Now that I’m a fair way in to developing a social deduction game I find myself trying to work out how to analyse social deduction game states. A competitive game should be winnable by either side, depending on their skilful choices and decision making, I want to get more into how social deduction games achieve this goal as a means to working out what to include in the people’s orchestra game.
A quick intro to social deduction for the newbies (Skip to the next para if you’re not): In a social deduction game the players work together to defeat some scum. The twist is that the scum is among them, as one or more of the players and contributes to their efforts. None of the other players know who they are so the scum can try to trick the group into eliminating one of their honest contributing members instead.
Most social deduction games can track their origin to mafia (or werewolf, same game, different name). In this game players first collectively vote on who to kill, eliminating someone from play and then the scum secretly kill one player of their choice. Because the players have greater numbers if each side kills correctly every day the village will win, but since the scum get to talk and vote around the village kills they’re able to win the game by persuading the village to kill several of its own.
This game can then be spruced up by adding special powers. One player might get to make an independent kill every other turn, one might get to save a life, one might get to find out who someone else is and so on. I’ve played an awful lot of mafia and we tried all sorts of exotic combinations of powers. Some combinations were lots of fun and lead us to interesting games, others were clearly broken and meant that one side or the other could always win. This post is about analysing the differences between games that are broken in that way versus ones that are not.
The goal is to look for ways in which one group of players or the other can execute a strategy resulting in a guaranteed win. Here are a few examples:
Mathematical Guarantee: Imagine a game with three scum and three villagers. On the voting phase the scum can vote together, meaning that at best the villagers can tie and will fail to kill. Then the scum use their kill repeatedly with no opposition until they win. Trivial version of this are easy to avoid, but it can arise as a result of other complications.
Role Combination: Imagine a game in which one player can save a life each night, one can check an identity each night and one can target a person to power them up to hit multiple targets. Individually none of these break the game, but in combination the checker could reveal themself, knowing that the hidden life saver would preserve them for at least one night and the empowerer could make them find enough information in that night to push the game into a mathematical guarantee situation.
Mass Roleclaim: Imagine a game in which every player has a unique power and the list of powers is known. The town could open the game by asking everyone to state which power they have, then any person who does not have someone duplicating their claim can be guaranteed to be innocent. Depending on other aspects of the setup this could also push the game into a mathematical guarantee situation. Alternatively, depending on the powers, the town could insist that one of each pair use the power in a certain way while the other did not use it and then work out who is lying in order to correctly target all of their votes in that fashion.
There are plenty of other ways in which mafia setups can be broken, but that gives an idea of the shape of the problems that a particular setup could have. There are also other ways in which setups can be bad (one with a great many killing roles that causes the game to end prematurely for instance) but this is a discussion of social deduction in general so let’s talk about how the game has evolved.
The genre has moved in a lot of different directions, producing versions with no player elimination such as Avalon or ones based around an unreliable narrator such as Nosferatu, but I’d like to talk about a specific direct descendant of the early mafia/werewolf games: One Night Ultimate Werewolf.
This game works in a very similar way to that described above, but instead of alternating day (voting) and night (scum kill and powers) phases the game has only one phase of each type. First each player may secretly use their power, then everyone collectively votes on who to eliminate. If they get a scum they win, otherwise the scum wins. It’s a wonderful paring down of the concept to distil its most interesting and exciting moments.
With no discussion before power use it’s almost impossible for coordinated power use to create a broken scenario. With a single vote, a mass roleclaim is also impossible as there’s no time to work out which of the duplicate claims are lies – furthermore some roles have duplicates which reduce the odds of getting a correct kill by targeting a group that have claimed the same role and some roles win if they are lynched (but cause the town to lose) so anything that looks like “guaranteed scum” behaviour can always be someone faking it.
While effective, I’d argue that these factors don’t guarantee avoiding a broken gamestate. Some roles are able to look at other roles cards, which can’t be faked effectively. In principle the discussion is the meat of the game, so if some combination of power use guarantees that the discussion will go in a particular direction very quickly then there’s still a problem. For instance if “I looked at Mike’s card and he’s a werewolf” terminates a discussion quickly, with a relatively quick resolution to a potential fake claim of “Nah, I am the seer and I looked at…” based on other knowledge then the game state still wouldn’t work.
ONUW undermines this problem by ensuring that the information gained is at least slightly unreliable. All roles that look at cards go before all roles that switch cards between players and the mere existence of these roles is sufficient for there to be the ability for scum to cover each other by claiming to have switched the card of the player under suspicion. This combination of factors keeps the majority of vote discussions interesting and dynamic.
There’re some similarities between this design and the orchestra design and some significant differences. Most striking is that rather than a team of scum there is a single villain in the shape of the Artiste and rather than a vote there is a single choice that comes from the Manager. This is mitigated by the Manager having exceptionally poor information gathering skills and needing to rely on the other players to learn the identity of the correct target and the other players generally winning if the Manager is correct but sometimes working at tangents to or in direct opposition to him.
The structure of the game is slightly different as the pre-discussion stage is longer (each player gets three opportunities to act rather than one) but also more interesting (you don’t spend 90% of it, or in fact any of it, with your eyes closed). However the volume of things that can happen in the pre-discussion phase runs the risk of reaching the discussion phase with the Manager in a position to immediately name the Artiste with no opportunity for the lies an discussions that usually make up the meat of social deduction games to take centre stage.
A lot of the straightforward measures are implemented. There are roles that side with the artiste providing some cover from accusations by deliberately looking suspicious. There are roles with ulterior motives allowing the artiste to throw doubt onto those who’ve genuinely seen her identity. However there remain some strategies that were still unnaturally strong.
The Artiste Swap: A player with the Artiste card switches cards with another player, giving a high likelihood of obtaining a card that wins if the Artiste is caught. They do this near to the end of the game, ideally after the victims final turn, allowing them to state what they’ve done to the manager and claim an easy win. In theory an Artiste could defend themselves by discrediting the person who did the swap, but in practice this proves difficult as the possible motivations for falsely making this claim are tenuous and may be disprovable by other players who’ve seen either of their cards in the past.
The Manager Grab: A player with a strong investigative role spends their first two actions finding information and then uses their third action to switch roles with the manager. Usually the manager needs to depend on information from other players as their own power doesn’t provide any, this allows a player to sidestep the limitation and make decisions based more on their own information than on the results of the discussion.
Switching Roleclaim: A traditional mass roleclaim is ineffective as players do not have a full list of roles that are in a particular game (They could know the ~45ish roles that exist but not which 3-9 are in use). However by encouraging everyone to switch as often as possible it’s possible for a few players to coordinate well enough to build a list and then insist that everyone indicate their role. The Artiste can try to match someone else’s claim but the game state can render this impossible if other cards that they might have will have been seen by more than one player.
These problems are all mitigated by a stricter sequence, mandating that each players actions will be power-switch-power in that sequence helps with all of them. There being an extra power use following any switch limits the effectiveness of the first two options and the switching roleclaim does less well if there’s only one switching opportunity. Roles with alternative victory objectives also mitigate to some extent, as a traitor player who wants the aritste to win means that it’s possible for the artiste to make a false accusation and be backed up – given the possibility of that it’s easier for the artiste to defend themselves against the sort of supported claim that can make it hard to escape some of the scenarios above.
There’s still some debate over whether the game would benefit from some duplicate roles to give a “default lie” for new artiste players, in the same way that the normal villagers suggest a course of action for new werewolf players. On the one hand it could add something to the game, but on the other hand playtest groups of newbies are still enjoying the game as it is and more vanilla cards would take away from the potential re-playability that having lots of different roles provides.