Calibration
The crucial alignment process ensuring that a model's predicted confidence score perfectly matches its actual empirical probability of being exactly correct.
Think of It Like This
Like an honest weather forecaster whose '80% chance of rain' prediction mathematically translates to it actually raining on 8 out of 10 identical days.
If a highly calibrated model predicts a 90% chance of rain across ten different days, it should actually rain on exactly nine of those days. Modern deep neural networks, despite their high accuracy, are notoriously overconfident and require post-processing techniques like temperature scaling to output trustworthy probabilities for downstream decision-making.