Setting Up Reproducible Analysis

This is the checklist you wish you'd had before that analysis you couldn't reproduce six months later. It walks through setting up an isolated worktree (with detection to avoid nesting them), pinning your environment with lockfiles, fixing random seeds globally, and making raw data immutable with checksums. The Step 0 worktree detection is especially thoughtful since it checks for existing isolation before creating more. Use it before running any analysis that matters, especially pre-registered work where reproducibility isn't optional. The emphasis on baseline verification (does the existing analysis still reproduce?) before adding new work is the kind of discipline that separates real science from notebook chaos.

Install to Claude Code

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