A trial summary can look better when two awkward files disappear from the sentence. Imagine a team testing an assistant on twenty incoming invoices. By the end of the rehearsal, fourteen need no changes, four have been corrected and two are still waiting for an answer.
Someone writes “fourteen out of eighteen” on the slide. The calculation isn’t wrong. The label may be doing too much hiding.
Which twenty did you begin with?
In this fictional exercise, fourteen of the eighteen finished files needed no changes. That is about 77.8% of completed files. Fourteen of all twenty submitted files needed no changes: 70% of the original trial.
The two percentages describe different sets. Calling the first simply “success rate” makes it easy to forget that the unfinished work has left the denominator. We would keep all twenty in the main record, then explain the completed subset separately.
The other four completed files matter too. They reached a usable result after correction. They belong in a correction count, with the work that was needed to finish them. An operator cannot plan the next queue from a percentage that makes review effort vanish.
Put the definition on the row
A useful trial record could say:
Input set: twenty fictional invoice files
No-change results: fourteen of twenty submitted, 70%
Corrected results: four of twenty submitted, 20%
Still unresolved: two of twenty submitted, 10%
Here, the three outcome counts add back to the original twenty. The labels describe this exercise; they do not establish how another set of invoices or a deployed system would perform.
Before calculating, agree what counts as “no change” and who checks it. A correct supplier name isn’t a completed invoice if the total still needs attention. Keep the individual file references behind the count so a reviewer can see which job moved categories and why.
This connects to the correction-time rehearsal: a file that needed work should retain that work in the record.
https://www.rebuildshow.com/p/count-the-correction-time
For your next small trial, write the input count before reviewing the outputs. When you report a percentage, put its numerator, denominator and outcome definition on the same line. Check that every original file still has a place.
The Rebuild Editorial


