1. Define success first
Write the constraints, required evidence, and acceptable uncertainty before model fluency can change your standard.
2. Separate claims from presentation
A polished answer may contain weak claims; a plain answer may be correct and useful. Review those dimensions separately.
3. Trace evidence
For factual claims, prefer current primary sources and confirm that each source actually supports the nearby claim.
4. Locate uncertainty
Ask what is not observed, what could confound the result, and what new evidence would change the decision.
5. Decide for this task
Choose the answer that best satisfies this prompt and risk level. Do not turn one example into a global model ranking.
External methodology reference: NIST AI Risk Management Framework. The guide is educational and should be adapted for domain-specific risk.