Multiple Choice Flashcards

Survey whether users like the loading experience instead of measuring retrieval freshness. Measure retrieval freshness, retrieval latency, generation latency, answer groundedness, and user task success separately. Evaluate the foundation model's public accuracy benchmark as a proxy for the deployed RAG pipeline. Review total monthly token spend and defer component-level testing until the budget changes.