Grade by hand
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.
[0.6, 0.8), [0.8, 1.0), [1.0], and the table below recomputes live.
| bin | n | mean conf | accuracy | gap |
|---|
Why accuracy alone lies
Ten predictions, every single one correct. Slide the stated confidence and watch accuracy stay perfect while ECE and Brier expose the mismatch.