How do I allow "stępień" while detecting Zalgo-isms?
n=1 is probably a reasonable falsehood to believe about names until someone points out that language X regularly has multiple combining modifiers in a row, at which point you can bump up N to somewhere around the maximum number of combining modifiers language X is likely to have, add a special case to say "this is probably language X so we don't look for Zalgos", or just give up and put some Zalgo in your test corpus, start looking for places where it breaks things, and fix whatever breaks in a way that isn't funny.
[0] I say conceptually because you might implement a form-insensitive Zalgo detector that looks at each non-combining codepoint, looks it up in the Unicode database to find how many combining codepoints one would need if canonically decomposing and call that `n`, then count from there all the following combining codepoints, and if that exceeds `N` then reject. This approach is fairly optimal because most of the time most characters in most strings don't decompose to more than one codepoint, and even if they do you save the cost of allocating a buffer to normalize into and the associated memory stores.
If you really think you need to programmatically detect and reject these (I'm dubious), there is probably a reasonable limit on the number of diacritics per character.
I think we based it on Lodash’ deburr source code. If deburr’s output is a-z and some common symbols, it passes (and we store the original value)