It's a way to validate the LLM output in a test scenario.