Forecast calibration compares groups of predictions with similar assigned probabilities against their observed outcomes. If events given a 70 percent chance occur roughly 70 percent of the time across a suitable set, those forecasts are calibrated at that level.
Scenarios without explicit probabilities can still be reviewed for timing and quantity error, but that is not full probabilistic calibration. Good practice preserves original claims, scores resolution consistently, and separates directional accuracy from exact dates and magnitudes.
ELI5
Forecast calibration checks whether stated probabilities match what happens over many predictions. Events given a 70 percent chance should occur about 70 percent of the time in a suitable group of forecasts.
For example, if only half of many events labeled 80 percent likely actually happen, the forecasts were too confident. Good calibration needs preserved original predictions, consistent outcome rules and enough cases to make the comparison meaningful.
