Research Judgment Checklist

When reviewing a claim, ask:

1. What exactly changed?
2. What exactly stayed fixed?
3. Is the baseline strong?
4. Is the metric aligned with the claim?
5. Are compute and data controlled?
6. Are results stable across seeds?
7. Are slice results reported?
8. Are there failure cases?
9. Is there any leakage or contamination risk?
10. What is the strongest conclusion the evidence actually supports?