Your question is Hallucination Evaluation Strategies. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
Compare different strategies for LLM evaluation, specifically when dealing with hallucinations in financial documentation.
Discuss how you would evaluate factual correctness, citation faithfulness, appropriate refusal, and harmful unsupported claims. Compare golden datasets, programmatic checks, human review, LLM-as-judge methods, adversarial testing, and production monitoring. Explain the tradeoffs among evaluation quality, coverage, cost, and latency.