For consistency, please s/_Note_/*Note*:

This isn’t specific to base64, it applies at least to hex, too:

> Buffer.byteLength('aa  ', 'hex')
2
> Buffer.from('aa  ', 'hex').length
1

(I think it’s only those two encodings but I didn’t double-check.)

@addaleax oh indeed. That makes wording it a bit more complicated...

nit: s/aren't/are not

nit: Maybe mention that the method doesn’t account for whitespace in the input … for some people that might be included in “invalid”, but that’s kind of subjective…

The return value could be smaller too (e.g. Buffer.byteLength('1345 4 ', 'hex') === 3, Buffer.from('1345 4 ').byteLength === 7), but then since the data is invalid it's just hard to say. Maybe just say something like the return value is meaningless and should not be used?

Hmm, wait, I forgot to pass hex to Buffer.from('1345 4 '), that one would just throws TypeError: Invalid hex string, so there is nothing to compare :/

Yeah, logically there is no way for it to be smaller. That would imply that you can encode more data by using an invalid string than a valid string.

It's not completely meaningless - e.g. if you allocate a Buffer with the size at least Buffer.byteLength, then the actual buffer will fit in it. Although I can't imagine a use case for it.