Yeah, that's pretty much it. I suppose it would be half-measure to still include sign characters in the integer and exponent part rather than going all the way and passing those as bool.

Encoding issues are also interesting. It's quite plausible that the input is a UTF-16 or UTF-32 string, and you'd want to parse that "directly". std::from_chars has no Unicode support at the moment, but that almost feels like a separate issue. I feel like it would be fine to have whatever functions come out of this still operate on character sequences with some encoding. Alternatively, you'd pass a sequence of std::bytes, each storing a digit or so, but that's just another encoding, so it's not any better than operating on ASCII/UTF-8 floating-point numbers.