check-reproducibility

What problem does it solve?

This Skill prevents silent changes in results by verifying that a full fresh-clone run reproduces the committed pipeline outputs without drift.

Core Features & Use Cases

• Fresh-clone simulation: Re-runs the entire pipeline in a cleaned environment (while preserving data/raw/) to emulate real reproducibility conditions.
• Output diffing with categorization: Diffs the newly generated output/ against a snapshot and classifies drift into FAIL/WARN/PASS, with special handling for numerical tables versus visual artifacts.
• Failure-safe reporting: Surfaces whether the pipeline itself failed, which files differ, what category of drift occurred, and likely suspects (e.g., seed randomness, package/version drift, missing committed inputs).

Use case example: After upgrading Stata or touching model code, run this Skill to confirm that tables (especially .csv) match exactly and figures did not unexpectedly change.

Quick Start

In your local repo, run the Skill command that triggers the reproducibility check and outputs a PASS/WARN/FAIL verdict with a diff summary.