• Welcome to ZD Forums! You must create an account and log in to see and participate in the Shoutbox chat on this main index page.

Game Thread Pendio Mafia XI

Ragnarokio

AVATAR NOT BY JIMMU
Joined
Jan 19, 2018
Gender
If you don't identify as the default options of Male/Female, you may enter your gender here.
Mafia451 has a "Mafia Academy" where people train to be better at reading. They use the Briar Score to keep track of their reads: https://forum.mafia451.com/t/how-to-measure-your-read-accuracy/4836

I know quite a few people who do this, and some of their scores are pretty impressive. Granted, I have some issues with the way accuracy is measured here, but that is a discussion for another time.

i'm more interested in what the samples are like. Scoring an individual readslist is easy (although i would probably have used a different method) but it won't be outside the range of realistic random variation on its own.

There are two big obvious concern with measuring a mafia players reads. The first would be concretely determinining concretely what a players reads are. Do the players you know who track their scores only evaluate unambiguous readslists or do they also score less ambiguous reads? (ie: maybe they take themselves saying "i don't want to lynch x today" as meaning they read x as town that day.) The second is selection bias. How do they choose which read lists to use when factoring in their career score and additionally how do they choose when to post reads lists?

There are some other factors which could conceivably affect the result to make a person much better at guessing than a random number generator, such as factoring role-info into reads (ie: if a person claims to be a doc and targeted a player and there was no nightkill you'll be reasonable in guessing that player is more likely to be town). This is still probably "reading" but it borders on solving via mechanical info and wouldn't translate into a vanilla game.
 
Joined
Sep 27, 2021
@Ragnarokio based on the link you assign a score from 0-1 to each other player (aka a percentage chance they are town) and use that to calculate the score

do this each day and you can track evolution over the course of a game

do this each game and you can track evolution over time
 
Joined
Jul 7, 2021
Gender
Male
i'm more interested in what the samples are like. Scoring an individual readslist is easy (although i would probably have used a different method) but it won't be outside the range of realistic random variation on its own.

There are two big obvious concern with measuring a mafia players reads. The first would be concretely determinining concretely what a players reads are. Do the players you know who track their scores only evaluate unambiguous readslists or do they also score less ambiguous reads? (ie: maybe they take themselves saying "i don't want to lynch x today" as meaning they read x as town that day.) The second is selection bias. How do they choose which read lists to use when factoring in their career score and additionally how do they choose when to post reads lists?

There are some other factors which could conceivably affect the result to make a person much better at guessing than a random number generator, such as factoring role-info into reads (ie: if a person claims to be a doc and targeted a player and there was no nightkill you'll be reasonable in guessing that player is more likely to be town). This is still probably "reading" but it borders on solving via mechanical info and wouldn't translate into a vanilla game.

Essentially what Fig said, but yeah when people are attempting to accurately track their reads with the method the use at the Mafia451 Academy, they have to give all living players a score between 0 and 100 (as percent chance they're town) every game day, and then at the end of the game when they know who is mafia, they evaluate their reads every day and add that data to their other games. Of course, when they rand mafia, all this data is ignored:P
 

Ragnarokio

AVATAR NOT BY JIMMU
Joined
Jan 19, 2018
Gender
If you don't identify as the default options of Male/Female, you may enter your gender here.
Yeah if you do it every day every game then that would eliminate most selection bias. Variances within days would create some potential bias but the data would still be really interesting. Unless the player played a lot of vanilla You'd easily get statistically significant results because there are times when a player is mechanically confirmed as town or non-town but if the game records are easily available it could still be studied.
 

Ragnarokio

AVATAR NOT BY JIMMU
Joined
Jan 19, 2018
Gender
If you don't identify as the default options of Male/Female, you may enter your gender here.
and if you know anyone who is tracking their data with anywhere in the vicinity of that much rigor and also believes their result is significantly above what would emerge from randomness i'd be interested in looking at their stats and reading over their games
 

Ragnarokio

AVATAR NOT BY JIMMU
Joined
Jan 19, 2018
Gender
If you don't identify as the default options of Male/Female, you may enter your gender here.
the idea is interesting though, i'll probably start posting weighted readslists at the start of every day
 
Joined
Jul 7, 2021
Gender
Male
and if you know anyone who is tracking their data with anywhere in the vicinity of that much rigor and also believes their result is significantly above what would emerge from randomness i'd be interested in looking at their stats and reading over their games
If you want to see some actual data, I would ask Ellibereth (he runs the Mafia451 Academy).
 
Joined
Sep 27, 2021
it kinda seems like the site hasn't had any new posts in a few days

so I'm not sure if it's just dead?
 

Users who are viewing this thread

Top Bottom