General knowledge QA

Results are only compared within a matching group, so a model tested one way is never ranked against a model tested another way.