I mostly wrote this post as an excuse to try the freshly-minted and excellent pydantic-evals framework for LLM evaluations but one intere...