Preregistering Analysis

This enforces the discipline of locking your hypotheses, analysis plan, and decision rules before you touch outcome data. The idea is simple: if you decided what test to run after seeing the results, you can't tell whether the data shaped your conclusions. It gives you a template for writing predictions, walks you through verifying they're actually falsifiable, and commits everything to git with a timestamp before you execute. The "looked at it? it's exploratory now" rule has no exceptions, which will feel rigid until you've watched a few analyses that seemed obviously confirmatory turn out to hinge entirely on choices made after peeking. Think of it as test-driven development for science: write the test before the code so you know the test tests something real.

Install to Claude Code

npx -y skills add k-dense-ai/science-superpowers --skill preregistering-analysis --agent claude-code