The useful question is not whether the model was confident. It is whether this agent may perform this action, on behalf of this principal, under the authority that applies right now.
When that authority is sufficient, the action may proceed. When it is not, a named person can authorize it. The reason is written. So is a refusal.
The systems you already run stay. Nexus sits on the authority decision before the action executes.
