The number that improves may be measuring the part someone learned to display.

People get better at being measured before they get better at the thing. When a number is watched, attention moves to the surface the number touches: the tidy report, the responses inside the window, the parts of the work that show. Improvement in that surface is real improvement in the surface and tells you little about the rest.

Given a figure that has improved, ask what someone would do differently to make it rise without the underlying thing changing. Then go and look at the underlying thing directly, in one case, in detail. Keep at least one measure that nobody is scored on.