Assuming the documents are digitized, would probably be pretty easy to put together a neural network that pegs "interesting" documents for review. Not that it should be used as an authoritative judgement, but it sounds like so much stuff gets through the system just because there aren't enough eyes to check everything for the obvious markers.