Use independently checked records, not flags alone. To find the fraction of errors caught, divide flagged errors by ALL checked errors, including missed ones. To find the fraction of flags that identify errors, divide flagged errors by ALL checked flags, including false alarms. These fractions answer different questions. State which group is counted. Missing checks cannot be labeled correct or wrong. These invented checked batches do not establish performance on future records.
Worked example
In a separate checked batch, there are 5 actual errors. A rule flags 4 of them and also flags 6 correct records. It catches 4/5 of the errors. Among its 10 flags, 4/10 identify errors. Both fractions use 4 on top, but the denominators are different.
Checked batch
In an invented batch of 20 independently checked records, there are 4 errors and 16 correct records. Rule A flags 3 errors and 5 correct records. Rule B flags 2 errors and 2 correct records. Neither rule changes the records. Separately, 3 more flagged records have no independent checks; they are NOT included in the 20 checked records.