There should be some sort of "taint" checking that will indicate how many retracted papers any given one is based on. It shouldn't be too difficult since publications are highly structured.

With LLMs we can read all the papers at once and flag the most strongly derived for further review.