trust-threshold
How much can your model decide on its own? Paste its confidences and whether each answer was right. You get the threshold that meets your target accuracy, the share of traffic that clears it, and a split-half check of whether that number holds up.
Data
The example is synthetic: one calibrated model and one overconfident model.
%
Everything runs in this tab. Nothing is uploaded.
Paste data or load the example.