Courts IT wants docket summaries shown to clerks only when citation IDs are present. Which generative optimization helps catch missing citations?

Courts IT that wants docket summaries shown to clerks only when citation IDs are present should implement a self-critique pass that checks citation IDs before display. Clerks need summaries that point back to docket sources. A critique gate that verifies citation fields are populated (and well-formed) blocks incomplete drafts from reaching the desk. That generative optimization is a quality/gate pattern: draft, critique for required citations, then show or revise. Tuning sampling temperature to enforce citation ID presence in clerk summaries fails because temperature only controls randomness, it cannot verify or enforce required fields, so missing IDs still pass through with no display gate to catch them. Adding a docket reranking step that scores summaries with citation IDs higher fails because it only reorders candidates by preference score, it cannot verify required fields or block display, so missing IDs still pass through with no completeness gate to catch them. Compressing retrieved dockets to fit citation IDs and longer histories into context fails because fitting more into context does not verify IDs are present before display, so incomplete summaries still reach clerks. Exam caveat: citation self-critique is a pre-display completeness gate for clerk summaries—not tuning sampling temperature to enforce citation presence, reranking that scores summaries with citation IDs higher, or compressing dockets to fit IDs and histories. Operational check: require citation ID checks in the critique pass, withhold clerk display when IDs are missing, log critique failures for courts IT review, and sample displayed summaries to confirm every one includes citation references.