Do the prices come true?
Every resolved contract, grouped by the price actually paid for it and compared against how often it really settled yes.
Actual vs price paid
Calibration error by price
About this page — read before quoting any number
Contract-weighted throughout, and that is not the file's default. The published implied_prob column is trade-weighted while the win rate beside it is contract-weighted; differencing the two compares different populations. Everything here uses the contract-weighted pair, derived from calib_error_qty, so the price on the x-axis and the win rate on the y-axis describe the same contracts.
Why the election sits in its own bucket. The 2024 election is a single correlated event carrying an enormous share of all volume ever traded here. Pooled into an all-contracts curve it dominates the result, so the default view excludes it and it can be inspected on its own.
One settlement is one observation. Intervals are clustered so that many contracts resolving on the same underlying outcome count once, not once per print. The effective cluster count — which additionally accounts for how concentrated volume is across them — is what sizes each dot.
Whose price this is. ForecastEx publishes no aggressor flag, so this is a leg-price curve rather than a taker curve and should not be differenced against Kalshi's taker-side series.