Finalize Analysis

Organize messy exploratory work into a clean reproducible project with Snakemake pipeline and publication-ready outputs. Create one figure or table at a time with user approval.

Preconditions

Check for docs/analysis-plan.md and docs/codebook.md (from /init-research-project). If found, use them to guide organization. If not found, ask user about research goals and data before proceeding.

When to Use

• •Exploratory analysis done, need clean reproducible outputs
• •Messy scripts need organizing into numbered pipeline
• •User says "clean up", "finalize", "make reproducible", "prepare for publication"
• •Ready to create publication-quality figures and tables

When NOT to Use

• •Starting from scratch with new data (use /init-research-project)
• •Single plot or table (use /viz, /coefficient-plot, etc.)
• •Already have clean pipeline, just need paper (use /draft-paper)

Workflow

Step 1: Scope Decision

AskUserQuestion: New directory or restructure existing? What to preserve? What to discard?

Done when: User confirms scope choice.

Step 2: Survey Existing Work

Scan project for scripts, figures, tables, data files. Present inventory as a table. Ask what to keep, drop, or revise.

Done when: Keep/drop decisions made for all existing outputs.

Step 3: Plan Structure

Propose numbered script sequence following the standard pipeline (see references/project-script-workflow.md). Confirm with user.

Done when: User approves script sequence.

Step 4: Iterative Output Creation (core loop)

For each figure or table, one at a time:

1. •Discuss: AskUserQuestion about what the output shows, data source, design preferences. Suggest ideas for key tables.
2. •Write script: Create numbered R script. Chain to /viz, /coefficient-plot, /marginaleffects, /survey-analysis as appropriate.
3. •Execute and show: Run script, display result to user.
4. •Iterate: Ask for tweaks, loop until approved.
5. •Document: Write companion .md file (see references/companion-md-template.md).
6. •Update pipeline: Add Snakemake rule. Update data cleaning/measure creation scripts if needed.

Done when: User approves the output and companion .md is written.

Step 5: Output Formats

Every figure saved as PNG (300dpi) + PDF. Every table saved as CSV + LaTeX (modelsummary). Optionally Excel. See references/output-formats.md for dimensions and code patterns.

Step 6: Finalize

Update CLAUDE.md with final script list and output inventory. Verify full pipeline:

snakemake -n          # dry run
snakemake --cores 4   # full run

Done when: Pipeline runs end-to-end without error.

Quick Reference

Common Mistakes