> encode a grapheme cluster as a constant size […] totally possible with damn near every language that people actually use This is not true. For a concrete exa

> encode a grapheme cluster as a constant size […] totally possible with damn near every language that people actually use This is not true. For a concrete example: the languages Hindi and Marathi, with ~500 million speakers, use the Devanagari script (also used by Nepali and Sanskrit), in which a grapheme cluster is (usually) a sequence of consonants followed by a vowel. For instance, something like "bhuktvā" (भुक्त्वा) would be two grapheme clusters, one (भु) for "bhu" and one (क्त्वा) for "ktvā". In Unicode each vowel and consonant (here, bh, u, k, t, v, ā) is separately encoded, which is the only reasonable thing to do, and inevitably means that grapheme clusters can have different lengths (number of code points). The alternative would have been to encode every possible (sequence of consonants + vowel) as a single codepoint, which gets ridiculous quickly: these sequences can be up to 5 consonants long, so you'd end up having to encode (33^5 * 13 ≈ 500M) codepoints for Devanagari alone (or completely prevent certain sequences of consonants from being expressed, which makes no sense either), not to mention that most of the scripts of the Indian subcontinent and south-east Asia follow the same principle and have similar issues (e.g. Bengali with 250M speakers, Telugu, Javanese, Punjabi, Kannada, Gujarati, Thai with over 50M speakers each, etc). (See chapters 12–17 of the Unicode standard, currently version 15: https://www.unicode.org/versions/Unicode15.0.0/ch12.pdf https://www.unicode.org/versions/Unicode15.0.0/ch12.pdf)