General Mafia Theory Discussion

  1. I merged your posts. Testing that feature out. Hope you don’t mind. In general I think it’s probably healthy for obviously mergeable cases like this?

  2. Disagreement with the absolutes in the above aside, isn’t it probably better to be fake-claiming investigative power as VT over non-VT given that the expected penalty for incorrectness right now is death?

Smite bad players and commend good ones. This isn’t limited to “lying” or any specific behavior. It works on a general scale as well.

1 Like

The issue with this is that Mafia is a probabilistic game. Plays that are usually very good can lead to very bad results and often times you can’t even really blame someone for going for a 50/50 in many situations. Exclusively analyzing and reacting based on the result gets mob-like very fast and in practice I don’t think it really helps anyone.

If someone is consistently making worse plays over a long period of time they’ll eventually have a lower win rate. I think at that point if some’s into improving they can look into/reach out to others for what’s going wrong. If they’re not - they just belong in a different demographic than the players who do care about every player playing as optimally as possible.

In terms of site relevance, we’ll be experimenting with enforcing competitive vs. casual (perhaps worded differently) after our first growth push. It doesn’t make sense with our low player base atm.

So is finance and banking, but CEOs and bosses there also analyze performance based on results.

I do believe that examining a players track record is more accurate than the results of a single game, yes. Also, reputation matters a lot in mafia, since it is a community game. However, not analyzing based on results at all would make metrics pretty difficult.

I think it is not accurate to judge an individual player through win rate because of many factors: The proficiency of the scum team, The proficiency of other town, the fact that a better town player will always be night killed earlier than weaker townies and therefore have less time to make an impact on the game, low sample size due to the length of mafia game etc.

A weaker player playing on Mafia451 would invariably have a higher win rate than a stronger player who plays on MU or MS.net, because, admittedly, Town winrates over here are significantly higher than those other sites (as of 2018).

  1. Mafia is a game, not business!

  2. They’re analyzed based on their decisions/results. However “This was a 90/10 situation and the 10 percent happened” is the analysis, not “you were wrong”.

  3. Proficiency ones balance out over sample size. I think the “better town” being night killed thing is a fallacy: if you fail to make a sufficient impact in the time you’re alive you’re not better town. There’s probably a fairly elegant “weighted average based on results when alive” that can measure this. But yes - I do agree it all boils back to sample size. Since most people don’t have a large enough sample, it makes even more sense to not vilify people too much.

  4. Also true. The nuance here is that even within sites there are “mafia games” that in reality are completely different games (MU turbos vs. MS’s long opens vs. Nanook’s lovely setups here, etc.).

We’re working on making this all better! A few standardized setups with high replay value that are conducive to finishing relatively fast would help us get a lot better at judging things within that specific realm of mafia.

But I think until that exists, in this fuzzy hard to quantize mafia world we live in, it pays and makes more sense to be a bit nicer and less absolute on what’s good/bad since we all really don’t know what we’re talking about. The only thing we can talk about with absolute confidence are certain GTO role coordinations in specific open/semi-open setups. Commonly held believes have been getting flipped on their head for years. :stuck_out_tongue:

Yes, I agree with your conclusion. Though, question for your second point: How can you arrive at a probabilistic analysis of the situation, if one has not played enough games with that specific situation to even obtain a correct set of statistics and probabilities? Simply: How did you arrive at a number of 90/10, if 100 games with that situation have not been played?

Unless one is an expert at mafia (in this case, you are, but a significant majority of players are not), one is not able to accurately and correct analyze the win potential of a play, simply put, “they are in no position to judge”. It’s like an ordinary person making speculations about the stock market (I’m sorry for the business references!)

With this in mind, “This is a 90/10 situation and the 10 percent happened”, coming from any ordinary player, is more of being comforting and nice and polite, but likely not indicative of reality.

Hence, most players aren’t necessarily wrong to judge based on results, since it is simply a representation of the facts.

I probably need a tldr; on this topic.

  • Sometimes fake-claims are good. Hidden element often involves real life constraints.
  • Sample size matters. It’s hard to discover and draw conclusions without it. Can’t gather sample size on plays without executing them.
  • It’s important to have distinct casual/competitive games so people can have the freedom to test their ideas. Hyper-competitive results based environments and experimental/creative environments should both be encouraged.
  • If we want a version of mafia that’s more well-suited for the measuring tools we have, the rules themselves need to be modified.
1 Like

That’s a good question.

The way I look at is that most people are still gathering their sample. When people comment on their experiences and results to each other they’re kinda merging together their experiences into one larger sample.

I know it’s hard - but I think the attitude is often a lot more about justifying that someone’s right rather than working together to figure out what’s going on. The former is what leads to a lot of sweeping generalizations and negativity. I know the mental gymnastics involved in switching from an in-game mindset to an out-of-game mindset is a big part of what makes it hard.

On a personal level, a lot of the conclusions I’ve drawn about the game are based around and would not have existed without years and years of repeated failed experiments. (And the failing experiments are still happening!) I’m thankful for the people who dealt with me when I was figuring things out because my results were disastrous!

1 Like

You also have to take into account that other players do not know X’s alignment when they are performing said experiment, making the mental gymnastics even more difficult.

Ah yeah - I’m mainly talking about how people discuss things out of game. In game is a completely different story given how many hidden motivations every action could have.

How do you guys think we should organize this topic?
Keep a running directory of topics in the OP?
When a topic gets sufficiently enough discussion to move the posts out to its own topic?

yes.

I’ve found the reception to fakeclaiming as town to be largely a cultural norm.

There is a reason that the game has “lynch all liars” as a meme. As a general base set the game involves an informed minority verses an uninformed majority. Naturally, each group may be somewhat paranoid for differing reasons. As the majority town is generally uninformed of particular alignments the building of appropriate trust can be fundamental toward resolving paranoia and solving the game. Fakeclaiming in that alignment is understandably problematic, yet with mitigating factors, such as game setup, player meta, and cultural expectation.

For example, some mafia cultures regularly employ a claim to cop/seer cover, where several to most players give fake views in an attempt to allow a true cop/seer to post actual views without immediate outing themselves to the mafia. It’s a cultural thing where one has to have several players at least guess somewhat correctly to give adequate cover. In my first tourney game at twoplustwo we had half the field quite believably fakeclaim cop/seer, including 4 town and 2 mafia, with even a non-claimant (me) nightkilled early by mafia in being thought cop/seer. We had 3 scum in that game, with the final 2 giving up as caught.

However, I’ve seen town fakeclaiming backfire horribly too, unwittingly outing not one but several town PRs, even breaking a game more than once due to conflating factors. Interestingly enough, these have usually come from players known to engage town fakeclaiming as a strategy. The key there was often either an unawareness in players of knowing that player meta or of being receptive to lying as acceptable town behaviour.

Thereto, there is also sometimes a clash between individual player strategies and/or player/cultural norms on the innate value of PRs or even of mass-claiming as town strategy. Some may fakeclaim either a PR or a VT in hopes of furthering an individual strategem they believe is ‘best’ for town, regardless of a town consensus otherwise. Evenso, these are often not a matter of if but of when accurate roles ought best to be claimed. When there is a clash between town player expectations, and a town player fakeclaims in support of what they think is advantageous over against the impetus of most other players in mass-claiming, it often defeats the main aspect of the mass-claim for town, esp. particular to figuring game balance and mechanical structure. But part of the game too is in getting all of town on the same page in either strategy agreement or compromise of such toward unity.

And then there are jesters.

I’d agree that “drawing the line” can be somewhat circumstantial, with the caveat that honesty is generally the best policy for town.

Hi, I have a question about metareading: how to do that? I heard Mr Elli makes people’s profiles but Im not really sure how to do that

1 Like

There are players who keep detailed spreadsheets on all manner of things, sometimes including every game they’ve ever played. I’m sure it can be quite effective, but it’s a lot of work.

Some, maybe even most or all players get into patterns of behaviour. Often those behaviours can be alignment indicative for them personally. I’ve variously been told I’ve had several of those over the years, but I’ve also been told I’m difficult to read. I think it’s likely for consciously trying to mix up my patterns, and maybe for usually seeming very scummy as town and towny as scum, which in itself is a pattern. Some patterns can be consciously swung back-and-fourth - so if I’m seen as jokey and lighthearted as town then I mimic that as scum; or if I’m read as scum for being mean or gaslighting as scum then I consciously do that as town. However, many players actually prefer to have obvious personal town tells they rarely use as scum.

Most players have a preference of alignment and so play accordingly. That’s usually a big meta read, as enjoyment can be difficult to sincerely fake. But often individual meta reads are a matter of how a particular person differentiates playing different roles, and so establishing certain tells for future considerations.

To the in-topic directory or the parsing out independent topics?
If we parse them out the category basically becomes the directory yeah?

@GTacc I’ll get to you in a bit, meta’s a topic dear to my heart.

1 Like

And a catch-all for topics not of strong enough interest to merit stand-alone independence?

But, yeah, I can see where one or the other may be ideal. I think having a general theory hodgepodge milieu is welcome, but also topicfying elements of theory discussion worthy of stand-alone status.

You’ll have to either forgive or bear with my personal tendencies to meander and hailing from home boards noteworthy of meandering tangentially in tumbling windswept unparseable trippinesses.

I’m just now thinking of calling it mafia mushroom theory.

Whoops - forgot about this - sorry for taking so long. I was considering recording a video about this.

I break down meta into two subcategories:

  1. Hard Data
  2. Personality Analysis

Hard Data includes things like word choice, post structure, post times, relative post times, post density, etc. In other words, things you can scrape and store from a set of posts. Analyzing Hard Data is mainly about looking for patterns for how things relate to alignment. Keep in mind these patterns tend to be probabilistic.

Personality Analysis is fuzzier - you’re answering “What sort of person are they?” You’re judging things like their age, emotional maturity, level of education, etc. For a lot of people, you have to make some educated guesses based on what they’ve revealed about themselves and how they post - it certainly helps to have archetypes to refer to so you can compare expectation. Having a good personality profile on a player helps you figure out what behaviours are deviations/disingenuous/unexpected.

I go very deep on both, working programmatically for the former and spending a lot of time researching the latter. This isn’t practical for most people! I was developing less time intensive techniques for myself (which are probably going to be delayed because of my incoming foray to video mafia), but I’ll outline some ways a player can apply these ideas in practice.

  1. Sample Size is important. Hard data really only works when someone has a huge amount of games and posts. Seeing some pattern over 3 games is likely meaningless.
  2. Notice quirks in someone’s word choice. Search for patterns for those specifically. 99% of the time it leads to nothing but it takes very little time and it gets you into the habit of looking for things.
  3. Research people’s activity/relative game activity. This is some of the least effort-intensive data to look up and see if there are obvious patterns in. Search is ready-made to help you look at things this way.
  4. Check other people’s meta-work! Don’t just take their word for it. Usually it’s not very deep. Oftentimes you’ll find that the pattern they claimed they found doesn’t actually exist.
  5. The core lesson from personality profiles is that the baseline expectations you should have for a player should be based on them - not based on you. Understanding how someone thinks and reacts to things is the core of “normal” read-based mafia. People tend to not have good reasons for forming the set of expectations they have for most players. I tend to think about people I know in real life that are a similar profile to whatever target I’m analyzing and use that person as a model. A helpful bonus is that people often freely talk about their life and background on forums. These are not things you cite, but things you should think about on your own when analyzing. Just the act of reminding yourself about what a player’s personality profile is like before thinking about what the set of their expected reactions is will already help most players a lot.
  6. Remember that you don’t need to be in a game to be practising these things! You can try and research or profile any player that catches your fancy. As a matter of fact, most of the work I do is when I’m not in any game at all. These are things where it’s easy to “check your work”. When you’re reading games - guess the response a particular player would have to a statement if they were town or if they were mafia and compare your guess to their actual responses. Likewise, if you think you noticed some greater than random hard data pattern - check as many games of the player as you can to verify it.
  7. When you’re just starting, a lot of the stuff you’re going to find is going to be >rand, but barely so. Let the two types of meta form your read together, along with whatever other effective tools you might have. Meta is powerful because it’s reusable, and if the target doesn’t know the content of your work - hard to mitigate. Don’t be shy about “making up” reasons or doing whatever finagling to get your scumread lynch. Meta techniques are about getting correct reads. The act of them getting lynched is something different altogether (and often involve various degrees of lying/deception).

As a final note - I obviously (and demonstrably) find meta very effective. Make sure to rigorously inspect all your “tools”, meta or not. Is whatever tell actually better than random? Is whatever technique actually getting me results better than average? When I did this I found that most things just didn’t work so I had to keep experimenting! There’s definitely a lot that hasn’t been discovered yet.

2 Likes

Wow thanks a lot, Ill try this now

I’ve seen a lot of talk about what to do when you’re at various, usually extreme, positions on the read accuracy by charisma plane. This 2 axis view point is bit too simplistic: for instance, there’s also a time axis in every game, and reads and charisma aren’t independent. (We’ll ignore the former by considering frozen snapshots). Regardless, we’ll talk about it a bit here.

Some definitions and conditions:

  • I think people tend to have many slightly different definitions for “Charisma” in a mafia context. I define it as the ability to make others act in some intended way.
  • If someone has demonstrable record of sufficiently high read accuracy - said record can be cited. Convincing others isn’t something particularly interesting in this scenario. We’ll mainly talk about scenarios where this doesn’t apply. In other words there is no player that most other players would just sheep.
  • What does it mean for someone’s accuracy (let’s call them Tom) to be sufficiently high? Roughly, if from the point of view of any other player, the percent chance that Tom is town times Tom’s expected accuracy is at least perceived to be higher than the expected accuracy of that player when they’re town, Tom’s accuracy is sufficiently high.

And some basic practical advice . In the following, assume that the game is in a state where no player (or set of players) should be sheeped. Identifying when you are in such a game state is a skill, and a topic for another day. Obviously these are general rules - there are always exceptions.

If you’re relatively charismatic and have low expected read accuracy:

  • Try to sculpt the flow of the game conversation to get some desired distribution of thoughts/ideas/contribution from the other players. Figuring out the best desired distribution is also a topic for another day, but for most situations/levels, going for a uniform distribution (every player has the same amount) is good enough (but not best because there usually exists some set of players that you’d rather have more or less from given the expected accuracy of others).
  • Facilitate clear communication, good organization, and make sure ideas are being understood. Examples of this can include proving clear, readable summaries or breakdowns (don’t make them too long!) and potentially going out of your way to clear up misunderstandings. You’re trying to move the game state to one where there is an obvious path to take. Doing the above makes it more likely to happen for the vast majority of player-lists. Someone may complain that you made it less likely for them to get some read/ruined some technique - but in fuzzy game states it’s usually better to cater to the majority of player’s and not a particular one.

If you have you’re relatively uncharismatic and have okay expected read accuracy:

  • Work on improving your charisma throughout the game! You’re getting live feedback on how convincing you are. Every time you’re insufficiently convincing try to present your idea in a different way the next time. It’s okay to experiment to find what works. A lot of charisma is about knowing your audience - understand that the purpose of your posts is NOT to make the idea clear to you, but to a specific set of others.
  • Think about the future. You aren’t necessarily insufficiently charismatic for the entire game, make well-positioned clear notes aimed at the future. It’s likely the more convincing people will get their way now, so you should set your groundwork for future days (whether you’re still alive or dead). Consider and analyze various future game states, and try to make your posts sufficiently memorable/readable that people will look back on them if you’re not around to cite them later.
1 Like