Etl as text is good, because you can save it in version control. (Is it “code” or “json” is irrelevant for the vcs)

Edit: save in vcs stringly implies usability of ‘diff’ and ‘grep’