[SETUP] [EV] Monty Hall 7p

[RadiantCowbells]

this comes out as a lot more scumsided than you're thinking it is.

[RadiantCowbells]

keep in mind that a single night protection is essentially worthless
you have two top townreads on d1 and both of them are dead by d3, all town gets to do is change the order.

[RadiantCowbells]

like yes the monty hall mechanic does help but in the end all it really changes is that town's wincon is top 3 townreads be town as opposed to top 4 townreads be town

[RadiantCowbells]

hell it's not even that and that explains why it's worse than that cuz

top 3 town can be town, but if scum are townier than 2 townies then they win

town path to victory is insanely narrow here.

[Mathdino]

How does that shake out against the 50/50 mechanical EV? Do you think 2:7 here is actually more balanced?

- D3 is 1:2 LyLo, which is 2/3 likely to be won by town. It seems really difficult for mafia to win the 1:2 without getting the protection vote.

- It seems like it should have a really similar gametate progression to Marked for Death itself, which also allows mafia to kill top townreads.

- Town still retains probability info from previous days, which mechanically benefits them.

I'm a little biased toward running it as 2:5 right now given site activity and speed of completion but I could definitely do a 2:7 after to keep testing.

[Mathdino]

top 3 town can be town, but if scum are townier than 2 townies then they win

So if you have {1, 2} scum and {3, 4, 5, 6, 7} town, say 1 gets protected and 3 and 4 are killed/lynched. The next day, say 2 gets protected and 5 marked for death. Doesn't that lead to "why are you still alive" discrediting against 1?

And if town gets 1 in this scenario, the next day you have {2, 6, 7} with one townie protected and one IC, leading to what's almost mechanically an auto-loss for scum.

I think the primary source of these wacky EV calculations is the anti-swinginess of LyLo. To a weird degree, it almost punishes scum getting townread TOO hard early on rather than later.

[RadiantCowbells]

- D3 is 1:2 LyLo, which is 2/3 likely to be won by town. It seems really difficult for mafia to win the 1:2 without getting the protection vote.

from a certain point of view this is true
from another point of view, the top two towniest towns are already nightkilled and all you're actually asking for mafia to be the towniest player in that lylo is for 1/2 of the mafia to be in the top 3 towniest players

[RadiantCowbells]

Doesn't that lead to "why are you still alive" discrediting against 1?

or what if they left him alive for this reason :thonk:

this goes both ways, really.

[RadiantCowbells]

like you understand that 10/14ths of the time with scum evenly distributed in the playerlist scum is going into that lylo as the towniest player

[RadiantCowbells]

and that mechanic is essentially irrelevant in non-lylo days because the nightkilled player isn't really going to be on the lynch list.

[Mathdino]

I see your points, and I do think Mountainous creates an inherent scumsidedness (unless scum have no NK). That said, per pops's argument, I'm not really comfortable running a 2:7 experimental setup with 57% mechanical EV in what's apparently a townsided site meta (?), especially when I'm used to trying to balance games around ~40% mechanical.

Is 2:7 scumsided, balanced, or townsided?

[RadiantCowbells]

i think that as 2:N approaches infinity the setup is always scumsided.

[RadiantCowbells]

we have heavy empirical evidence that increasing the player count in mountainous doesn't really help. that meshes with both how I intuit the game playing out and how i calculate setups focusing on how townread each player is.

[RadiantCowbells]

hell

you know how this is balanced? 2:3.

[Mathdino]

2:3 comes out to 35.6% with random lynches.

[RadiantCowbells]

significantly better odds than 2:5 or 2:7

[RadiantCowbells]

EV may disagree with me but EV doesn't at all factor in the disproportionately scum sided effect of the fact that the nightkill is targeted at a townie chosen by scum.
EV also doesn't factor in that people aren't equally likely to get lynched, there's voting dynamics that come into play and etc.

I think EV is essentially worthless.

[Mathdino]

For Mountainous setups, would you feel more comfortable with a model that randomly ranked players from towniest to scummiest and lynched/NK'd accordingly?

Redoing the EV project with that in mind would be pretty trivial. You could call the status quo the Random Lynch Model and yours the Random Reads Model.

[RadiantCowbells]

and in 5 way it plays closer to nightless because you only need the 1 correct townread, it gets locked in by your save, and I think town are disproportionately likely to be correct on their top townreads and wrong on their top scumreads, so I think the whole fact that you only have to pick 1 scum out of 3 with a correct townread is a saving grace for scum

intuitively if you ignore EV doesn't starting in lylo feel better for town's odds?

[RadiantCowbells]

For Mountainous setups, would you feel more comfortable with a model that randomly ranked players from towniest to scummiest and lynched/NK'd accordingly?

Redoing the EV project with that in mind would be pretty trivial. You could call the status quo the Random Lynch Model and yours the Random Reads Model.

For Mountainous setups, would you feel more comfortable with a model that randomly ranked players from towniest to scummiest and lynched/NK'd accordingly?Redoing the EV project with that in mind would be pretty trivial. You could call the status quo the Random Lynch Model and yours the Random Reads Model.

I run this both for mountainous and setups with roles

[RadiantCowbells]

i also simulate read variance, slots actually voting each other, etc.

[Mathdino]

i mean
do you have the math anywhere onsite or
have you been keeping this to yourself

because uh
BIG if true

[RadiantCowbells]

i've brought this up before, i know i've gotten into arguments with people about ev before

[Mathdino]

right but do you have the numbers already set up or shall i get on that

[RadiantCowbells]

you mean for this setup?