Protocol fit
What should never become trusted state blindly?
Tell us where AI, agents, or automated systems produce meaning that needs evidence, authority, replay, and a clear commit boundary before anyone can rely on it.
Useful context
- What proposes the state today: model, agent, workflow, user, or outside system
- What evidence or source material should bind the proposed meaning
- Who has authority to approve, deny, hold, or escalate it
- What receipt, audit trail, or replay behavior would make it inspectable later