Start here · Lesson 5 of 10
What calibration proves
Calibration is the test of whether our probabilities mean what they say. Take every time we published roughly a six-in-ten chance of finishing up. If reality is honest, about six in ten of those actually finished up. Plot that for every probability level and you get a calibration diagram.
This is the number that matters, and it is why we score every forecast automatically against what actually happened. A tool that never checks itself is asking for trust it has not earned.
See the live record on the calibration page — and note that early on the ranges around it are wide, because few forecasts have resolved yet. We publish it anyway.
See also: calibrationWhether our probabilities match reality — of all the times we said an outcome was, say, three-in-five likely, did about three in five actually happen?, brier scoreA single score for probability accuracy — lower is better; it rewards being both confident and right.