Brier score, from 0 (all probability on the right answer) to 1 (all on a wrong one): squared error, scaled to 0–1 for any number of options.
The probability the model gave its own pick: how sure it was.
Picked the right option, side, or level.
Log loss of the right answer, capped so one sure miss can't dominate.
The probability the model gave the right answer.
The model declined to answer.
How one decision did. Probability measures are null when the model gave none.