If you’ve been solving inconsistent LLM behavior by continually expanding your prompts, you may be optimizing the wrong variable. As…