How do I create an evidence package for an AI-generated result?
AI output is not evidence. Re-checkable methods are.
1. The problem
Models produce confident prose about deploys, agents, and status. Partners and you still need: timestamped methods, raw results, and clear separation of “model said” vs “chain/RPC returned.”
2. What a good package contains
- The claim (one sentence)
- Method used (RPC, explorer, script name)
- Result (raw or summarized without spin)
- Confidence + unknowns
- What was not checked
QPF’s posture: prepare with AI if you want; authority and verification stay human-readable and re-runnable.
3. Example shape
Certificate #000 style: verified chain and code presence; mint/LP explicitly gated; revenue and adoption not verified.