Tag
This paper investigates evidence utility in retrieval-augmented generation, finding that reader-specific ordinal preferences are stable but do not reliably predict intervention outcomes across different readers.