@rysiek @xpil
vv would appear like w and score bad, qg could still be a nightmare depending on the attack/use, cirrilic users would less likely be tricked in clicking on latin codes and the other way around, there's no ownership or who came first issues, but attacks on established domain names (based on user trust) could hopefully more easily be spotted.

@rysiek @xpil apparently you have spent much time thinking about this problem, and I'm completely new, but intrigued. Now I can't think of catch all solution but maybe this is something that might help a little.

@rysiek @xpil
What if a browser saves a graphic representation of every domain visible in the address bar (as the user sees it) , and compares it graphically (with the prior recorded domain representation images) when a domain is about to be opened (as in, seen by the user in the address bar) and warns when similarity is above a given (yet to be discovered) threshold AND the domain names in unicode don't match (or differ enough based on textual difference)

@rysiek @xpil

Assuming that most attacks are based on trusted (by the user) domains,
Assuming that a centralized option is not preferable (for all good reasons set out elsewhere in this thread),
Assuming that taking user browsing preferences are important,
Given that the total of glyph combinations is too large too handle effectively on large scale,