Verifying Results Before Claiming

This is the pre-commit hook for your brain before you report any finding. It enforces one iron law: run the analysis fresh, read the actual output, then make the claim. No "it was significant earlier," no "the script ran fine," no trusting what a subagent reported. You identify what command proves your claim, execute it from raw data with a fixed seed, read the estimate and interval, check assumptions, and only then state the result with evidence attached. The source is uncompromising about this, calling out common rationalizations like conflating training accuracy with model quality or reporting p-values without effect sizes. If you work with data and have ever caught yourself saying "it should work" before actually checking, this skill is the corrective.

Install to Claude Code

npx -y skills add k-dense-ai/science-superpowers --skill verifying-results-before-claiming --agent claude-code