A social deduction game deals most of a lobby one goal and a hidden few the opposite one, then gets out of the way. What settles the match is the argument: what people claim, what they were seen doing, and who the room believes. The design work is all in what the game gives them to argue about.
| Role | Knows | Wants | How they win |
|---|---|---|---|
| Crew | Only that they themselves are innocent | Evidence, and a group that will listen to it | Finish the shared work, or eject every impostor |
| Investigator | One verified fact, usually per night | To be believed without being killed for it | With the crew — but claiming early gets them killed |
| Impostor | Who the other impostors are | Doubt, and somebody else in the chair | Reach parity with the crew, or run the clock out |
| Neutral | Their own private condition | Something neither side is playing for | On their own terms — a jester wins by being ejected |
The asymmetry is informational, not mechanical. Impostors are rarely stronger; they hold a map of the lobby while everyone else holds one square of it. So the crew's real job is to assemble scattered observations into a shared picture faster than a liar can pollute it, and the crew's burden is that an innocent player is being asked to prove a negative.
It assigns roles, keeps the secrets, runs the night, calls the meeting, counts the votes and cuts the timer off. That list is the whole job — and it is exactly the job Mafia needed a person for. Dmitry Davidoff devised Mafia at Moscow State University in 1986: no board, barely any components, and one player out of the game to moderate it. Andrew Plotkin's 1997 Werewolf retheme changed the fiction and kept the machinery intact.
Automating the moderator is why a digital lobby can seat everyone who turned up rather than sacrificing one of them — Among Us runs 4 to 15 players. But nothing about that automation moved the game itself, which still happens in the gap between what someone says and what they did. Software's real contribution was the second half of the design: something for the innocent to do.
Tasks are not content. They are the evidence supply, and the genre's central design rule governs them: tasks must be observable enough to build an alibi on, fakeable enough that watching someone do one proves nothing, and spread out enough that the group genuinely loses track of each other.
Miss on the first count and votes become coin flips. Miss on the second and innocence is trivially provable, which kills the argument outright — Among Us exposes precisely this as a host setting. Visual tasks, the ones that play an animation only a real crewmate can produce, hand the crew hard proof, and plenty of lobbies switch them off to keep the room uncertain. One checkbox moves the entire evidentiary standard of the game.
There is a second way to build the supply, and it is a genuine fork rather than a detail. The Resistance (Don Eskridge, 2009) has no tasks at all: evidence is the outcome of missions the group votes to staff, so the record is public, the deceit is in the staffing, and nobody has to be anywhere in particular. Watched behaviour or read outcomes — pick one before you write a role.
Somebody goes out first, and in the folk version going out means watching other people enjoy themselves. Three fixes are worth stealing. Among Us keeps ghosts working, because a dead crewmate's tasks still count towards the crew win. Blood on the Clocktower keeps the dead in the conversation and gives each of them one vote to spend across the rest of the game. The Resistance deletes the problem by never eliminating anyone.
The other cost has no design fix and deserves saying plainly: with strangers, no voices and one confident bulldozer, votes stop tracking evidence and start tracking charisma. That is a lobby problem, and no ruleset has solved it. Strip the traitors out and the shared chore loop that remains is co-op chaos, which is why sabotage stings — it attacks work you were already invested in. Move the deceit from an alibi into a body and you have hide-and-seek.
Type: Social deduction (hidden roles; the software moderates, the argument decides)
Core loop: Work where you can be seen, watch who can't be accounted for, then argue and vote
Perspective: Top-down or first-person, a lobby small enough to track — Among Us seats 4 to 15
Win / lose: Crew finish the work or eject every impostor; impostors reach parity or run the clock out
Session: 5-15 minutes a round, and a group will play a dozen without repeating one
Progression: Nothing carries between rounds — cosmetics and role variants are the whole meta
Roles are the easy part, and they are what everyone designs first: a long list is quick to write and reads like depth. The evidence rules underneath decide whether any of those roles will have anything to say.