Even better would be subsidized prediction markets. That way people train themselves as necessary to get results.

There'd be some kinks to work out, since you don't always get the answer handed to you after all the bets are in, but I think this would be a solvable problem. You could find ways of having occasional payouts based on cases where there is slam dunk evidence which is withheld and the professional bettors don't know if their decision will affect the defendant's sentence or just their pay. You could also try rewarding consistency between separate prediction markets in the hope that "our actual best guess" is the most salient Schelling point.