LLM inference is nearly deterministic. We use this to audit providers | Not Hacker News!