There is a bug where pages that start with numbers end up getting put under headers on the category pages that consist of weird Unicode characters, rather than the appropriate number. I don't have an example of this offhand, but it is mentioned at T8948#1919137. I think this only happens under ICU collation, but need to verify.
It would be for icu collation modified to use numeric sorting. It might depend on version of libicu. Im more assuming it happens. Havent really tested
It would be caused by a mismatch in the logic of firstLetterHeader over how numbers sort at a primary strength.