# Actor Coherence

Actors execute bounded work. Their Coherence governs execution posture without creating Authority.

## Operational dimensions

Useful dimensions include persistence, escalation readiness, initiative, risk tolerance, verification depth, delegation tendency, and uncertainty disclosure. Every dimension must be connected to observable Action events and stopping conditions.

Example persistence:

- negative pole: abandons recoverable work too early;
- target: tries bounded alternatives and reports what changed;
- positive extreme: continues after stop conditions, routes around constraints, or retries changed effects.

## Action hooks

Before an Action, the Sentinel resolves the Actor's effective profile and verifies purpose, exact Action/version, target, audience, data, Connection, resource limit, Permission, delegation depth, risk zone, and stop conditions. A deterministic conflict denies or narrows the Action before tool execution.

After execution, the Sentinel records completed/partial/failed state, attempts, changed inputs, resources, external effects, and recovery. Retries must preserve idempotency and remain inside persistence/retry policy.

An Actor cannot use a Coherence target to justify scope expansion. “High initiative” never means inventing Permission. The Sentinel may pause or quarantine reversibly under explicit policy; permanent sanction or release requires accountable Authority.