meleantonio Replication And Research Hygiene
Trigger
Use when the task involves:
- making results rerunnable by someone else
- deciding which scripts or notebooks are canonical
- packaging tables, figures, simulations, or model outputs for handoff
- checking for hidden path, seed, dependency, or environment assumptions
Workflow
- Inventory inputs, outputs, environment assumptions, and any required secrets or credentials.
- Distinguish exploratory artifacts from canonical pipelines.
- Verify paths, seeds, configuration values, and result-generation steps.
- Write a short replication note with commands, dependencies, expected outputs, and remaining blockers.
Guardrails
- Do not claim a result is reproducible if it depends on undocumented manual edits.
- Avoid absolute paths and hidden local dependencies.
- Keep raw data immutable.
- Name the canonical script or entry point that regenerates each table, figure, or model output.
Output
Return:
- the canonical entry points
- required inputs and environment assumptions
- commands or steps to regenerate outputs
- unresolved reproducibility gaps
Replication Note Template
## Replication Note
- Entry point: `path/to/script.py`
- Inputs: `data/raw/...`
- Environment: Python 3.11, MATLAB R2025a
- Outputs: `results/tables/table_1.csv`, `results/figures/figure_2.png`
- Remaining blockers: none / [list blockers]