First identify all checked errors and all checked flags. The fraction of errors caught uses all errors as its denominator. The fraction of flags that identify errors uses all flags as its denominator. Include missed errors in the first denominator and false alarms in the second. Compare the same checked batch. A larger catch fraction need not mean a larger correct fraction among flags. Unknown checks stay unresolved. Do not claim that one invented batch predicts future performance.
Worked example
In a separate checked batch, a rule flags 3 errors, misses 3 errors, and flags 2 correct records. It catches 3/6 of the errors. Its 5 flags contain 3 errors, so 3/5 of the flags identify errors. A fraction must name the group it describes.
Checked batch
An invented batch has 30 independently checked records: 6 errors and 24 correct records. Rule C flags 4 errors and 4 correct records. Rule D flags all 6 errors and 12 correct records. A different set of 4 flagged records lacks independent checks and is NOT included in this checked batch.