LLM evaluation report

Write an LLM evaluation report for an AI help-center answer bot, comparing three candidate models. Purpose: decide which model to use and whether the feature is ready to launch. Requirements: open with the recommendation, describe the eval set's sources, size and scenario mix, define metrics such as accuracy, factuality and format compliance and how they're scored, compare quality, latency and cost per request in a table with a breakdown by scenario, categorize failures with representative examples, report safety test results, and close with launch criteria, a staged rollout plan and production monitoring metrics. Base everything on the eval data and don't invent scores.