Intelligence
Generated output does not ship without human review
Retrieval and models can accelerate operators. They cannot be the last control before a citizen or a commander sees a result.
WF-INS-2026-02
An intelligence layer is useful when it shortens the distance between a question and a governed answer. It is dangerous when it becomes an unsupervised publisher.
The pattern we use is narrow. Source systems are ingested under role constraints. Retrieval is limited to what the operator is already allowed to see. A model may draft. A person accepts, edits, or rejects before anything leaves the system.
That last step is not theater. It is the control that lets a program say, in an authorization package, that generated language is not an official record until a human has reviewed it. Logging who reviewed, and against which sources, matters more than the novelty of the model.
If a workflow cannot name the reviewer, it is not ready for operational use.