Query-Aware Source-Risk Triage for Retrieval-Augmented Generation

Retrieval-augmented generation (RAG) pipelines typically rank evidence without asking whether a source has a material relationship to the query. We study a pre-generation triage layer that treats this relationship as query dependent. The method first routes canonical query families for enhanced review and then assigns retrieved pages to pass, contextualize, exclude, or review. It combines a four-dimension page score, rank-discounted family aggregation, intent-preserving query mutations, and a family-held-out router. A manual audit of real URLs fixes calibration anchors; a constructed scenario derived from those anchors is used for workload and robustness analysis. An oracle page gate then defines the risk–coverage operating target for a future learned classifier. The evaluation shows why page-level frequency cannot substitute for family-level exposure and quantifies how calibration changes review load. The result is an auditable triage method and implementation contract. The present evidence does not estimate live-Web prevalence or downstream answer improvement.

• Machine intelligence