A dataset or codebase is often the most reusable part of a study, and the least credited. A record shows who made it, who checked it, which versions changed what, and every paper and page that relies on it.
This matters most when the data is reused. A figure traced back to a dataset tells a reader where the numbers came from; the dataset's own record tells them who assembled it, what changed between versions, and whether the version cited is the one still recommended. The people who built it get named for the work every time it is used again.
Versions are tracked rather than replaced. A dataset corrected after publication leaves the earlier version visible with a note on what changed, because papers already built on it need to stay checkable. Anyone citing the data points at a specific version, and the record shows whether that version is still the one to use. A record that nobody can version is a record nobody can cite precisely, which is how a reused dataset ends up supporting two papers that disagree.