Live Action Mafia
http://mafia.mit.edu/

Investigator Strategy
http://mafia.mit.edu/viewtopic.php?f=190&t=4676
Page 1 of 1

Author:  ksedlar [ Sun May 22, 2016 8:17 pm ]
Post subject:  Investigator Strategy

We should probably wank about this. My current thoughts:
1. Investigators should report results relatively soon after they investigate, only delaying to obscure who would have been able to investigate quickly. No keeping results secret until the next day.
2. Lots of proxy chaining should be done. I think we should aim to have the average chain length be 3 (investigator tells 1 person, who tells 1 more person, who tells 1 more person, who posts the result). Mafia only has one set-a-trap, so they really don't want to waste it on low-probability guesses. However, we should have chains be short enough that if we need to trace back and find the original investigator, we can do it. Off the top of my head, I think we want people to pass on an investigation with probability 75% (including the original investigator). People should also just post the investigation if the chain loops back to them at any point. This randomization process should only be done for the first investigation; after that people should use the same proxy chain every time. It should also be clear which investigations are coming from the same investigator.

Author:  aok [ Sun May 22, 2016 8:31 pm ]
Post subject:  Re: Investigator Strategy

I think that it's even more important to obscure the role info to the mafia. I think investigators should act like kingpins in a normal game; trying to sway debate without revealing.

Author:  ksedlar [ Sun May 22, 2016 8:45 pm ]
Post subject:  Re: Investigator Strategy

I think chaining does a sufficiently good job of obscuring role info (not that chaining can't be done in Resistance; if people could get a random unknown person to give information to, who knew the identity of the person telling them the information, I think a kingpin WOULD want to info chain through this method).

I think that keeping investigations secret and just trying to softly sway things is going to confuse town's information a lot. Especially because a "soft info" type statement might be "I don't think we should lynch X," (where X investigated negative for the kill). But probably the investigator is more suspicious of X than average, because they investigated X in the first place. So they're not communicating their suspicions well in that case. And if someone tries to do any more than softly swaying publicly, the fact that they're an investigator would be MUCH more obvious than if they had just chained.

Author:  Daniel Grazian [ Mon May 23, 2016 12:42 am ]
Post subject:  Re: Investigator Strategy

I'm actually not convinced that investigators should give away their results at all. Set-a-trap is incredibly brutal here, and the evidence we get from investigators is not terribly strong.

If investigators should in fact give their results, then it should definitely be behind a large proxy chain. Nobody should no for sure who the investigator actually is. My concern is that due to the need for extremely heavy obfuscation, the mafia can impersonate an investigator too easily.

Author:  ksedlar [ Mon May 23, 2016 4:32 am ]
Post subject:  Re: Investigator Strategy

The fakeability of investigators seems unlikely to be worse than the fakeability of stalkers. Slightly more fakeable in that both mafia could fake investigator, but slightly less fakeable in that there is a chain, albiet a long one, to go back to the investigator.

At the very least, having weak information from investigators is better than no information. And it is far from useless, IMO. Someone investigated as innocent on 3 kills in a row looks pretty good, given the "If the same mafia member makes three kills in a row, all other mafia die," rule. Investigators can also investigate kills as many times as they want, so we could end up with a kill with say, 5 reasonable suspects, 4 of whom an investigator says didn't do it. And of course, an investigation itself gives Baynesian updating. If some person keeps not showing up as innocent for a kill, then we should lynch them.

I think it's unlikely that mafia would set a trap unless it's the end of the game or they're at least 50% confident (they only have 1). A good proxy chain makes it much less certain than 50% that the person telling you results is an investigator.

Author:  ksedlar [ Mon May 23, 2016 5:17 pm ]
Post subject:  Re: Investigator Strategy

We should also discuss how investigations should be reported.
Method A: X did not kill Y
Method B: I asked about the set {X, Z} for Y's death, and was given the result that X did not kill Y.

Method A gets the most important part of the investigation. However, Method B also allows us better Baynesian updating, and makes it somewhat more likely that Z killed Y. The main downside of Method B is that investigators will need to be even more vigilant about sometimes reporting a set that includes themself (and doing this means they can't report their real investigation), so that mafia can't discover the investigator by process of elimination.

What do people think?

Author:  dylanhen [ Mon May 23, 2016 5:22 pm ]
Post subject:  Re: Investigator Strategy

Getting that X is innocent only gives a 2:1 likelihood ratio for Z making the kill, so it takes about 3 of these to have reason to significantly suspect Z. Since we'll probably never have that many overlapping investigations, I think we should use Method A.

Author:  pravinas [ Mon May 23, 2016 6:13 pm ]
Post subject:  Re: Investigator Strategy

I disagree. Given that townies would have this information, there's no reason whatsoever not to share it, so people can use the Bayesian reasoning thing anyways.

Author:  dylanhen [ Mon May 23, 2016 7:24 pm ]
Post subject:  Re: Investigator Strategy

It makes it easier for mafia to figure out who the investigators are. In particular, investigators have to investigate themselves with some probability to avoid giving themselves away, which makes the investigations less useful.

Author:  ksedlar [ Mon May 23, 2016 10:01 pm ]
Post subject:  Re: Investigator Strategy

The people proxying to Alex should either post that they are the new proxy for [insert Professor Oak / Officer Jenny] or they should pick a new person to be the head of the proxy chain.

Author:  ksedlar [ Mon May 23, 2016 10:32 pm ]
Post subject:  Re: Investigator Strategy

Alex said before he died that Lucy had asked him to proxy for her before she died. Therefore, it could be useful to know if she was part of any chains, or if her request to Alex was independent of the current three claimed investigators. Thus, if Lucy was part of chain, people in chains should pass down the info saying she was. Otherwise she is more likely to have actually been an investigator, which would make our current three "investigators" look sketchier.

Author:  ksedlar [ Mon May 23, 2016 10:33 pm ]
Post subject:  Re: Investigator Strategy

Also, in case it wasn't obvious from him not claiming as such before his lynch, Alex was NOT an investigator.

Author:  ksedlar [ Tue May 24, 2016 10:27 pm ]
Post subject:  Re: Investigator Strategy

Can people push investigation results through the chains more quickly? Hopefully everyone investigated yesterday too. Also, if Sasha was an investigator, that should be posted. Presumably if Sasha was proxying to you and you don't get any day 2 results for a while (like by midway through tomorrow), then they were an investigator.

Matt was not one (he said he didn't have any important information to give to town before he died).

Author:  ksedlar [ Wed May 25, 2016 1:44 pm ]
Post subject:  Re: Investigator Strategy

I would like to see Nurse Joy investigate {Dgrazian, ermain}->sashacf today.

Given last night (and the confidence that Pravina was actually out of town / that the Squares people couldn't have missed one of them being gone), I'm pretty confident that one of the following is true:
a. Matt was mafia and killed Sasha
b. Erin was mafia and killed Sasha, and Professor Oak is mafia
c. Aok and Dgrazian are both mafia, one of them killed Sasha, and Professor Oak is mafia
(In b. Professor Oak is mafia for having lied about ermain being innocent, and in c. Professor Oak is mafia because otherwise aok wouldn't have proposed Dgrazian as the lynch.)

Making a further bet that Nurse Joy and Professor Oak are probably not both mafia (i.e. that we have at least one alive, active investigator), we can get some further cases upon my suggested investigation:

Case 1 (Nurse Joy investigates {Dgrazian, ermain}->sashacf and gets that ermain is innocent):
a. Matt was mafia and killed Sasha
b. Aok and Dgrazian are mafia, Dgrazian killed Sasha, and Professor Oak is mafia
If we get this result and Professor Oak's proxy chain does not start with either aok or Dgrazian, then a. is true.

Case 2 (Nurse Joy investigates {Dgrazian, ermain}->sashacf and gets that ermain is innocent):
a. Matt was mafia and killed Sasha
b. Erin was mafia and killed Sasha, and Professor Oak is mafia
If we get this result, then b seems more likely to be true (Joy's investigation revealing that aok was innocent was presumably on {ermain, aok}, so while ermain could have been unlucky twice, b looks right).

Author:  ksedlar [ Wed May 25, 2016 1:49 pm ]
Post subject:  Re: Investigator Strategy

Wow I totally copy-pasted Case 2 wrong. Case 2 is Nurse Joy investigates {Dgrazian, ermain}->sashacf and gets that Dgrazian is innocent.

Author:  aok [ Wed May 25, 2016 2:03 pm ]
Post subject:  Re: Investigator Strategy

It's not clear to me that Nurse Joy checked ermain vs aok.

Author:  aok [ Wed May 25, 2016 3:03 pm ]
Post subject:  Re: Investigator Strategy

Case 2 happened. I think we want to know whether Joy checked vs ermain or not? I think this leaks nothing about investigator identity to the mafia and can help inform our decisions.

Author:  ksedlar [ Wed May 25, 2016 3:08 pm ]
Post subject:  Re: Investigator Strategy

++ aok

Page 1 of 1 All times are UTC - 5 hours
Powered by phpBB® Forum Software © phpBB Group
http://www.phpbb.com/