[waldrews]

Well, in this case we're literally asking if the model can remember new facts, not generalize, so seems like a legit first level test; second level might be, can it answer a question incorporating that specific knowledge in a broader question.