Parametric Memory Boundary. The rule that a model's internal learned associations may help language, ranking, or synthesis, but may not be treated as admissible evidence for a consequential claim.
Parametric memory is useful, but it is not a source of record. It has no provenance the system can inspect, no freshness guarantee, no approval state, and no preserved citation edge. Law V therefore draws a boundary around it: the model may reason, but the facts that ground the reasoning must come from governed evidence.
Crossing this boundary turns capability into ungrounded confidence. Retrieval-grounded reasoning keeps the boundary intact by putting admitted evidence in front of the model before generation and preserving the derivation afterward.