Our number, their price, on every market worth trading.
Two products already price Bitcoin. This one prices everything else Polymarket runs — politics, macro, geopolitics, sports, culture — and states the gap between our forecast and the market in percentage points.
Most of the board agrees with the market.
Absolute gap between our published forecast and the de-vigged mid, across every quotable market on the board right now. It is the honest shape: a deep book on a well-traded question is hard to disagree with, and a forecaster that disagreed everywhere would be broken rather than clever.
- median gap
- 0.0pp
- 90th percentile
- 1.0pp
- widest
- 19.6pp
154 of 171 sit inside a point of the price. The board is a ranked list of the ones that do not.
Three layers, and each row says which produced it.
The price, undistorted.
De-vig against the live book, coherence across the legs of a mutually exclusive event, then a recalibration curve — applied only where one beat the market out of sample on a temporal split, inside the band a forecast can actually be traded in.
What needs no model.
A first-passage probability from the same bootstrap the Levels product sells. A station's climatology over fifteen years, and the forecast where the date is in range. The meeting calendar, read from the source. The settlement page the criteria name.
Blind to the price.
A researcher with web access, given the evidence and the archetype's playbook, and never given the market price — a model shown the price returns the price. It must report the figures it used and where they came from, and it is blended at its own stated confidence.
Where the depth is.
The deepest markets in each category, re-ranked on every sweep. A market with no two-sided book gets no forecast at all — an edge nobody can fill is not an edge.
Bitcoin is deliberately absent. It already has two products of its own, fitted to years of its own data, and one forecast sold twice is one forecast too many.
No edge is demonstrated yet, and this page will say so until one is.
Every forecast is appended to a ledger the moment it is published, with the price it was made against and the time — before the outcome is known. Scoring joins that ledger to settlements afterwards and reports both sides on the same rows, so the comparison cannot be won by choosing which markets to count.
131 markets have settled, carrying 1975 published forecasts. Paired against the market on the same rows and resampled by market, the difference is +0.0019 nats, CI [-0.0028, 0.0094].
Pooled across 7,409 observations, a fitted curve scores 0.070 nats worse out of sample than simply believing the price. Two buckets of forty-five earned the right to be applied; the rest are recorded and ignored.
That is the same answer this project already published for Bitcoin. A liquid prediction market is well calibrated, and a product that pretended otherwise would be selling the pretence.