Negative scores Positive scores Threshold t (draggable)

Confusion Matrix Threshold Lab

A classifier's accuracy claim is only ever true at one specific threshold. This simulator makes that threshold something you can grab: two overlapping synthetic score distributions (one per class) sit above a real 2x2 confusion matrix that recounts every sample against the threshold on every move — no cached curve, no lookup table, just true positives, false positives, false negatives and true negatives tallied from the actual data each time the line moves. Precision, recall, F1 and accuracy are then read straight off those four counts, so you can watch exactly how tightening or loosening the threshold trades one error type for the other, and where accuracy quietly stops telling the whole story.