Mainstream RAG evaluation frameworks include RAGAS, TruLens, and DeepEval, which typically score across two dimensions—retrieval quality and generation quality—but lack built-in support for specialized requirements such as exact identifier matching and rule clause completeness