Recall
Measurement
The fraction of actual positive cases that a system correctly identifies.
Recall equals true positives divided by true positives plus false negatives. If positive means an annotation error, recall measures the share of real errors caught by review. High precision alone cannot show this. Estimating recall requires reference judgments that also cover items the system did not flag.
In practice
If a reference sample contains 50 errors and review finds 40, error detection recall is 80%.