Eval Dataset Design

Design evaluation datasets for LLM or agent systems with representative task coverage, edge cases, clear labels, and stable scoring intent. Use when Codex builds or improves eval suites.