Auto-Execute
The action matches your policy and the state checks out. It proceeds, and the verdict is recorded with what it was checked against.
Platform
Verdicts and audit trail
Observe
The proposed action is permitted to proceed but is flagged for monitoring; the action executes while a structured observation is logged for subsequent review.
Escalate
The action needs a human: it crosses a value threshold, hits a risk limit, or lacks a sign-off. It holds and routes to the person who owns that call, with the full context attached.
Deny
The action breaks policy, or the state it depends on is not what the agent assumed. It stops before it executes, and the agent receives the reason.
Auto-Execute
The action matches your policy and the state checks out. It proceeds, and the verdict is recorded with what it was checked against.
Observe
The proposed action is permitted to proceed but is flagged for monitoring; the action executes while a structured observation is logged for subsequent review.
Escalate
The action needs a human: it crosses a value threshold, hits a risk limit, or lacks a sign-off. It holds and routes to the person who owns that call, with the full context attached.
Deny
The action breaks policy, or the state it depends on is not what the agent assumed. It stops before it executes, and the agent receives the reason.
The record is the point for audit.
A blocked action with the state attached shows what was true when the decision was made, not what the record looked like afterwards. Because the enforcement layer sits outside the agent and is owned by a different team, the audit trail answers who granted the permission separately from who exercised it.
