research-data-analysis

Research Data Analysis

Analysis should be repeatable, inspectable, and connected to claims.

Read First

• references/repository-contract.md
• references/output-contracts.md
• references/workflow-stage-contracts.md

Workflow

1. Run npm run workflow:analysis when available.
2. Read the contribution-local analysis.yaml, blocker summary, data paths, figure catalog, stats appendix, and paper-export folder.
3. If required preflight fields are missing, write only the blocker summary.
4. Identify raw, interim, and processed data paths.
5. Read data dictionary and provenance notes.
6. Write reusable analysis logic in src/, not only notebooks.
7. Use scripts or CLI entrypoints for repeatable runs.
8. Put generated tables and figures in the analysis bundle or declared output paths.
9. Record important findings in the contribution report, claim map, or wiki pages.
10. Link paper claims to exact table, figure, or experiment artifacts.

Statistical Discipline

• Match tests to design, distribution, sample size, and hypothesis.
• Report effect sizes and confidence intervals when relevant.
• Distinguish statistical significance from practical significance.
• Flag multiple comparisons, missing data, selection bias, and confounds.
• Do not treat exploratory analysis as confirmatory.

Figure Discipline

• Use readable labels, units, captions, and colorblind-safe palettes.
• Avoid misleading axes and decorative chart types.
• Save source data for final figures when practical.
• For publication figures, record command and input dataset.

Notebook Policy

Notebooks can explore and explain. If the logic becomes part of the result, move it into src/ and leave the notebook as a consumer.

Project Quality

Keep raw, interim, processed, exploratory, analysis, and final paper-facing outputs separated. Promote a table, figure, statistic, or model result only when the input data, command or procedure, environment notes, validation check, and linked claim or experiment are recorded. If an analysis supports an artifact or reproducibility claim, add the evidence path and validation status to artifacts/badge-evidence-ledger.csv.

Review Loop And Handoff

Run, validate, review, fix, and re-review the analysis until method, data, metric, interpretation, and clean-copy issues are resolved. Handoff reviewed generated outputs to research-results-reporting and paper-facing assets to publication-figures-tables.