Answer only from the supplied original sources. Emit short canonical claims, with your own claim IDs and exact supporting source passages. Use only the source IDs and temporal target IDs provided in the inputs. Report abstention, unavailable sources and terminal failures explicitly. The answer label files and private judge observations are unavailable to your tools. Include every actual attempt and auxiliary call in usage receipts; unobserved telemetry is null.
