arXiv:2611.15700v1 Announce Type: cross Abstract: AI-based question-answering systems are increasingly used for user support, including at high-performance computing (HPC) facilities. However, existing systems lack a systematic, deployment-oriented approach for evaluating final-answer quality and attributing failures to specific components.
