Experts in SPR, experts in the field in question, or only experts in both?

Experts in the field in question. But even then, selective defection is probably a bad idea unless there is an obvious countervailing factor, like a prisoner recently became a quadripelegic and hence is very unlikely to commit another violent crime no matter what an SPR says. If experts are inclined to tweak the results based on theoretical knowledge, that knowledge should probably be incorporated directly into the model.

I added a citation on this point.