Enforcement Modes

Each policy scorer can operate in one of three enforcement modes. The mode determines what GOVERN does when a scorer exceeds its threshold.

Log

The inference continues normally. The violation is recorded in the audit trail and the assessment score is reduced. No alert is sent unless your alert configuration includes log-level violations.

Use Log for: new scorers you are calibrating, low-risk systems, or behavior you want to monitor before enforcing.

Flag

The inference continues normally. The violation is recorded and an alert is sent to your configured notification channels. The assessment is marked as flagged in the dashboard.

Use Flag for: violations that require human review but should not block operations, or medium-risk systems where you want visibility without friction.

Block

The inference is stopped. GOVERN returns an error to the caller before the inference reaches the upstream model. The violation is recorded and an alert is sent immediately.

Use Block for: critical violations (prompt injection, prohibited content), high-risk systems, or regulated use cases where a non-compliant inference must never reach end users.

Escalation

You can configure enforcement escalation: start in Log mode, promote to Flag after N violations in a rolling window, and promote to Block after M violations. This lets you ramp up enforcement without disrupting operations on day one.