Test set
A test set is past cases with your reviewers' decisions and reasons, rerun after every change to the harness to show whether the change helped.
Start it from one workflow's past work, failures included. After that, each failure also becomes a case in the test set.
It's the one record that tells you whether a change made the harness better or worse, so ask a vendor about the test set first. When a model upgrade makes some checks stale, the test set shows which ones to remove. It needs no engineers.
The test set is the past cases. An evaluation harness is the setup that runs tests and grades the results. Loop 6 uses one, and so does testing the checking agent on past cases.
Example
In a paid social example, the brief-writing skill had not yet told the agent to take every claim from the approved claims list. So one claim reached legal in the agent's own words. The team adds that step to the skill and the case to the test set, which is rerun after every change to the harness.