Grade by hand

Ten confidences, one reliability curve

Same 10 stated confidences (0.6 to 1.0) as the worked example. Click a card to flip that answer between correct and wrong, or load one of the two presets.

Click a card to flip correct/wrong. Predictions group into three bins, [0.6, 0.8), [0.8, 1.0), [1.0], and the table below recomputes live.
binnmean confaccuracygap
Mean confidence
0.00
Accuracy
0.00
ECE
0.000
Brier
0.000

Why accuracy alone lies

A model that is always right, but never sure

Ten predictions, every single one correct. Slide the stated confidence and watch accuracy stay perfect while ECE and Brier expose the mismatch.

Accuracy
1.00
always correct
ECE
0.000
gap to perfect
Brier
0.000
squared error