AI Operations · Glossary
Top-K Policy Injection
Top-K Policy Injection is the mechanism by which the most relevant Policies for the decision at hand are surfaced into the decision context assembled by Dynamic Context Engineering.
A mature policy library carries far more Policies than fit into a single decision's context. Top-K Policy Injection selects the K Policies most relevant to the current decision and injects them into the context the model and the enforcement layer operate against. Relevance is computed against the Routine step, the Variables, the recent trace, and authored Policy metadata. Persistent Policies (agent-wide scope) and Continuous Policies (per-turn cadence) bypass the Top-K selection and are always evaluated. Top-K applies to the remaining policy surface. The K is operator-tunable. The injected set is recorded in the production trace, supporting per-decision traceability of which Policies were visible to the decision.