I did find an old thread that vaguely describes the empty string change, but the wording is too broad for me to understand the necessity of the technical rationale:
Variants, notes, and references are also not the same as translations (every segment ID does not have an annotation), so I’m not sure if that rule was supposed to apply to them.
Non-sutta translations (essays, blurbs, sutta tree names, interface elements, etc) are also a different category, one where it is more likely for there to be partial/incomplete translations.
Afaict, all of the bugs from the empty string root cause are in those two categories. Perhaps a partial revert could make sense too
