Trust tells

by Collin Lysford


I played the social deduction game Mafia (also commonly known as Werewolf) competitively off-and-on for a decade or so. Traditionally, Mafia is a short and sweet party game. Each game day is perhaps ten minutes long, players brashly accuse each other so they can look deep in each other’s eyes and try to spot the liar from their reaction. But I played by forum, where a game day might last three real-life weeks, with more the air of a courtroom drama as players laboriously hunted through the transcript thus far looking for contradictions. (You were, of course, forbidden from editing your posts.)

There was a time on this forum where a few players would sometimes post in red and allege that the words in red text were incontrovertibly true, a reference to the murder mystery visual novel series Umineko When They Cry. The games they were playing weren’t themed games based on Umineko When They Cry, and the moderator wasn’t enforcing the truth of the red text. So this might seem to be a pointless exercise, the equivalent of a player saying I pinky-swear I’m telling the truth!” even as they lie their pants off. That’s not what happened with the red text of truth, though. The players themselves ensured that they were telling the truth each and every time they used the red text.

From a surrogation lens, you might think that eventually the pressure to betray the signal would be too strong, and it would partially lose it’s predictive power as some red texters pulled off high-impact lies. Instead, though, the moderators banned the red text of truth by fiat. In fact, they actually generalized their ban to trust tells”, any sort of signal of truthfulness that you build up over multiple games such that you can point to your past history and say When I do this, it’s always truthful.” In a game about lying, why would external authorities need to punish you for telling the truth in a transparent way?


In Board games are a social construct, Neil wrote:

But there’s an even larger problem. It’s not just that winning at all costs might be a dick move. Sometimes people don’t agree what winning is.

An individual game of Mafia makes it objectively clear which players won and which players didn’t. But however well-defined the winners of a game of Mafia are, the games are all embedded in the broader context of the site itself. What does it mean to succeed as a Mafia player?

At first blush, it may seem like the answer is simply win as high a proportion of games as possible”. But it’s different to win as town vs. mafia. The town is an uninformed majority, having the numbers to win the game but lacking the information to use their numbers effectively. The mafia is an informed minority, with knowledge of the team they need to keep alive to win, but lacking the numbers to fully control the machinery of the town. Because townies are a majority, you’re more likely to play as a townie than as a mafioso in any given game. Anything that increases your townie winrate at the expense of your mafia winrate will increase your site winrate overall.

As a thought experiment, imagine a player who truthfully claims their alignment in their first post. If this made you win every town game and lose every mafia game - well, your teammates would be awfully cheesed off at you for those mafia games, but your site winrate would be extremely high overall. To prevent this, all mafia games have a strict rule that you must play to your win condition. If you’re a mafioso, then too bad about your site-wide win rate - you have to try your hardest to win this particular game, here and now, and you’ll be banned if you truthfully claim that you’re in the mafia.

Now we can see the problem with trust tells: they’re a subtle cousin of claiming your alignment. It’s not that you’re hard forfeiting the games where you play as a member of the mafia. But you get an extra little boost to your town games by having the ace of truthful statements up your sleeve, at the extent of an extra little drag on your mafia games (”They’re someone who uses trust tells, so the fact they haven’t done one this game is pretty suspicious…”). While it’s a lot less egregious than truthfully claiming your alignment, it still crosses the magic circle of each game being an independent, self-contained thing. If you’re someone who uses trust tells, your goal is not 100% winning this particular game - some of those percentage points get diverted to the iterated game, where you want to point at your history of honoring trust tells in future games, even if it lowers your chances to win this particular game. That reputational concern is at odds with the idea of playing to win condition, and that’s why moderators need to stop it.


But really, what’s the point of playing Mafia? Everyone has their own answer, but for most people, a large part is that it’s fun to work our muscles for deception and detecting deceit in a consequence-free zone. It’s interesting to learn which of your friends are good liars in a context where those lies were for a temporary purpose instead of a long-term breach of your trust.

Players using trust tells may seem to be playing Mafia at a higher level” than those naively focused on a single game, since they’re considering their whole winrate and not just the individual game at hand. But what happens if every player on the site is like that? Well, each game would last about ten seconds, as every player truthfully claims their role and town always wins. The best player on the site would be whoever happened to roll town the most consistently, and then everyone would get bored pretty quick and leave.

The lesson to be learned here is that the individual games are not inherently independent. Games are embedded in contexts just like everything else, and that context may ensure that one game has a different relationship to the win condition than the next. The magic circle separating each game from the next is aspirational: it exists not as a straightforward implication of game design, but because the community wishes it to be so, and agrees to play under the thumb of moderators who prohibit behaviors that erode it.

So - when you see some sort of iterated game, remember that you shouldn’t assume that each game is an isolated contest unto itself. Instead, ask yourself: what work is being done to keep the iterations independent, and how well does it seem to be working?

Thanks to reformed trust teller Terry T. for filling in some of my gaps on the history.

Links to this post
100th post anniversary