Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[SpaceAnemone]

Here's my table or pair-wise results, based on re-reads. I'm disappointed that so many pairs appear to have a medium-to-high likelihood. I also feel like I could spend hours more chewing over the various comments, but I'm afraid that's time I don't have... it's already after 1am here, and I need sleep :-(

Jake-Robz	Very High
LL-Robz	Very High
Calamitas-EFHW	High
Calamitas-Robz	Medium-High
LL-Calamitas	Medium-High
LL-Jake	Middling
Jake-Calamitas	Medium-Low
LL-EFHW	Low
Jake-EFHW	Very Low
Robz-EFHW	Null

I meant to add that in this context, I'm using "null" to mean "I don't know how to assign a likelihood here". I think Robz and EFHW are two very experienced players so looking at their interactions is considerably less likely to yield useful tells. It does not mean I don't think they're possible as a scumteam.

[EFHW]

My scumreads on Jake and Lalight are weakening. Still on the fence about Robz. Going with vote: Calamitas.

Surely if he's scum he'd have hammered you by now, no?

He can't because he said the case against me was ridiculous. I'm sure that's part of where my initial townread on him started, but he's been so inactive.  I'm really at a loss here.  I did notice that Calamitas is in 3 of the top 5 of your pairings.

[SpaceAnemone]

Also, the length of time EFHW's been at L-1 makes me feel like she's more likely to be town.

This is exactly backwards, though. It's scum who languish at L-1, because their partners don't vote for them, and they try to subtly undermine the lynch. This should be an argument in favor of EFHW being scum.

3/5 of my lynchpool is already voting EFHW. The only person that leaves is Calamitas. I have Calamitas-EFHW as a high-likelihood pairing, but with anyone else it looks considerably more unlikely to me, so overall out of my possible scum-paring reading, the marginal probability of EFHW being scum is much lower than for other people.

Yeah but I think the LL-EFHW pair should be very high or at least high

I was responding to Robz's particular comment about how scum languish at L-1 because their partners don't vote for them.

[EFHW]

I should say 2 out of 4, since one of those pairings is with me!

[SpaceAnemone]

My scumreads on Jake and Lalight are weakening. Still on the fence about Robz. Going with vote: Calamitas.

Surely if he's scum he'd have hammered you by now, no?

He can't because he said the case against me was ridiculous. I'm sure that's part of where my initial townread on him started, but he's been so inactive.  I'm really at a loss here.  I did notice that Calamitas is in 3 of the top 5 of your pairings.

Robz is in three of the top four. And you and Robz have incomplete information because I just didn't want to pick a number.

[LaLight]

I suspect everyone here.
Less so Calamitas. Others are on the very level "Will be glad to lynch"

PPE:3

[EFHW]

My scumreads on Jake and Lalight are weakening. Still on the fence about Robz. Going with vote: Calamitas.

Surely if he's scum he'd have hammered you by now, no?

He can't because he said the case against me was ridiculous. I'm sure that's part of where my initial townread on him started, but he's been so inactive.  I'm really at a loss here.  I did notice that Calamitas is in 3 of the top 5 of your pairings.

Robz is in three of the top four. And you and Robz have incomplete information because I just didn't want to pick a number.

Still willing to lynch Robz.

[EFHW]

vote: Robz

[Calamitas]

My scumreads on Jake and Lalight are weakening. Still on the fence about Robz. Going with vote: Calamitas.

Surely if he's scum he'd have hammered you by now, no?

He can't because he said the case against me was ridiculous. I'm sure that's part of where my initial townread on him started, but he's been so inactive.  I'm really at a loss here.  I did notice that Calamitas is in 3 of the top 5 of your pairings.

Robz is in three of the top four. And you and Robz have incomplete information because I just didn't want to pick a number.

Still willing to lynch Robz.

What is the case there?

[EFHW]

POE and feeling unable to read him.

[SpaceAnemone]

Sorry... wrote this just after my last post, but work is hectic and I really need to stop trying to keep up with the forum just now, but it's too interesting :-P

So looking at my mean marginals on the numerical version of my table in my notes, I seem to have Robz > Calamitas > LL > Jake > EFHW for my scum ordering. I'm not sure that's the optimal way of calculating it, but it doesn't feel a million miles out.

I think I'd like to hear more from Robz in a way that analyses play in this game, rather than talks about how scum/town tend to play on average.

[LaLight]

I can see no single reason for SA to townread EFHW.

[Calamitas]

Sorry... wrote this just after my last post, but work is hectic and I really need to stop trying to keep up with the forum just now, but it's too interesting :-P

So looking at my mean marginals on the numerical version of my table in my notes, I seem to have Robz > Calamitas > LL > Jake > EFHW for my scum ordering. I'm not sure that's the optimal way of calculating it, but it doesn't feel a million miles out.

I think I'd like to hear more from Robz in a way that analyses play in this game, rather than talks about how scum/town tend to play on average.

How did you gather the pair-reads from above? Just be looking at the interactions or did you take your scum-priors into account?

[JaketheBaseballGod22]

I can see no single reason for SA to townread EFHW.

I agree why do you think EFHW is town SA

[SpaceAnemone]

Sorry... wrote this just after my last post, but work is hectic and I really need to stop trying to keep up with the forum just now, but it's too interesting :-P

So looking at my mean marginals on the numerical version of my table in my notes, I seem to have Robz > Calamitas > LL > Jake > EFHW for my scum ordering. I'm not sure that's the optimal way of calculating it, but it doesn't feel a million miles out.

I think I'd like to hear more from Robz in a way that analyses play in this game, rather than talks about how scum/town tend to play on average.

How did you gather the pair-reads from above? Just be looking at the interactions or did you take your scum-priors into account?

The pairwise reads were done looking at how pairs of people interacted with each other, and folding in information about their voting patters on each other when viewed in isolation from the actual content (since I think scums kind of generate noise around their motives, but the action of voting is more clear.

The ordering above is a marginal over the scum pairings (i.e. take a mean score of each of the four pairs involving each of the five of you), so it's not totally reliable in that respect, especially since I had to put a number on Robz-EFHW in order to sum it.

I'm not totally comfortable with what the scores themselves mean in a physical sense. Amount of expected surprise on seeing a particular pairing confirmed was the closest explanation I could fit to it in my head..

[gkrieg13]

Every time you get the preferred pronouns wrong for someone in a post, you get 1 strike. 3 strikes and you're out.[/color][/b]

You don't need to modkill anyone on my account.  I don't take it personally - I'm usually the only she, and new players assume I'm male until they hear otherwise.  I take it like I'm "one of the guys".  I correct them, or let others do it usually now, because I want to be known and don't want them to be embarrassed later, but it's not an insult to me.

Getting used to the singular "they" is hard - I still always imagine a crowd of people when I hear someone use "they" to refer to an individual. It's disrespectful not to make an effort, but I think punishment may be going too far.  That's for Space to say, I guess, since their feeling of inclusion is what's important here.

It isn't really on your account.  Part of it is because even if it is hard, Jake should be able to get it right.  A second part of it is if someone who actually doesn't want to be referred to as "he" comes around, I want them to comfortably be able to play in a game with Jake.  Considering that we actually have people like that in our mafia community (people who are trans/non-binary), I feel like it is something to support.

Either way, that is 1 strike Jake

[Calamitas]

Sorry... wrote this just after my last post, but work is hectic and I really need to stop trying to keep up with the forum just now, but it's too interesting :-P

So looking at my mean marginals on the numerical version of my table in my notes, I seem to have Robz > Calamitas > LL > Jake > EFHW for my scum ordering. I'm not sure that's the optimal way of calculating it, but it doesn't feel a million miles out.

I think I'd like to hear more from Robz in a way that analyses play in this game, rather than talks about how scum/town tend to play on average.

How did you gather the pair-reads from above? Just be looking at the interactions or did you take your scum-priors into account?

The pairwise reads were done looking at how pairs of people interacted with each other, and folding in information about their voting patters on each other when viewed in isolation from the actual content (since I think scums kind of generate noise around their motives, but the action of voting is more clear.

The ordering above is a marginal over the scum pairings (i.e. take a mean score of each of the four pairs involving each of the five of you), so it's not totally reliable in that respect, especially since I had to put a number on Robz-EFHW in order to sum it.

I'm not totally comfortable with what the scores themselves mean in a physical sense. Amount of expected surprise on seeing a particular pairing confirmed was the closest explanation I could fit to it in my head..

Do multiple scummy interactions even imply anything then?

[SpaceAnemone]

Do multiple scummy interactions even imply anything then?

Is your question about my initial pairwise method, or the marginalisation? I'm not really sure what you're asking sine my pairwise read was more looking for lack of conflict or non-interaction than our was looking for anything actively scummy. Does that make sense?

[Calamitas]

Do multiple scummy interactions even imply anything then?

Is your question about my initial pairwise method, or the marginalisation? I'm not really sure what you're asking sine my pairwise read was more looking for lack of conflict or non-interaction than our was looking for anything actively scummy. Does that make sense?

What I meant is that having multiple "high" entries doesn't make someone scummy here. Because if one has a scummy (in regard to the pairwise reading method) interaction with multiple people it becomes a no tell. Those entries work only on a relative scale, not on an absolute if I understood your method correctly.

[SpaceAnemone]

Do multiple scummy interactions even imply anything then?

Is your question about my initial pairwise method, or the marginalisation? I'm not really sure what you're asking sine my pairwise read was more looking for lack of conflict or non-interaction than our was looking for anything actively scummy. Does that make sense?

What I meant is that having multiple "high" entries doesn't make someone scummy here. Because if one has a scummy (in regard to the pairwise reading method) interaction with multiple people it becomes a no tell. Those entries work only on a relative scale, not on an absolute if I understood your method correctly.

Okay, I see where you're going with that now; you're arguing that if I don't normalise by how likely an individual is to have a "scummy-looking" possible pairing overall, then the scores shouldn't be directly comparable, yes?

I'm not sure I agree. The way I formulate the evaluation in my head was "how surprised would I be at the end of the game if this pair turns out to have been the scums", on a scale of 0 to 1. So I thing averaging in that sense is still valid without additional scaling. It's just hard to fold this with other correlated info like how scummy a person has seemed from things they've said in isolation.

[SpaceAnemone]

A second part of it is if someone who actually doesn't want to be referred to as "he" comes around, I want them to comfortably be able to play in a game with Jake.  Considering that we actually have people like that in our mafia community (people who are trans/non-binary), I feel like it is something to support.

I really appreciate this, and I'd like to see f.ds be this kind of inclusive community :-)

[Calamitas]

Do multiple scummy interactions even imply anything then?

Is your question about my initial pairwise method, or the marginalisation? I'm not really sure what you're asking sine my pairwise read was more looking for lack of conflict or non-interaction than our was looking for anything actively scummy. Does that make sense?

What I meant is that having multiple "high" entries doesn't make someone scummy here. Because if one has a scummy (in regard to the pairwise reading method) interaction with multiple people it becomes a no tell. Those entries work only on a relative scale, not on an absolute if I understood your method correctly.

Okay, I see where you're going with that now; you're arguing that if I don't normalise by how likely an individual is to have a "scummy-looking" possible pairing overall, then the scores shouldn't be directly comparable, yes?

I'm not sure I agree. The way I formulate the evaluation in my head was "how surprised would I be at the end of the game if this pair turns out to have been the scums", on a scale of 0 to 1. So I thing averaging in that sense is still valid without additional scaling. It's just hard to fold this with other correlated info like how scummy a person has seemed from things they've said in isolation.

In that case you did include priors since you compared posteriors.

[Calamitas]

Suppose we have someone really scummy (scumslip) who does interact with basically everyone quite towny. Based on the interactions only you wouldn't expect any scum pair with that person. Based on their scumslip you expect them to be scum though. So the "how much i'm going to be surprised" criterium resolves in much higher scum ratings than the conversations alone would.
And if one actually analyzes the pairs and the scuminess of their interactions one has to look only at the discrepancy between different possible partners. Suppose person A interacts scummy with B, C and D. And person B interacts towny with B,C and scummy with D.
The only information you can get there is:
- A seems to interact scummy with everyone, no tell in regard to their potential partner
- The estimate of B&D being scum has risen a lot

[SpaceAnemone]

Suppose we have someone really scummy (scumslip) who does interact with basically everyone quite towny. Based on the interactions only you wouldn't expect any scum pair with that person. Based on their scumslip you expect them to be scum though. So the "how much i'm going to be surprised" criterium resolves in much higher scum ratings than the conversations alone would.
And if one actually analyzes the pairs and the scuminess of their interactions one has to look only at the discrepancy between different possible partners. Suppose person A interacts scummy with B, C and D. And person B interacts towny with B,C and scummy with D.
The only information you can get there is:
- A seems to interact scummy with everyone, no tell in regard to their potential partner
- The estimate of B&D being scum has risen a lot

If I had strong first-order information (reads on individuals, scumslips etc), I wouldn't have needed to resort to the much more costly second-order information (pairwise interactions, particular voting histories). So let's just assume a flat prior on the five of you.

I think I object to you calling interactions "scummy", because it makes it sound like there are definite pieces of evidence that I'm integrating over a set of interactions or something. That's not the case: the pair-wise tells seem to be much more subtle, and difficult to quantify, which is why this is hard. I can do inference and mathematical modelling really well; it's reading people I'm less good at :-P

[2.71828.....]

With 7 alive, it takes 4 to lynch. Day 1 ends Sunday Nov 19 at 12 pm

Reread tomorrow. I will not be around on Sunday at lunch, but will be around about an hour or so before. Hopefully we can get a lynch done Saturday though