The plain-English version
How closely predicted probabilities correspond to observed frequencies over comparable cases.
Why it matters
It matters when a product uses confidence to decide whether to act or defer.
Let’s make it click
Among cases assigned an 80% probability of the positive class, about 80% should actually belong to that class on representative data.
Vibe Bro’s heads-up
A confident-looking number is not automatically a calibrated probability, especially when it is generated as ordinary text.
The docs behind it
Documentation links checked .
Editorial review · October 28, 2026
Corrected: Calibration concerns observed class frequency at a predicted probability; an unspecified fraction of predictions being correct can instead describe accuracy and depends on the decision threshold.