[SETUP] Hollywood Mafia

Lycanfire · Post Post #0 (ISO) » Thu Apr 13, 2017 7:09 pm

Hollywood Mafia 13p

10v3
x4 celebrity
x6 VT

x1 mafia celebrity
x2 mafia goon

Celebrity:

Your celebrity status is confirmed at the start of the game

gigabyteTroubadour · Post Post #1 (ISO) » Fri Apr 14, 2017 9:22 am

seems really swingy

10:3 mountainous is scumsided, but if a mafia celebrity is lynched town gets 4 ICs which townsides the setup

i would try it \_O_/

have you ran it offsite?

Lycanfire · Post Post #2 (ISO) » Fri Apr 14, 2017 2:52 pm

Part of the game dichotomy is that there's a higher chance of scum in the VT claims than the known celebrities, and the trade off of hunting in celebrities is the conftown. While it seems swingy, it's my opinion that it is likely still scumsided.

I haven't modded this before, but I would like to.

Randomnamechange · Post Post #3 (ISO) » Sat Apr 15, 2017 1:30 am

i feel like the optimum strategy is to ignore groups until one scum is lynched?

Lycanfire · Post Post #4 (ISO) » Sat Apr 15, 2017 4:15 pm

In post 3, Randomnamechange wrote:i feel like the optimum strategy is to ignore groups until one scum is lynched?

This is an interesting thought, but entertaining it from then on comes down to outguessing who the mafia would kill. At what point does the town get tempted to play the set up? Scum already dictate night, what if they drive day play? I think this kind of problem and how there's no clear strategy is what makes the set up valid. A procedural mafia game that has one route (like follow the cop games) are awful to play.

Randomnamechange · Post Post #5 (ISO) » Sun Apr 16, 2017 2:03 am

yeah fair enough. The variation in lynching/NKs is going to make it valid.
does this set up benefit town enough? 10v3 mountainous is scumsided right?

gigabyteTroubadour · Post Post #6 (ISO) » Sun Apr 16, 2017 2:08 am

if i felt like it i could probably write a simulation random

Randomnamechange · Post Post #7 (ISO) » Sun Apr 16, 2017 5:16 am

i was thinking about it as well, but the different strategies makes it kinda weird to accurately write a program for.

gigabyteTroubadour · Post Post #8 (ISO) » Sun Apr 16, 2017 5:29 am

i would have gone for "lynch randomly until 1 mafia celeb is lynched or two goons are"

after which that group would be ignored

Randomnamechange · Post Post #9 (ISO) » Sun Apr 16, 2017 11:04 am

fair enough.

Lycanfire · Post Post #10 (ISO) » Mon Apr 17, 2017 1:17 pm

In post 5, Randomnamechange wrote:yeah fair enough. The variation in lynching/NKs is going to make it valid.
does this set up benefit town enough? 10v3 mountainous is scumsided right?

You're right by saying 10v3 mountainous is scumsided-it's very scumsided, and that was the point in this set up. I was trying to create a mountainous-esque game while providing the clears necessary to give town a chance, if they can do the leg work for it.

I think mathematically it's still more than fairly scumsided-but I feel like lynchpools and PoE are potentially damning to scum, math aside. The informed minority SHOULD be strong enough to outplay with favorable odds, and most cases favor scum by a margin. I would be interested in modding or playing in a 10v3 set up to see how the two groups would approach it.

Randomnamechange · Post Post #11 (ISO) » Wed Apr 19, 2017 5:19 am

its logically v scumsided but in practice could be able to swing it back towards balanced as scummier players make town more likely to lynch in the right group.

Randomnamechange · Post Post #12 (ISO) » Wed Apr 19, 2017 8:48 am

writing a program for it atm

Umlaut · Post Post #13 (ISO) » Fri Apr 21, 2017 1:51 pm

I'd play it.

Umlaut · Post Post #14 (ISO) » Fri Apr 21, 2017 2:29 pm

Just ran a 1,000,000-fold simulation with the following assumptions.

Lynches are according to the following algorithm:
- If all living celebrities are mafia, lynch a random celebrity.
- If all living non-celebrities are mafia, lynch a random non-celebrity.
- If all mafia celebrities are dead, lynch a random non-celebrity.
- If all mafia goons are dead, lynch a random celebrity.
- Otherwise, lynch a random player.
Nightkills are according to the following algorithm:
- If all mafia celebrities are dead, or if all vanilla townies are dead, shoot a random town celebrity.
- If all mafia goons are dead, or if all town celebrities are dead, shoot a random vanilla townie.
- Otherwise, shoot a random town player.
- However, never shoot a player if he is the last town celebrity or the last vanilla townie, and there are living mafia in that group.

Under these rules, the town EV is

48.4%

. That's actually pretty well balanced.

I ought to see if there are any tweaks to the lynch strategy that raise the town EV, or any tweaks to the nightkill strategy that lower it, but this is probably not a bad estimate.

Umlaut · Post Post #15 (ISO) » Fri Apr 21, 2017 3:13 pm

I've edited the above post three or four times as I've found mistakes in my code, but I'm pretty confident it's right this time.

It's simple, it's interesting, and it has close to 50% EV. I say run it.

Lycanfire · Post Post #16 (ISO) » Fri Apr 21, 2017 8:32 pm

Thank you for the analysis. Will definitely point to that in Open Review when my current game (Open 685) ends. Of course, someone else is free to fast track it, just be sure to pre-in me

Randomnamechange · Post Post #17 (ISO) » Sat Apr 22, 2017 3:17 am

I did the same and got a 54.1 % win rate for town but mine had slightly less mafia strats than yours so i would say it's balanced.

mhsmith0 · Post Post #18 (ISO) » Sun May 14, 2017 1:27 pm

I kinda feel like anything at 45% or above EV through random lynching is townsided, unless you just think that towns are totally hopeless.

Umlaut · Post Post #19 (ISO) » Sun May 14, 2017 1:33 pm

Win-lose records suggest towns do worse than random in general.

LlamaFluff · Post Post #20 (ISO) » Thu May 18, 2017 2:25 pm

Doesn't mass claim day one hurt scum chances way too much?

D1 massclaim (known 6VT, 2 Goon)... assume scum tries to kill the worse odd group each night and scum mislynches from worse odds pool...

D1 - VT lynch (25% vs 20%)
N1 - Celeb kill to create (28.5% vs 25%)
D2 - VT lynch (2/7 vs 1/4) -> 2/6 v 1/4
N2 - Celeb kill to create (33% vs 33%)
D3 - Celeb lynch (33% vs 33%)
N3 - VT kill to create 2/5 v 1/2

So if both alignments simply play optimally, there is a 100% chance of the worst thing that can happen for scum a lylo where one of them is in a 50-50 caught chance.

The other town option is to just try to power lynch Celebs (higher risk/reward), as if they hit one quick you start forcing a lot of kills for scum. D1 scum celeb lynch I think the game is almost over as there are now four forced NKs

Toto · Post Post #21 (ISO) » Fri May 19, 2017 7:01 am

In post 14, Umlaut wrote:Otherwise, shoot a random town player.

This is non-optimal. Scum should shoot the group with largest town to scum ratio in most cases.

According to a mix-max strategy if both team play optimally then EV is 27%, if my code is right.

Here are the suggested actions in non-trivial mixed living players cases (if all remaining players are either Vanilla or Celeb, then it is a standard game mechanic):

Spoiler: actions

mhsmith0 · Post Post #22 (ISO) » Fri May 19, 2017 7:02 am

In post 19, Umlaut wrote:Win-lose records suggest towns do worse than random in general.

Is that actually true? Like pure random lynching gives atrocious town win records; I guess the question is how towns do compared to random lynching given the power assigned to them, which I don't know the actual answer for.

Toto · Post Post #23 (ISO) » Fri May 19, 2017 7:12 am

In post 22, mhsmith0 wrote:
In post 19, Umlaut wrote:Win-lose records suggest towns do worse than random in general.
Is that actually true? Like pure random lynching gives atrocious town win records; I guess the question is how towns do compared to random lynching given the power assigned to them, which I don't know the actual answer for.

Newbies D1 stats are pretty close to random lynching. However, this is bad considering there are PRs that should never be lynched.

Toto · Post Post #24 (ISO) » Fri May 19, 2017 7:16 am

In post 23, Toto wrote:
In post 22, mhsmith0 wrote:
In post 19, Umlaut wrote:Win-lose records suggest towns do worse than random in general.
Is that actually true? Like pure random lynching gives atrocious town win records; I guess the question is how towns do compared to random lynching given the power assigned to them, which I don't know the actual answer for.
Newbies D1 stats are pretty close to random lynching. However, this is bad considering there are PRs that should never be lynched.

I have a theory that towns are worse than average before the first scum flip, but better afterwards.