Automatic Balance Tester - Page 4

On January 11 2014 06:05 gonzaw wrote:
I think....we could just use good ol' probabilities and statistics, based on all possible scenarios.

For example, you can balance a game, knowing that in the worst case scenario for town, town loses on D3 after 2 misslynches and 2 NK going through. Then you have the worst case scenario for scum, where town lynches scum 3 days in a row, the medic saves shots 2 nights in a row, cop gets checks 2 nights in a row, etc.
But what about the stuff in between? You can't balance JUST using worst case scenarios, because those will NEVER happen, they are just a rough guide on what "could" happen.
I think it's more important to balance according to what WILL indeed happen.

Maybe the best way to model that, is just count all possible scenarios, assign a probability to them, or maybe some other factor (like a Balance Number perhaps?), and then determine the overall balance of the game, knowing the probability of each of these scenarios.
Most likely there would be heuristics to determine these probabilities/factors, or some other theory that can help.

For instance, yes, the worst case scenario for scum is if the Town Vig shoots scum on N1, the medic saves 2 townies on N1 and N2, and town lynches scum on D1, D2 and D3. Good, but that will surely NOT happen at all. What will most likely happen, is town misslynches on D1, town vig shoots a townie on N1, town lynches scum on D2, medic saves townie on N2, bla bla bla. THAT is the "normal" flow of the game you should be balancing. If town has it so easy, that in that scenario they win almost all the time, then the game is balanced wrong, even if there exists a "worst case scenario" for town that seems it can fuck them up pretty badly. The point is that it's more probable that the "normal" flow of the game happens than that worst case scenario.

Different roles have different balancing factors as well, that's not just a +1 or -1. A town vig can shoot a townie or a scum, he has the power to greatly favor scum or town. That has a lot of variance, so when a vig is inserted in a game, it should introduce variance on the chances each faction can win, and thus balance.
Idem when you have a cop and a framer or GF for instance. It introduces variance, since the cop may check the GF. That greatly changes the game, since the GF could be considered almost confirmed town. Or maybe town is good and catch him nevertheless. That's even more variance (how town can react to it).

I think something like that might work better: introduce the average chance each faction has to win the game, and the variance of it, and balance (somehow) according to that).

Seems to me there can be better ways to handle balance, to be as accurate as possible.

On January 11 2014 06:48 kitaman27 wrote:


I think considering worst case scenarios is helpful because it allows you to eliminate undesirable scenarios. In mini games, the worst case (or close to worst case) scenarios aren't so uncommon that you never see them play out that way since you're usually only dealing with 1-2 roles.

Suppose you come up with a setup that is 55/45 in balance and each faction must pull off three correct lynches before they have a chance of winning.

You have a have another setup that is 50/50 in balance, but once out of every ten times, a game snowballs with a night one loss for a faction.

I'd probably want to run the first setup where you can ensure you aren't going to end up with the occasional dud, even if it is slightly less balanced.

It's difficult to calculate the average chance each faction has to win. You could try to calculate out every single scenario and take an average of all outcomes, but that's assuming lynches and night actions are randomized, which isn't the case. You could try to come up with odds of each lynch going a certain way based on past games, but I don't think the sample size is large enough for it to give us an accurate weighting.

On January 24 2014 05:53 HiroPro wrote:
Merc Micro isn't a normal setup. The players start out with items (so they aren't actually VTs) and money to buy other items.

On January 27 2014 00:43 raynpelikoneet wrote:
I think the optimal/suboptimal play is the most decideing factor in a game. Agreed, probably impossible to balance since you can't know or predict where the suboptimal play happens (is it going to be mafia, town??) and it's impossible to calculate.

If you look at Really Small mini mafia for example: 1 doctor, 4 VT vs 2 goon (+ instant majority lynch which usually favors mafia). Your program gives the following chances:
Town 45.16 %
Mafia 54.84 %
-> Overwhelming town victory

After the game i talked with Mocsta about the game. He asked me "how is mafia supposed to win that game?". I don't know, if we assume all the players played as optimally they did in that game, regardless who was mafia i think mafia wouldn't have stood a chance. When townies look town there are no mislynces. So what do you do? You most likely lose because you need to make up things as there is no bad play to punish and good players will see through your bullshit.

So all in all i think the play of players is the biggest factor in games, regardless of the roles, but it is impossible to balance as you can't predict the "level of play" beforehand.

On January 27 2014 04:56 Xatalos wrote:


That's an interesting example.

I agree that it gets harder for Mafia if everyone is playing "perfectly". On the other hand, if everyone *is* playing perfectly, then Mafia should be favored, since it would be just random chance to hit Mafia (which is more unlikely). But it's a lot harder for Mafia than for town to really appear town in an environment where every townie appears townish.

Setup balance does still play a role and should be focused on - since it's the only truly predictable part of game balance. Another way might be to somehow balance the players (an equal share of "good" players for both teams?), but then that would create a new area of setup speculation Maybe some hosts already do balancing like this and I'm just not familiar with it. It doesn't actually seem like that bad of an idea considering the fairness/balance of games.

On February 14 2014 11:32 Promethelax wrote:


Some hosts do balance that way, more often with vets rather than good players but the idea remains the same, I don't condone it but some do.

On February 16 2014 20:03 Xatalos wrote:


Hm. Isn't it better if not all the veteran players are on the same team? Pure randomness could produce games where one of the teams never stood a chance, I'd think?

On February 17 2014 11:58 Promethelax wrote:


You would think that but individual play is what wins and loses games, sometimes vets rotfl stomp newbies (fruity) and sometimes newbies destroy vets (les mafia). I prefer these random draws because it gives town one less thing to game, there is already setup speculation and I personally find vet balance speculation totally worthless because it is not mafia.

On February 17 2014 11:58 Promethelax wrote:


You would think that but individual play is what wins and loses games, sometimes vets rotfl stomp newbies (fruity) and sometimes newbies destroy vets (les mafia). I prefer these random draws because it gives town one less thing to game, there is already setup speculation and I personally find vet balance speculation totally worthless because it is not mafia.