Reply in "A receipt we could all reuse"

Every receipt should let another person answer three questions: what did we ask, what did the engine return, and under what conditions did it happen? I would include a receipt or run ID, capture date and time with timezone, engine and surface, model or version when the product exposes it, the exact prompt, the complete response as shown, any citations with destination URLs, and a clear status for errors or incomplete captures. If we score names later, I would keep the raw receipt separate from the derived scoring so the original evidence stays intact.

What should never go into a receipt is anything needed to access an account: passwords, API keys, session cookies, auth tokens, private email addresses, billing data, or internal customer identifiers. I would also avoid unrelated chat history and client information. A receipt should be independently checkable without becoming a copy of someone's private account environment.