Architecture of bounded authority

Identity identifies; it does not authorize

An agent identity tells a relying party which software participant is involved. It does not, by itself, establish the economic principal behind the action or permission to spend, disclose, modify or accept anything.

The proposed architecture therefore keeps AgentIdentity, Principal and Mandate as distinct entities.

The mission is the unit of delegation

A mission mandate describes the permitted scope of a specific task. It should define relevant actions, participants, limits, deadlines, revocation conditions and any required human decision.

A supplier profile may be reusable. A previous buyer’s permission is not. Each principal must make an independent authorization decision for its own mission.

Execution and acceptance are different events

A technically successful action may still fail the buyer’s agreed commercial criteria. Conversely, an acceptance decision may require human judgment that cannot be reduced to an execution log.

APA separates the execution record, the permitted evidence, verification against criteria and the buyer’s acceptance or dispute. That separation creates a more explicit account of responsibility.

The enforcement boundary

A record of a mandate is only part of a control system. External agents must use compatible execution controls or binding integrations for action boundaries to be enforced.

The planned prototype must test revocation, expiry, tenant isolation, event replay, evidence integrity and recovery from partial failure. No claim of enforced control over arbitrary agents follows from this architecture.

Source and publication status