Eval sets grow, and growth is not free. Every case added changes the denominator, so a set that grows quietly makes every past score impossible to compare against.
About 1 min of reading
Stamp and bump a set
Stamp a new case
Before you run it, say out loud what it will print. Then open the answer.
Bump the set version in the name
Before you run it, say out loud what it will print. Then open the answer.
Mark a case retired
Before you run it, say out loud what it will print. Then open the answer.
The set changes at the same rate as the model. Both need a version, or the pair of them cannot be reproduced.
Cost is per case per run. A nightly run over a large set is the most expensive loop in the building.
Two cases added between runs, nobody noticed, and the score rose four points. The model did not improve.
The date added and who added it. A case with no owner is a case nobody can defend later.
A case that always passes is retired, not deleted. It stays in the file and out of the score, so the reason it was retired survives.
Each one is defined here on this site. Open the entry for the full definition and the other words it points at.