user = User.new(password: "あ" * 25) # 25 characters, 75 bytes

And misunderstandings so easily occur here, when we're talking about encodings, but not those encodings, the other encoding, which is really charset. And it's especially hard because you can't visually tell the difference and in so many cases everything still works even though it is wrong.

The btoa() function takes a JavaScript string as a parameter. In JavaScript strings are represented using the UTF-16 character encoding: in this encoding, strings are represented as a sequence of 16-bit (2 byte) units. Every ASCII character fits into the first byte of one of these units, but many other characters don't. Base64, by design, expects binary data as its input. In terms of JavaScript strings, this means strings in which each character occupies only one byte. So if you pass a string into btoa() containing characters that occupy more than one byte, you will get an error, because this is not considered binary data:

If you need to encode Unicode text as ASCII using btoa(), one option is to convert the string such that each 16-bit unit occupies only one byte.

The character exists in Unicode/ISO 10646, but not in the character encoding used for the document. In this case, use Numeric Character References (NCRs, example: 噸).

�Yes, but how will we ever keep track of such a large project?�

The use of U+212B 'Angstrom sign', which was encoded due to round-trip mapping compatibility with an East-Asian character encoding, is discouraged, and the preferred representation is U+00C5 'capital letter A with ring above', which has the same glyph.