something like correctness^2 x novel information content rank?

Actually now thinking about it, incorrect information has negative value so the metric should probably reflect that.