Observability

Here’s how most people build an eval. They open a file, write an LLM judge prompt that says …