I need help to find a non-parametric accuracy test for the model with multiple prediction outputs

#1
Hello!
Thanks for taking time to read my post
I have a model that gives multiple predictions, and I need help to find a non-parametric accuracy test for this case. All predictions can be ranked (they have a stability coefficient). Ideally, the test or precedures hould penalize if not the most stable forecast turns out to be correct. Is there a way to introduce some sort of a scoring element?
I will be very gratfeul for any hints