Reproducibility

Reproducibility means that someone else (or you, six months later) can take your data and your code and obtain exactly the same results. It is the difference between a one-off lucky finding and a trustworthy, checkable analysis.

Achieving it in practice means recording every step in code rather than clicking around by hand, keeping the raw data untouched and transforming copies, and noting the versions of the tools you used. It matters because unreproducible results cannot really be verified or built upon, and across science the inability to reproduce published findings is a recognised crisis. A useful related idea is replicability: getting the same conclusion from a fresh, independent dataset, which is an even stronger test.