Base91 Encoder/Decoder

basE91 uses 91 of the printable ASCII characters, against base64's 64, and with that the growth drops from 33% to roughly 23%. Its mechanics differ from the other bases: instead of grouping into fixed blocks, it accumulates bits in a register and emits two characters whenever 13 or 14 bits are available — the count varying according to whether the value fits in 13 bits.

Type the text and get the encoding. That 13-or-14-bit variation is what makes the format more efficient than base85, and also what makes it harder to implement correctly: an error in the threshold produces plausible, undecodable output. The implementation was checked against known vectors — "test" encodes to fPNKd and "a" to GB.

The price of that efficiency is the alphabet, which includes double quotes, the dollar sign, the backtick and other characters needing escapes in nearly every context: inside code strings, in URLs, in XML, in CSV. That is why basE91 shows up rarely in practice despite being measurably smaller than base64 — the saving seldom justifies the extra care in transport.