Also, Daniel says that in Cyrillic, variants are word-based rather than character-based and gives an example of E and Ё in one word. I’m not sure I follow the example and logic and tend to disagree, although the exception of “обед” given later on makes a point. I’ll be happy to have a separate discussion with the Cyrillic group to clarify this, but as a linguist and native Russian speaker, I do not see a problem with Ё using E forming variant domain names. There is always a character layer, pure spelling with no pronunciation issues, and that’s what we need to focus on, as that’s what makes up an FQDN.

So, taking on board Siavash’s advice, I’ve made up a short list of working definitions for the purpose of this discussion, just to make myself clear. For me, an atomic unit here is a specific character within a specific language, and the variations this character produces when forming a (domain) name. Then “variant” can be a string of characters that is similar and interchangeable with another string; all “variant” strings form a “bundle”, an atomic domain unit that can be treated as one – cf. the SC & TC treatments in ccTLD registries. If two strings are similar but one cannot be mistaken for another, they are not variants.