Loading
Public and permanent. One page per resolved forecast. This is the reason to believe anything else on this site.
Public data runs 30 days behind. A call is published before its outcome is known, evaluated automatically at the end of its stated window, and never edited afterwards.
0 graded, 43 topics still being watched without a call.
No resolved calls to count yet.
Lower is better. Published once calls have resolved.
A published call stays, whatever happened to it.
Forecasts grouped by the probability we published, against how often they actually resolved. Close to equal is calibrated. Claimed above observed means we are overconfident, and the fix is recalibration, not rewording.
Dated when made, evaluated when due, kept either way.
A forecast needs enough history to publish and a closed window to evaluate. Until both have happened there is no accuracy figure to claim, and an empty table is the honest state of it.
See what is being watchedThe threshold is declared when the forecast is made. At the end of the window the call is graded automatically, against registration volume above a rolling baseline. No one can move a marginal call into the hit column.