A threshold is a choice made before reading the metric. Moving it changes the exact confusion counts, so the table recomputes each operating point instead of echoing labels.

highlighted = computed this step

Thresholds are choices

The threshold is stated before the counts are read. This toy table uses thresholds 5, 3, and 1. Lowering the threshold marks more rows positive, so the confusion counts change by exact integer recomputation from the same scored rows.

thresholds 5,3,1\text{thresholds }5,3,1
Operating pointsExact thresholds produce exact metric rows.discrete exact operating pointsthresholdTPFNFPTNprecisionrecall5120311/3321122/32/3130213/51discrete exact operating points only; no smooth curvethreshold fixed first; metric alone is not a claim; NOT training

Counts move exactly

At threshold 5, the counts are TP=1, FN=2, FP=0, TN=3. At threshold 1, they become TP=3, FN=0, FP=2, TN=1. Nothing is interpolated between rows; each operating point is its own exact recount.

TP,FN,FP,TN recompute at each thresholdTP,FN,FP,TN\text{ recompute at each threshold}
Operating pointsExact thresholds produce exact metric rows.discrete exact operating pointsthresholdTPFNFPTNprecisionrecall5120311/3321122/32/3130213/51discrete exact operating points only; no smooth curvethreshold fixed first; metric alone is not a claim; NOT training

Summary

The threshold is chosen, not fitted inside this evaluation table. The diagram shows discrete exact operating points only; it does not draw a smooth curve or imply unshown threshold values.

discrete operating points only\text{discrete operating points only}
Operating pointsExact thresholds produce exact metric rows.discrete exact operating pointsthresholdTPFNFPTNprecisionrecall5120311/3321122/32/3130213/51discrete exact operating points only; no smooth curvethreshold fixed first; metric alone is not a claim; NOT training