An article with a proof on optimal mafia play.

[Psyche]

i'm sure some people do
just maybe not 43 people

[AlwaysInnocent]

I'm not actually sure that that is true. I think the real reason is that no one want's to play a 43 player game.

That is probably the main reason, but do you mean that it is not actually true that innocents being able to identify mafia better than chance is the reason for not doing 40:3, or that innocents cannot identify mafia better than chance in practice?

[Ythan]

Think the large normal queue'd accept a mountainous game of undetermined size, with scum ratio determined by a formula, and accepting as many entrants care to join?

[Ythan]

Because if so I'll totally enable this train wreck to see how it works in practice.

[chamber]

I'm not actually sure that that is true. I think the real reason is that no one want's to play a 43 player game.

That is probably the main reason, but do you mean that it is not actually true that innocents being able to identify mafia better than chance is the reason for not doing 40:3, or that innocents cannot identify mafia better than chance in practice?

Day 1 numbers support it being true for day 1. And then power roles get in the way of any meaningful sample size. But I think the nightkill in mountainous is extremely powerful, so I'm not sure that a town can lynch better than random in mountainous games.

[Ythan]

Nightless

mountainous still best setup.

[AlwaysInnocent]

I'm not actually sure that that is true. I think the real reason is that no one want's to play a 43 player game.

That is probably the main reason, but do you mean that it is not actually true that innocents being able to identify mafia better than chance is the reason for not doing 40:3, or that innocents cannot identify mafia better than chance in practice?

Day 1 numbers support it being true for day 1. And then power roles get in the way of any meaningful sample size. But I think the nightkill in mountainous is extremely powerful, so I'm not sure that a town can lynch better than random in mountainous games.

Yes, with the exception of day 1, of course. According to my own simulations, there is a difference. I cannot exclude any mistakes, but so far this seems to be an actual significant result:



Statistics:

Sample size: 1,000
P: 7
m: 2

Explanation of symbols:

u_m = average utility for mafia (i.e., winning rate)
u_i = average utility for innocents (i.e., winning rate)
R4 = the game includes nights (so not-R4 means nightless games)

Strategies:
random = self-explanatory
informed = mafia only vote for innocents
analytic = innocents use some AI algorithm to guess wolves (which simply assumes that wolves play informed rather than random)
mixed = mafia play random β of the time and play informed (1 - β) of the time

I have tried to include other strategies, but they do not really seem to diverge much from these basic strategies.

P.S. This was before I started reading about MafiaScum terminology.

[LicketyQuickety]

Why has this thread suddenly exploded???

[Ythan]

Is there a better possible answer than rtft?

[Psyche]

almost definitely

[Ythan]

Let's see what you got.

[LicketyQuickety]

People are not robots who are incapable of picking up cues of personality and/or mood and much can be known to some extent of the guilt of a guilty party. That said, reading behavior does not as often get people lynched because of peoples need for tangible concrete evidence.

[AlwaysInnocent]

People are not robots who are incapable of picking up cues of personality and/or mood and much can be known to some extent of the guilt of a guilty party. That said, reading behavior does not as often get people lynched because of peoples need for tangible concrete evidence.

True. But even without communication a lot can be said about the voting behavior! And even robots should be able to do so.

[callforjudgement]

40:3 would be terrible because after like 10 day/night cycles, the Mafia would have killed off all the town players who either a) have any scumhunting skill, or b) cause any sort of activity/interest/fun in the game. The result would be that the town would be completely demoralized (by definition, they're nearly always going to lynch townies), and the scum would be on edge and not enjoying things much either (because they're hugely hurt by even a single lynch on scum, meaning that they need to get something like 10-15 mislynches in a row before they can start to feel like they're actually winning instead of coasting).

2:11 has similar problems (although not as bad as they would be in 40:3). For what it's worth, I believe the best town strategy in 2:11 is to policy lynch on day 1 (and possibly day 2) regardless of who you actually think is scum, aiming to eliminate players with a weak towngame. You probably wouldn't hit scum anyway, a mislynch that early in a theoretically balanced mountainous hardly hurts town, and you badly need to drive the townquality of the playerlist up because you know that the scum will be trying to drive it down. (Plus you have the same odds of hitting scum by chance as you would lynching anyone else.)

[LicketyQuickety]

People are not robots who are incapable of picking up cues of personality and/or mood and much can be known to some extent of the guilt of a guilty party. That said, reading behavior does not as often get people lynched because of peoples need for tangible concrete evidence.

True. But even without communication a lot can be said about the voting behavior! And even robots should be able to do so.

I agree. My only comment is that paying attention to how people vote may not be as easy as just noticing behavior for some players. I know this is true for me. Unless the VCA is laid out in one post I usually never even bother to consciously pay attention to how people vote alone. The reason VCA works is because people are not currently thinking about manipulation info for VCA.

[AlwaysInnocent]

People are not robots who are incapable of picking up cues of personality and/or mood and much can be known to some extent of the guilt of a guilty party. That said, reading behavior does not as often get people lynched because of peoples need for tangible concrete evidence.

True. But even without communication a lot can be said about the voting behavior! And even robots should be able to do so.

I agree. My only comment is that paying attention to how people vote may not be as easy as just noticing behavior for some players. I know this is true for me. Unless the VCA is laid out in one post I usually never even bother to consciously pay attention to how people vote alone. The reason VCA works is because people are not currently thinking about manipulation info for VCA.

Damn. People here really use a lot of abbreviations that are assumed to be common knowledge. What does VCA mean exactly? "Vote Count Aggregate"? This time I could not find it on the Wiki.

[Psyche]

i'm not making the wiki article until im sure it's not a waste of time

[LicketyQuickety]

People are not robots who are incapable of picking up cues of personality and/or mood and much can be known to some extent of the guilt of a guilty party. That said, reading behavior does not as often get people lynched because of peoples need for tangible concrete evidence.

True. But even without communication a lot can be said about the voting behavior! And even robots should be able to do so.

I agree. My only comment is that paying attention to how people vote may not be as easy as just noticing behavior for some players. I know this is true for me. Unless the VCA is laid out in one post I usually never even bother to consciously pay attention to how people vote alone. The reason VCA works is because people are not currently thinking about manipulation info for VCA.

Damn. People here really use a lot of abbreviations that are assumed to be common knowledge. What does VCA mean exactly? "Vote Count Aggregate"? This time I could not find it on the Wiki.

VCA stands for Vote Count Analysis. Its basically a way to determine who Scum are late game.

[AlwaysInnocent]

People are not robots who are incapable of picking up cues of personality and/or mood and much can be known to some extent of the guilt of a guilty party. That said, reading behavior does not as often get people lynched because of peoples need for tangible concrete evidence.

True. But even without communication a lot can be said about the voting behavior! And even robots should be able to do so.

I agree. My only comment is that paying attention to how people vote may not be as easy as just noticing behavior for some players. I know this is true for me. Unless the VCA is laid out in one post I usually never even bother to consciously pay attention to how people vote alone. The reason VCA works is because people are not currently thinking about manipulation info for VCA.

Damn. People here really use a lot of abbreviations that are assumed to be common knowledge. What does VCA mean exactly? "Vote Count Aggregate"? This time I could not find it on the Wiki.

VCA stands for Vote Count Analysis. Its basically a way to determine who Scum are late game.

Ah! I was close, though.

The reason VCA works is because people are not currently thinking about manipulation info for VCA.

Also because manipulating VCA often means voting for fellow mafia, therefore making it harder to win. Sometimes it is necessary, but in my experience it is really something mafia do not like to do. According to my experiments, there is a point after which it is no longer beneficial to vote deceptively. Which makes sense because killing mafia to prevent other people from killing mafia cannot work if you do a better job of killing mafia than them.

[AlwaysInnocent]

P.S. By voting I mean voting to

lynch

. Impotent votes are ignored.

[Ankamius]

The only time I managed to make VCA be strong is the one time I was specifically looking for patterns.

[LicketyQuickety]

The only time I managed to make VCA be strong is the one time I was specifically looking for patterns.

This statement is juicy af.

[LicketyQuickety]

Bumpity.