How DoppelUp works

Two verified adults cooperate against one AI. The server is the quizmaster: it sets topics, delivers mail, slips in counterfeits, and keeps score.

A turn

Each message that looks like it came from your teammate has its own reply address. You have two choices.

Reply with only the word AI if you think that message is counterfeit. Case does not matter, and extra space is fine.
Reply with anything else to accept it as human. That reply becomes your next message, unless you were replying to a fake. A fake reply is not forwarded.

A message that starts with AI but says more than that is ordinary game text. The word AI on its own is reserved, so don't try to send that as a real message.

Each message can be resolved once. A second reply does not change the score.

Scoring

Catch a fake: humans +1, and we tell you.
Accept a fake: AI +1. Your reply stays with us.
Accuse a real message: humans −1. It is not sent again.
Clearly off topic: humans −1. Tangents are allowed. If the judge is unsure, the message stands.

Humans win if their team score is strictly greater than the AI score. A tie is a draw. The rule is fixed for the game when it starts.

Length

A game has up to 20 accepted human messages, 5 AI interventions, and 2 topics. It ends when the fakes are all resolved or the human-message limit is reached.

A quiet game gets a nudge after a day, another after three days, and expires after 7 days. Nobody is penalised for an abandoned game.

What the AI is allowed to do

It only sees the current game. It does not build a profile from your other games, and we do not train a custom model on your messages. It must not invent emergencies, ask for money or passwords, threaten anyone, or drag the conversation somewhere explicit or abusive.

The mail comes from DoppelUp, not from your friend's personal address. We never pretend their domain authenticated the message.