinely showing a set time before it resolved — recovered from Polymarket's own order-book price history, not from the settled closing price. A settled market already knows the answer, so scoring it measures nothing.
The accuracy archive couldn't be reached just now. Nothing is estimated in its place — reload to try again.
Recently Resolved
0 archivedNo resolved markets archived yet — check back soon.
Frequently Asked Questions
Where does this data come from?
Real, closed Polymarket events plus targeted Kalshi series — each market's actual final outcome and the real closing price before it settled, archived permanently once a day. Nothing here is estimated or backfilled.
Why score prices from before the market closed?
Because a settled market already knows the answer. Polymarket's closing price for a resolved market collapses to 0 or 100 as the result becomes known, so scoring it measures settlement, not prediction — it makes any market look near-omniscient. Every figure above is instead scored against the real price the market was showing 1 hour, 1 day and 1 week before it closed, recovered from Polymarket's own order-book price history.
What's a Brier score, and what's the skill score?
The Brier score is the average squared difference between the real probability and the real 0/1 outcome: 0 means every call was perfect, 1 maximally wrong. On its own it's hard to read, because an easy sample scores well no matter who forecasts it. The skill score fixes that by comparing the market against a forecaster that knows only the base rate and repeats it every time — above 0% means the market genuinely added information, 0% means it added none.
What is discrimination (AUC), and why can it disagree with the Brier score?
AUC asks only whether the market ranked winners above losers — the chance a randomly picked winning market was priced higher than a randomly picked losing one. A market can rank perfectly (AUC 1.0) while being badly calibrated, if it says 30% for things that always happen. Calibration and ranking are genuinely different skills, so both are reported rather than folded into one headline number.