Dataset versioning is keeping a record of what the data looked like at a specific point in time, so that an analysis can be repeated with the same result. Without it, a common situation arises: a report run today shows different numbers than a month ago, even though nothing was supposed to change, because records were continuously corrected and supplemented. For a one-off analysis this does not matter, but for regular reporting and for training models it is a fundamental problem. The solution is to keep snapshots of the data at the end of each period, record what processing rules were used, and always state alongside the result which version of the data it is based on. It is also worth recording changes to definitions and cleaning rules. For models, the data version is just as important as the version of the model itself.
See also: Data dictionary, Data quality in analytics, Technical documentation of an AI system.
