But I do wonder: If NSA does not know what it is doing because it is so good at collecting intelligence, is it possible that the entire SIGINT architecture is somehow emergent in the sense that it is too complex to be subject to the type of oversight that a reasonable person who does not understand SIGINT would think is appropriate?

This isn’t a philosophical question. If the NSA doesn’t know what it’s doing, or if it is not possible for NSA to know what it is doing, then should it design systems that err on the side of over-compliance? In other words, should the “bias” of its policies be to proactively screen out information even if it might include valuable foreign intelligence because NSA is probably overcollecting? Or are the American people willing to allow a certain amount of technological pipe shifting, even if it includes their communication, so long as the NSA is legitimately trying to make it work?

How eager is the National Security Agency to share its compliance problems with Congress? And how much information SHOULD they share with members of Congress who aren’t members of the intelligence committees?