Skip to content
Doefin

Calibration

What is calibration?

Whether a forecaster's stated uncertainty matches outcomes: whether its 95% intervals contain the outcome about 95% of the time across many forecasts.

A forecaster can be accurate on average and still badly calibrated, by quoting intervals that are far too narrow. Its 95% intervals then contain the outcome much less often than 95% of the time.

Quick answers

What is the difference between accuracy and calibration?
Accuracy is about the central number: how close it is to the outcome. Calibration is about the stated uncertainty: whether a 95% interval contains the outcome about 95% of the time.
How is calibration assessed?
Only over a run of forecasts, never from a single one. That is why a dated public archive of forecasts matters more than any individual prediction.

Last reviewed 2026-08-20

Last reviewed . Figures marked illustrative are placeholders, not live data.