By name: reproducible-analysis-reporting description: Turn data science notebooks, scripts, and results into reproducible reports or handoffs. Use when Codex needs to rerun an analysis, structure notebooks, produce written findings, create reproducibility instructions, document data lineage, set seeds, compare generated outputs, prepare stakeholder reports, or communicate results with caveats and clear figures.
Reproducible Analysis Reporting
Treat a report as both communication and an executable claim.
Domain Context Requirement
Use the Domain Context Contract to shape the report narrative, caveats, audience language, metrics, and next-step recommendations. Reports must distinguish user-provided domain facts, researched facts, data evidence, and inferences. If the domain understanding changed during analysis, include the updated contract.
Reproducibility Procedure
- Identify raw inputs, derived inputs, scripts/notebooks, environment files, and generated outputs.
- Ensure the analysis can run from raw or declared intermediate inputs.
- Remove hidden notebook state or document the exact execution order.
- Set random seeds where stochastic steps affect results.
- Capture package/runtime versions using project-native tools.
- Regenerate outputs and compare key metrics, tables, or hashes where practical.
- Write a concise caveat section that distinguishes evidence from interpretation.
Reporting Procedure
- Lead with the answer to the framed question and the stakeholder decision it supports.
- Show only tables and figures that support the answer or important caveats.
- Label axes, units, denominators, sample sizes, filters, and time windows.
- State uncertainty and limitations near the claim they qualify.
- Separate exploratory findings from confirmatory or predictive results.
- End with decision implications and recommended next data checks or experiments.
Output Template
Domain context used:
Answer:
Evidence:
Uncertainty:
Limitations:
Reproducibility commands:
Inputs:
Outputs:
Environment:
Residual risk: