AI Operations · Glossary
Continuous Policy
A Continuous Policy is a Policy defined by cadence: it evaluates on every turn of an Interactive Agent's session, not only at Routine entry or at named checkpoints.
Continuous Policies are defined by cadence. The cadence is per-turn evaluation against the full state of the conversation, the agent memory, and the trace. Cadence is the defining axis of this class, independent of scope. A Continuous Policy can be scoped agent-wide (also a Persistent Policy) or scoped to a single Routine. The two axes, scope and cadence, are independent. Use cases are policy surfaces where a violation can emerge at any moment: responsible-gaming triggers, sentiment escalation thresholds, regulatory-disclosure obligations, fraud signals. A Continuous Policy that fires can pause the Routine, redirect to a different Routine, escalate to a human, or invoke a deterministic enforcement action. The class is named in the policy authoring surface and recorded per-turn in the production trace.