• 2.1 The Golden Dataset
• 2.2 Retrieval Metrics (for RAG systems)
• 2.3 Generation Metrics
• 2.4 Agent-Specific Metrics
• 2.5 LLM-as-Judge (the technique that powers most of the above)

2.6 Frameworks to Name

• RAGAS — the most-cited OSS RAG eval library. Implements Faithfulness, Answer Relevancy, Context Precision/Recall out of the box.
• DeepEval / Promptfoo — general LLM eval frameworks, support pytest-style assertions.
• Vertex AI Gen AI Evaluation Service — Google's managed offering, computational and model-based metrics, integrates with Vertex AI Pipelines. Name this in GCP-flavored answers.
• LangSmith / Langfuse — used for both evals and tracing; the dataset + experiments features are eval-flavored.
• TruLens — eval + observability oriented around RAG triad (groundedness, context relevance, answer relevance).