Actually that's wrong, in the morning I convinced myself that given a sequence of code points which forms a valid Unicode character (an extended grapheme cluster), any non-empty prefix of it also forms a valid Unicode character.