Quality checklist
Use this checklist to reduce mistakes and “confident wrong answers”.
60-second check (fast)
- Goal match: does it answer the task you asked?
- No invention: dates/names/numbers are not invented.
- Assumptions labeled: unknowns are marked clearly.
- Format OK: requested structure (table/sections) is respected.
- Actionability: next steps are concrete (if requested).
5-minute check (safer)
A) Factual integrity
- Any claims that need verification are labeled [CHECK].
- No hidden “new facts” were introduced.
- Definitions match the context (no term drift).
B) Completeness
- All constraints were respected (length, tone, required points).
- Missing inputs are listed as questions or assumptions.
C) Risk & privacy
- Sensitive information is not exposed.
- The output avoids overconfident legal/medical/financial claims.
D) Clarity
- Short sentences, minimal jargon.
- Headings reflect content.
Copy-paste prompt (QA pass)
Run a quality pass on the output below.
Rules:
- do not add new facts
- flag [CHECK] items that require verification
Output:
1) Issues list (severity: High/Med/Low)
2) Proposed safe rewrites
3) Questions for missing inputs
Text:
<PASTE>
When to stop and re-ask
- The model is guessing critical details.
- The answer depends on missing data.
- The format is wrong twice.