Original analysis · updated September 6, 2026 · by the ContractTax team
A prediction market only works if its prices mean what they say, a contract at 70¢ should win about 70% of the time. We tested that on Kalshi by matching our full harvested history of settled markets to the prices they traded at, then checking how often each price level actually resolved YES. Here is what 13,462 settled observations show.
Even a well-calibrated market usually shows a small, well-documented distortion: longshots are overbet and favorites are underbet. Our data shows only a faint version of this.
Kalshi is a different forecaster depending on what it is forecasting: a market full of economists pricing CPI is not the same crowd as a market full of fans pricing a game. So we publish a separate measured curve for each subject. If you trade one category, its own curve is the number that matters to you, not the blended one above.
We continuously harvest Kalshi markets and record their prices, then match each to its settled outcome. Markets are grouped into ten price buckets by their recorded pre-settlement price, and for each bucket we compute the share that resolved YES. The headline figure is the sample-weighted mean absolute difference between the actual YES rate and the price-implied rate, across buckets with at least 20 observations.
Caveats: the dataset grows daily, so figures update over time. Extreme-price buckets (0–9¢ and 90–99¢) can be affected by prices recorded very close to settlement, so we read them with more caution. Prices are our own scanner’s observations, not resold exchange data. See the live view in the Truth Machine, or grab the raw numbers from our open datasets. Not affiliated with Kalshi.