I suspect C++ uses the locale collection facet (a mapping/ordering of characters) instead of relying on their actual byte values. Need to see how easy it is to implement it in Typesense since we have to store the records in pre-sorted form.