Ranked Vengeful (Developing a rating system based on per game player ranking) Discussion

Archive Completed Signups
,
#122

Mafia breaks ties

#123

Is it not possible for the mafia to unjustly break the tie?

#124

I think tiebreakers are pretty problematic, esp in small setups like vengeful. There are a few situations where gamethrowing becomes your best option as town. Take these two scenarios, and assume you’re omnipotent / good at mafia:

  1. You haven’t voted yet, and two people are at L-1. You’re certain that one is the godfather, and the other one is you. Who do you vote for to maximize your ranking?
  2. There’s a townie at L-1. He hasn’t voted yet and you’re certain that the other townie is already on the wagon. If you hammer the townie, what placement do you get?

I’m sure there are other scenarios.
I think that rather than ranking townies against each other, you should award Rating based on the points they get in the game, independent of how other members of the town perform.
If you go this route, it also opens to door to awarding rating to mafia more granularly (did you win d1, did the gf get lynched, etc).

1 Like
#125

#teamUrist

1 Like
#126

Sounds like something I read at Ellibereth.com one time

#127

Where blogs go to die

#128

@Urist for the first situation voting the godfather is guaranteed 3rd while voting self is at best 3rd so the right play still makes sense, but I understand your poin. Makes sense to do away with tiebreaks within action sets I think and just make the final calculation a bit more robust (e.g. not necessarily top 2 gain, bottom 2 lose, 1 2-2 4 final ranks possible, will have to flesh that out.).

It’s hard to implement this in a way that’s not super ugly/straight up doesn’t work:

Its important to treat mafia and town as one entity for final calculation otherwise it opens the door to some very degenerate strategies. If you keep them in the same pool “weighing” items from the different alignments next to each other becomes hard (also hard when you have to compute something per action that makes sense as a rating.) Beyond that I’m convinced the right way to go is for movement to be a function of other people’s ratings and not just based on actions in a vacuum. In the limit someone that consistently gets less things right shouldn’t also tend towards infinity…(Consider a math test where there are 3 problems that progress in difficulty, someone getting all 3 right and someone getting +2 -1 everytime…would become a sample size race). Basically the “easiness” and “hardness” of actions needs to be looked at so don’t think can be independent.

There were other issues when I tried to make it work (actually did consider it!). But if you can come up with a concrete scheme that works we can try it. I think the relative ranked is actually fairly close to reflecting what I would want the format to value. Basically just the tiebreak components and the rating algorithm need to be refined imo.

#129

You know I wasn’t intending on any of that being public yet and key just posted it. -.-

#130

It’s in the public record now!

#131

but yeah in any case, I’ll make some changes to tiebreaks tonight and then fire game 2.

#132

but yeah, tldr to the previous.

There’s a lot of precedent for how to look at ratings well once you have people’s performance ordered. Accommodating for ties shouldn’t be the largest adjustment and end of the day people did perform better or worse than each other…just have to find the right definitions for it.

I think if we remove the tiebreak between cases and just let those be tied and then adjust the rating so in 1 2-2 3 cases 1 2-2 all get plus points stuff will make sense.

#133

The technically correct play is to vote for the goon, allow mafia to lynch you at DL, shoot the goon, and hope for the unlikely scenario where the other townies mess up and crossvote the next day. Worst case scenario is still 3rd but this way you can possibly get 2nd.

Simply removing the tiebreaks removes a lot of the edge cases I think. I’ll sleep on a formula for independent rating to see if I can come up w something that makes sense.

#134

ah yeah, that’s right.

Don’t spend too long working on an independent rating thing haha, spent quite a while on it and the math/game theory keeps breaking. :stuck_out_tongue:

Ordered list (even with ties) has the huge advantage that it’s relatively well established how to rate/do things with them so we don’t need to create the entire algo from scratch.

#135

Updating the OP to reflect possibility of town ties.

1 Like
#136

Okay I’m happy with firing g2 pending one more question:

Might make more sense to raise the lynch threshold from 24 hours to 48 since it seems like the threshold might act as an unofficial deadline?

#137

Without tiebreak is threshold necessary?

Any threshold is probably going to act as an artificial deadline though, so maybe just say it’s plurality at deadline if you’re attached to the no hammers before x time has passed idea?

#138

eh I think 24 or 48 hours with the option of going longer is better for flexibility reasons. The main questions is if 24 or 48 hours is the necessary time for everyone to get their vote on someone they’re happy with.

#139

24 is probably fine I think

#140

Note to self: need to make adjustment to rating system so that town slot in a slot alone does sightly more favorably in terms of gain/loss as compared to when it’s a tied slot.

#141

I think will test 48 next game. 24 felt a bit short to me, especially imagining when we extend system to 7 or 9 player setups.

1 Like