Every governance evaluation produces one of three verdicts. Understanding what each one means — and what to do about it — is the foundation of working with Nomotic.
ALLOW
The action passed evaluation across all 20 governance dimensions. It is consistent with the agent’s established behavior, within its authorized scope, and does not raise concerns on any dimension. The action proceeds.
ALLOW is the most common verdict for well-configured agents operating normally. A healthy agent running within its intended purpose should see a high allow rate. If you are seeing lower allow rates than expected, check your policy configuration or review recent evaluations in the audit trail.
ESCALATE
The action requires a human decision before it can proceed. Nomotic is not confident enough to ALLOW it automatically, but the situation does not clearly warrant a DENY either.
When an ESCALATE verdict fires, the action is paused and a review request appears in your approval queue. A team member with the appropriate role can review the details of the evaluation and either approve the action (it then proceeds) or deny it (it is blocked). Either way, the decision is recorded.
Escalations are not failures. They are the governance system working as intended. An agent that never escalates may actually be a concern — it could mean governance thresholds are too permissive.
Common reasons an action escalates include an unusual action type the agent has not taken before, an action that scores low on a specific dimension like authority verification, or a situation where the agent is operating in a new context that governance has not seen from it before.
DENY
The action failed evaluation and is blocked. The agent is notified that the action was not approved, the action does not execute, and the denial is recorded in the audit trail.
Denials are most common when an agent attempts something outside its authorized scope, when a veto dimension triggers, or when behavioral drift has pushed the agent outside acceptable parameters.
A single deny is informational. A pattern of denies for the same action type is a signal to review your agent’s configuration, its archetype assignment, or the policies applied to it.
Reading evaluations in the audit trail
Every evaluation — regardless of verdict — appears in Governance → Audit Trail. Clicking any evaluation shows the full breakdown: each of the 20 dimension scores, the UCS score, which tier produced the verdict, and the reasoning artifact.
This is your evidence layer. Every decision is here, every time.
Adjusting thresholds
If you are seeing too many escalations, or your allow rate is lower than you expect, you can adjust governance policy weights and thresholds in Governance → Policies. Changes apply to new evaluations immediately. Use counterfactual replay in Intelligence to see what your historical evaluations would have produced under a new configuration before applying it to live traffic.