Hidden Information in Secret Hitler
(This is essay is part one in a series. You can read part two here.)
As the wonkiest designer of the Secret Hitler team (I also write about philosophy semi-professionally as Philosophy Bro), I want to answer the question, “Holy shit, don’t we have enough social deduction games?” by talking about what I think is the single biggest contribution Secret Hitler makes to the genre: the policy deck.
Both puzzle and not
Secret Hitler is inspired by games like Werewolf and Resistance /Avalon, but those games can rely too much on the “social” aspect of social deduction games, and not enough on the “deduction” aspect.
We tried to solve that problem by making Secret Hitler feel like a puzzle that you could always almost solve if you pay close attention — but not quite. During the game, we make lots of information available, but you have to rely on social cues to sort it out from misinformation.
In this essay, I’ll talk about how we invented a new mechanic (the policy deck) to parcel out information to some of the players some of the time, and how that makes Secret Hitler feel as much like a collaborative puzzle as a social deduction exercise.
What is social deduction?
A brief overview: games readily defeat strict categorization, but here are the important commonalities to the social deduction games I’ll discuss:
- Players are assigned to a team in secret.
- There’s a Good Team, which has a majority, and a Bad Team, which has more information.
- The Bad Team must hurt the Good Team in secret.
- The Good Team must figure out who is on the Bad Team to prevent further interference, but they have limited information, which forces them to rely on social cues.
The asymmetry of more people vs. more information is, for me, the engine that makes social deduction games go. They are, at their core, (and this gets at the inherently political nature of the genre) about a well-coordinated minority convincing less-informed members of the majority to vote against their own interests. The most compelling mechanics in social deduction games — think of the Seer in Werewolf or Merlin in Avalon — are the ones that balance or change the flow of information in interesting ways.
How does Secret Hitler control the flow of information?
Secret Hitler uses a randomized policy deck. Here’s how it works:
- Each round, the table elects a President and a Chancellor to pass a policy.
- The President draws three cards from the policy deck.
- The President discards one policy face-down and passes the remaining two to the Chancellor.
- The Chancellor discards one policy face-down and enacts the remaining policy face up.
There are no guarantees about what the President will draw; it could be any combination from three fascist policies to three liberal policies, but at the start of the game the deck contains almost twice as many fascist policies as liberal policies. If a fascist policy is enacted (that is, if the Bad Team gets closer to its goal), then there are three possible explanations:
- The President drew three fascist policies (the technical term for this is “getting deck-fucked”).
- The President discarded a liberal policy and gave the Chancellor no choice but to enact a fascist policy.
- The President gave the Chancellor a liberal and a fascist policy, and the Chancellor enacted a fascist policy.
If a fascist policy gets passed, a fascist President can always blame the deck or the Chancellor, and a fascist Chancellor can always blame the President.
Sometimes entirely liberal pairs will be forced to pass a fascist policy. That means that even if you identify a fascist, you can’t work backwards very far; there’s always a chance that somewhere along the line, a bad draw happened to a good player.
By designing the game around one-on-one interactions that include some element of chance, we give individual players very specific information about whomever they’re entangled with, but we force them to convince the rest of the table. We also allow players to combine information in interesting ways.
Consider one memorable playtest: Max (President) and Andrew (Chancellor) passed a fascist policy, and disagreement erupted. Max said he passed a liberal policy to Andrew and Andrew discarded it, but Andrew said Max passed him two fascist policies.
Some players at the table (Katie and Bobby) had already been on governments that passed liberal policies with Max. They took his side, accusing Andrew of being a fascist. On the other hand, Trin had investigated Andrew, looked at his card, and claimed that he was a liberal.
Since I had been on a government with Trin, I believed her. But suddenly, based on that one interaction, I had information about three other people.
So if A and B are in a dispute, which one you decide to trust has implications well beyond that one pair. If you trust A, that looks really good for C and D and really bad for E and F. But maybe while B looks bad, E and F look really good. Is B’s statistically unlikely story actually true, or are E and F just very deep-cover fascists? The information is all related somehow, but there’s still an element of trust at the center of the web.
In our example game, if I believe Max is a fascist, all of the people entangled with him are suspect as well; if I want to believe Katie, I also have to question what I know about Trin.
How is this an improvement over existing social deduction games?
Existing social deduction games rely too much on “social” and not enough on “deduction.”
Werewolf/Mafia hides information from the Good Team by allowing the Bad Team to act at night; the Good Team learns almost nothing except what they can get from social cues. That purity — social cues rule everything around me — is probably what makes Werewolf such an enduring game. It explains why some people prefer Werewolf to The Resistance, which I think is a tremendous leap forward in game design. But also, holy shit is it exhausting.
In The Resistance (Don Eskridge’s The Resistance for long, just Resistance for short), information is hidden in increasingly large groups: the Bad Team acts as part of a team of (for example) four players, and we only see the results of that team’s actions. If something goes wrong, you know that one of these four people must be on the Bad Team, and it’s the Good Team’s job to use social cues to decide which one it is.
There is also no record of what anyone did. Even players who went on the mission don’t know anything except what they personally did.
Resistance’s solution to player elimination is truly genius game design, but I think there are two ways Resistance doesn’t go far enough:
First, the information you do get comes in vague clouds that are impossible to parse any farther. In early missions, one of three people betrays you, and it could be any one of those three people. In later missions one of four people has betrayed you, and it almost always could be any one of those four people.
After the mission is done, that information is destroyed. Even if you were on the mission and know you didn’t do it, you still have to decide among two to three possible traitors. If Werewolf’s problem is “who is best at yelling?” then Resistance only improves it to “who in this subset of players is best at yelling?”
Second, information doesn’t accumulate in a way that gives players any insight. Even if you think you know who failed the three-person mission, that tells you very little about the four-person mission. After three missions, you have three different puzzles with minimally-overlapping solutions.
Essentially, in Resistance, knowing two things doesn’t get you a third thing. There’s no reward for being right once; you have to be right about everything independently. So Resistance is still like 80% driven by reading people, and as someone who would rather be reading books, that has always felt like too much for me.
How will this affect my play experience?
First, Secret Hitler has greater replayability, because there are so many possible combinations not just of player roles, but also of the deck order.
But more importantly, I think the policy deck solves the biggest problem with the social deduction games that come before it: the “who is the loudest” problem. By fragmenting information instead of simply destroying it, we give members of the Good Team several entry points into the puzzle without allowing them to ignore social cues altogether. A player would have to shout down not just the person they’re entangled with, but also the actions of everyone else further down the web; this greatly reduces the advantage shouty players have, and puts power back into the hands of the attentive and clever — we restore the balance between “social” and “deduction.”
Later, I hope to talk about more specifics in the process of designing it, and to give you some insight into why that design fits so well with the other mechanics, in ways that surprised even us.