Reproducibility as a Discipline

A result that cannot be reproduced is not yet a result; we build every analysis so that someone else can run it and get the same number.

Reproducible or it does not count

Reproducibility is the property that another competent person, given our inputs and code, gets our output. We treat it as a precondition for a result to exist, not a courtesy added afterward. If we cannot reproduce our own number next year, we have no basis to ask anyone to believe it today.

What reproducibility requires in practice

• Versioned code, so the exact program that made a number can be recovered.
• Pinned environments, so libraries and compilers do not silently change the answer.
• Recorded inputs and random seeds, so a run is deterministic.
• A public archive with a permanent identifier, so the record does not depend on us.

Why we archive publicly

Our papers, data, and analysis are deposited with permanent DOIs under open licenses so the record outlives any internal system and anyone can check it. Reproducibility that only works on our machines is a promise; reproducibility anyone can exercise is a proof.

Distinct from the honesty of labelling claims, this is the mechanical discipline that makes the labels checkable. None of this is exotic; it is ordinary software and record-keeping discipline applied with unusual seriousness, precisely because the results it protects will be weighed by people who were not in the room when they were produced. A reproducible record is the closest thing a design study has to a witness: it can be questioned long after the work is done, and it will answer the same way every time.