Governance is the core of Nomotic — the runtime layer that evaluates every agent action before it executes. This article explains how governance works, how policies control it, and where to find your evaluation history.
How governance evaluation works
When your agent calls client.evaluate(), Nomotic receives the action details and runs them through a three-tier evaluation cascade:
The first tier handles clear-cut cases fast — actions that obviously comply or obviously violate hard boundaries. Most actions resolve here in microseconds.
The second tier applies weighted scoring across all 20 governance dimensions for actions that need more analysis. Each dimension scores the action from a different angle. The scores are combined into a UCS.
The third tier handles genuinely ambiguous cases requiring deeper analysis. Most actions never reach this tier.
The result is a verdict — ALLOW, ESCALATE, or DENY — delivered in under 3 milliseconds on average.
The 20 governance dimensions
Every action is scored across dimensions covering scope compliance, authority verification, resource boundaries, behavioral consistency, cascading impact, stakeholder impact, incident detection, isolation integrity, temporal compliance, precedent alignment, transparency requirements, human override, ethical alignment, jurisdictional compliance, and six advanced dimensions in the extended evaluation layer.
Several dimensions carry veto authority — they can block an action regardless of how well it scores elsewhere. A scope violation cannot be overridden by strong ethical alignment scores.
Policies
Policies are the configuration layer that controls how governance behaves for your agents. In Governance → Policies, you can adjust:
Dimension weights — how much each dimension influences the final UCS score.
Escalation thresholds — the UCS range that triggers an escalation rather than an automatic ALLOW or DENY.
Veto rules — which dimensions carry hard block authority for your use case.
Approval routing — which team members receive escalation notifications and are authorized to review them.
Policies can be applied per agent, per archetype, or organization-wide. Agent-level policies override archetype policies, which override org-wide policies.
The audit trail
Every evaluation is sealed in the audit trail immediately after it completes. The trail is hash-chained using SHA-256 — each record contains the hash of the previous record, making any tampering mathematically detectable.
From Governance → Audit Trail, you can search evaluations by agent, verdict type, date range, and dimension. Clicking any evaluation shows the full breakdown including all dimension scores, the UCS, the reasoning artifact, and the evaluation latency.