ReadonlyafterReadonlybeforeReadonlybetterReadonlydiffafter - before, the mean over rows.
ReadonlyhighReadonlylowIts 95% interval (a paired t interval); null with one row. When it includes 0, the change could be luck.
ReadonlymetricReadonlynReadonlysameReadonlyworse
One metric both evaluations have: the means, the paired difference with its 95% interval, and how many rows changed.