Yeah, my approach here is to use UTS#39, which is explicitly made with the intent of getting around the problem of confusables. I chose HighlyRestrictive for now since it still allows for things such as "지호Park", without maybe going into questionable territory.
In addition I have checks in place to only allow characters from the unicode classes alphabetical and numerical. So no random emoji, but you can have something like "二十五" as your username (which falls IIRC under the numeric class)

(moved accounts recently, not on @aumetra anymore :P)

Oh yeah, definitely I only do the validation on username selection, not when e.g. logging in a user or resolving webfinger.

But PRECIS is … precisely such a specification. Maybe not the best way to do it, I think @aumetra has a different approach based on unicode classes or something

This is a reasonable choice, since you had to come up with something for yourself. But for #GloballyInclusiveHandles to succeed long-term, I think there needs to be:

1. a specification which removes the burden from developers, of deciding which characters should be allowed and not allowed for creating handles. System admins should also be also be able restrict users to certain scripts within this specification.

2. Recognition that accepting handles is different from creating handles. It is important that most systems accept any handle which fits the whole specification, even if the system would not generate that handle itself. Otherwise users will get frustrated by their handle working then not working, depending on the system.

I was the one who attended the 2025-10 Fediforum, and proposed the topic then. I am persistent.

Cool overview, I didn’t know the same topic came up last time as well!

For flohmarkt, the current implementation uses a python module that implements https://www.rfc-editor.org/rfc/rfc8264 (the PRECIS framework), in particular the username profiles, and then does another check to prevent multiple scripts from being used (based on https://www.unicode.org/Public/UNIDATA/Scripts.txt )

though that’s just what I came up with myself.