CHARSET  [protocol]
RFC 2066's telnet option for agreeing an encoding. The reason a game's accented
names survive the trip, and the source of some subtle failures when it is
absent.
https://www.rfc-editor.org/rfc/rfc2066

Measured adoption
  114 of 886 listed games were observed offering it (12.9%)
  /games?protocol=CHARSET

  The games not counted here are not games without the protocol. A game is
  counted when we observed its server offering the option in a handshake; the
  rest are servers that did not offer it to us and servers whose handshake we
  have not read, and we cannot tell you which.

  By codebase, of the games we identified
    PennMUSH             28 of 34 offered it
    FluffOS              16 of 28 offered it
    RhostMUSH            5 of 40 offered it
    ROM                  2 of 55 offered it
    SMAUG                1 of 20 offered it
    AresMUSH             0 of 18 offered it
    CircleMUD            0 of 47 offered it
    CobraMUSH            0 of 2 offered it
    CoffeeMUD            0 of 32 offered it
    DikuMUD              0 of 35 offered it
    Evennia              0 of 11 offered it
    MUCK                 0 of 24 offered it
    tbaMUD               0 of 5 offered it
    TinyMUSH             0 of 9 offered it

CHARSET is telnet option 42, specified in RFC 2066. One side offers a list of
character sets, the other picks one, and both then agree on how bytes map to
characters.

In practice the negotiation settles on UTF-8 or is not held at all. The MUSH
family negotiates it noticeably more than the MUD family — TinyMUX, RhostMUSH
and PennMUSH all do — which reflects a population that writes prose with names
in it.

WHAT HAPPENS WITHOUT IT

A client has to guess, and the usual guess is either ASCII or Latin-1. Guess
ASCII and every byte above 0x7F becomes a question mark; guess Latin-1 on a
UTF-8 server and every accented character becomes two pieces of punctuation.
Both failures look like the game's fault and are not.

For a crawler this bites in a specific place. Our own telnet library defaults
its current encoding to ASCII, and that default is not inert — it is what every
byte is decoded with, for every server that never negotiates CHARSET, which is
most of them. We seed it deliberately for that reason.

THE ONE PLACE CHARSET DOES NOT REACH

MSSP field names and values are decoded as ASCII regardless of what CHARSET
settled on, because a subnegotiation is a command rather than text and the
specification scopes CHARSET to text. That is arguably conformant and it is
lossy: a game whose MSSP NAME is Café Noir reports Caf? Noir, and the original
bytes are gone before anything we control sees them.

If you see a mangled character in a declared field on this site and not in the
game's own output, that is why, and it is not recoverable from our side.

See also
  TTYPE and MTTS — /reference/protocols/ttype
  How to connect — /reference/connecting
  TinyMUX — /reference/codebases/tinymux