Team balances for high os players

[languid spire]

As I have shown in the image, the game is very unpleasant for high elo players. I think a new balance system is needed for this. Apart from me, there is not a player with a higher OS than the opponent in my team. What do I need to do to win such a game?I trust the balance team, they will definitely solve this problem.

<@&1014572756249083936>

[pale sleet]

Do what everybody else does
Start trolling and drop OS so you can enjoy the game again 😉

[languid spire]

Honestly my brother is thinking of starting the game, I can play games on his account sometimes if this problem is not solved 🙂

[silver laurel]

Bro you are literally #1 for large team OS. Kind of makes it pretty hard to balance you no? By a significant margin too. Everyone else is within a 20 OS range and you are 20 OS above the second highest player. What else do you expect it to do?

[languid spire]

After passing a certain os, the game becomes somewhat luck-based. More os don't necessarily bring more resources. Therefore, it's necessary to reduce the impact on team balance after passing a certain os.

i think snake draft could be good for team balance.

[silver laurel]

Thats what uncertainty is supposed to help? But it would seem to have such a high os you've consistently won against the predictions odds to do that. I'm not trying to take the piss I'm sure it is actually annoying but like, I don't know, if I cared about OS I would dodge that game as team 2 regardless of the predictions.

Do you mean like a captains mode alternating selections or do you mean snake draft like, alternating highest OS back and forth entirely.

[languid spire]

I had to dodge the game lol

like this

[pale sleet]

[silver laurel]

Very cool that you were prepared haha. I can agree to some degree logically that there is a ceiling of influence on the game that you cant exceed with just skill. I don't think balancing out each player more 1-1 as in equivilant matching each player as best as possible would help. All its done is put the winning favor to the player with the highest OS. And the entire enemy team would want to dodge now rather than just you.

[languid spire]

Honestly, I don't see a huge difference in balance between the teams. They are more balanced than when I first posted them

[silver laurel]

I do see your point, I think there is another thread with a similar premise. To balance around 50/50 OS prediction rather than trying to equal open skill / uncertainty totals?

#1416636313955733627 message

I can also see this solution makng you win even more often and making the situation even worse though lmao. Honestly the ability to even get that high is ludicrous, eg I've been top500 overwatch au/us combined. The top end gets to a point where you have to win over 6 games in a row to negate a single loss worth of climb.

[languid spire]

my chances of winning a game are around 20% right now if play all games

[silver laurel]

Which if I compare to my logic of high skill rating in top percentage of overwatch players is about correct, 1 in 5 to 1 in 6 wins

[languid spire]

yeah %5-%95 It's ridiculous

[silver laurel]

https://www.bar-stats.pro/playerstats?playerName=Emre&dateRange-0=1658102400000&dateRange-1=1758585599999&playerCount-0=2&playerCount-1=16&includeUnranked=false&maps=[]&gameTypes=["team"%2C"smallTeam"%2C"ffa"%2C"duel"] This suggests you shoud stop playing with Raikoh you lose with him 😛 hahaha

The fact that the game still has you somewhat close to a 50/50 win rate means it honestly isn't doing too badly though.

I do see a problem with excessively unbalanced games like 5/95 though

[pale sleet]

[silver laurel]

Anyway thank you for engaging with me about it. I'll probably see myself out since I'm literally a burnt out mid/intermediate player at best. I carry noob lobbies and get booted, and I can't tolerate high OS lobbies attitudes and boring meta/co-op anyway. (current 17os, peak 25 os, 1500 hours, for reference)

https://www.bar-stats.pro/playerstats?playerName=Emre&dateRange-0=1658102400000&dateRange-1=1758585599999&playerCount-0=12&playerCount-1=16&includeUnranked=false&maps=[]&gameTypes=["team"%2C"smallTeam"%2C"ffa"%2C"duel"]

also what changed at the 2500 game mark, is that when they changed how uncertainty was weighted?

but honestly, if you keep trying to climb its only going to get worse, if you are still dodging games at that rank to preserve OS you are kind of facilitating the problem yourself. If you want to have higher OS team mates you gonna have to let them get higher XD

[placid sequoia]

Lmao this is the first thing I saw in the chart

[dark bane]

emre we need spectator os, the dodgemasters will be so good at predicting winning team

game is just hard forcing you to 50% win rate, it's like death and taxes

Inherently in 8v8 after a certain point you have to dodge / chat your way to your teammates playing better, force team to play spawns they are good at / counter enemy blue/red expected spawns

[languid spire]

It's because I play well , but do I have to play well in every game? Maybe I want to try different things

[languid spire]

" chat your way to your teammates playing better"

[ashen osprey]

when chat really means asking 20 os player to leave game and give base

[placid sequoia]

Ya the tried and true, "Hello gameur, I see you have base there...can I borrow it permanently?" works wonders.

[pale sleet]

"Helo friend, if you want us to win you should leave the game right now, tnk you so much"
gg wp
ngl i tried it few times and it worked

sometimes it doesn't work, they stay playing the video game and you lose the game coz of it also, such is the life of BAR

[cedar void]

I think it gives you a +-4.16rating wiggle room for playing drunk to full sweat mode. But that still puts you far above everyone else :P

[solemn heath]

New meta ask low os for base. I think you can push for 90

"Wanna see me 1v4 please let me borrow your based this game 🤣"

[languid spire]

how my team %54 lol

[dark bane]

game expects you to be jesus wcyd

[pale sleet]

[proud wigeon]

From my perspective the system is working as intended, it just so happens that your rank is close to the realistic bound for the playerbase.

OpenSkill is not linear, You have a 99.9998% estimated winrate against their best player in a head to head, so to balance the game from a system perspective your team has to be far weaker than you otherwise you would wipe them every game.

Even if you make the (loosely incorrect) assumption of everyone only contributing equal 1/8 towards the outcome you end up with an even more balanced game of ~49.8% wr for your team.

While I understand it doesn't make for fun games, its the only way to get fair games given your rating no?

And if you are dodging games that don't look 'fair' it inflates your OS higher than where it possibly should be sitting, leading to this problem getting worse

[dark bane]

you can't escape forced 50% wr

[ashen osprey]

Raigeki FrozenSlayer KDBRA ...😉 wants to have a word to u about tht

[dark bane]

2 times when you can avoid it: when you are climbing, and when you quit

[lapis wing]

youre getting bad teams regardless of your os, because others dodge and inflate their os

[fast plank]

Should have said low os. But yea overall dodging can only be fixed with matchmaking.

[scarlet plank]

We could consider changing one of the current balance methods to "strict" and having it no longer respect parties or avoids. In the initial example there are 2, 3 player teams and while yes, they were broken up when you then look at the snake draft example it leads to the same team pairing result for the partied players. Having a balance:strictoption that you could vote to employ when things look bad in MM might at least help. Default Balance:Party and then have the other drop down be Balance:strict we could also do something like split-pro's that does something similar to split-noobs, so that each team has a roughly similar number of players above the average lobby OS which might also help with some of the 'feels bad' cases

Split-pro's would do something like Reserve the two highest OS players in the match - re-calculate avg match OS - trade the person closest above the avarge with the person closest to below the average and then place the above avg with the highest os and below avg with the 2nd highest OS, then redraft the team.

Using the intial image this would result in Platinumblonde being reserved for Emre and Baby_Doni reserved for Maladay, so 4 players reserved, you would then redraft the remaining 12 players normally. This should result in little change if there is a very tight avg OS between the players but help to reduce team stacking when outliers are present.

[thin path]

Mods will ban ypur ass

[scarlet plank]

Ex: !split-pros

That to me 'visibly' seems alot more fair. I see a 2v3 instead of a 1v4 if you get my meaning

[scarlet plank]

We would use -> get_initial_state/1 and then find top2 os, avg, above, below, remaining, then get_result/1 (anchor pros + redraft). Everything else you can borrow from SplitNoobs

I dunno if we would want some kind of qualifier for uncertitny since we are dropping uncertainty in split noobs.... I was just kind of assuming everyone at this rating would be pretty close in avg uncertainty or that it would be acceptable to ignore it

Ok so I mocked up the code this is what it spit out for the new teams using split-noobs as the base for a !split-pros call:

so pretty close to the "eyeballing it" example image, that would be no respect for parties or avoids or uncertainty

and I am simulating this in GPT for full disclosure if it was not obvious from the image above

[languid spire]

looking good

[scarlet plank]

K Im trying to just find test games at this point. Im gunna check some more from your match history and see what happens and then I also need to check some cases for sub 25os lobbies and see what kind of chaos it causes.

there are simlar cases for All Welcome lobbies when you have like a single 35 or 36 os player and the rest are like 20 something and down

[scarlet plank]

I probably need some actual direction for the Dev team on this if I look at other matches from my history with smaller outliers.... it can get alittle funky for exmaple:

It is doing its job per se, but not in a way that people would want in some cases. It may need some kind of separation qualifier or to have an additional reasonability check and throw out anything with more than 2 points of deviation from the original teams

That might really be the answer if the team spread deviation is greater than before split pros it just does nothing

@placid wedge Do you have any idea on how this might be expected to be handled internally as far as dropping uncertainty, throwing out bad results, or just locking the command to a certain gap between the top OS and the 2nd highest OS player? I don't really know who to talk to specifcally about this.

@languid spire I mean...it works when I use YOU pretty well lol....

Am I wrong in my assumption that most people really only consider the initial digit in OS when they do an eye ball balance "feels good" check. I would assume the results above would create a dodge simply because you have an additional ~40 os player even though the other player is only a tiny bit of OS off at 39.xx.

Seems like it really hinges on the specific gap between the closest above and the closest below average team OS.

In this case I feel like swapping the 41.30 player on Team A, with the 39.14 player on Team B would change the game from a dodge to everyone being happy...

[scarlet plank]

Ok... essentially I added something I called "hoarding Control" ; This follows the previous rules of the split noobs logic then just ensures that the top OS player does not end up with the most players in the most represented 10's place OS value. We are also going to count the highest OS player, the Anchor, as a member of that 10s place bin so they never end up with more.

[scarlet plank]

Ok here is an example list. If we look at example 2 team B you will see the 30's are stacked to one team(the lower anchor), as far as the balancer is concerned this is the correct way to do it. I am likely going to introduce deviation protection to this as well so if the teams move TOO FAR it simply tosses the results...but I feel like Ive taken the logic about as far as I am willing to without some kind of feedback.

I'll add this little context here to show that I am still respecting the overall balance when/if this is used in a match where it might not be appropriate such as Test Case 2 just keep in mind we are counting the anchors as members of the secondary bin which is why it says the 40os player is a 30s

[lapis wing]

is that yet another layer of random shit that pushes win prediction away from 50%?

[scarlet plank]

Yes but don't worry too much I tried to keep you in mind here as well, which is exactly why I also propose changes to the Balanced:advanced and Clan:skill as an ADDITION to this. I would suggest we creat a "strict" general balance algo to go with this that do not respect avoids/parties, this would be a voted option from the default Balance:Party which does respect those things, then !splitpros would be used to remove edge cases

[lapis wing]

[scarlet plank]

From what I can tell the two balance options in the drop downs are quite similar, they almost never effect the balance when changed between them.

so no reason to have two, lets keep the Make everyone happy in low OS we love parties balance as default, Allow for people to vote for "strict" when they want to, and then add a command like splitnoobs for splitpros so we get rid of "strict" cases where people feel like a single person is skewing balance beyond playability

[lapis wing]

I just think emre wouldnt have to dodge if we had proper balance every game regardless of splitpros splitnoobs etc. non 50% win prediction balance just compound on the next game and at some point you get people 10 os above or below their true performance due to balance rng

[scarlet plank]

We can take a look at this if you want in your thread, I can run some other test cases and we can compare what it looks like with a strict 50% since its kind of on the agenda anyway but it will have to be tomorrow for me

[cedar void]

Maybe we need to bring back rank balance. (I. E. Chevs not OS)

[outer hornet]

I think emre just wanted to mention he’s 80os

[dark bane]

i think they should weight super high os players differently, after a certain os (40-50) your os should only count for 50% in terms of team balancer. So you naturally get a team with 55-60% expected WR, when you win the OS doesn't give you as much.
It sucks for people just below that limit, but wcyd.

So you actually get some reward for playing game better and not 100% of your os is going into giving you worse teammates

[scarlet plank]

This is actually quite similar to what !splitpros is doing

It classifies the outlier as a member of the most represented os bin. That allows to scale whether you are a 36 playing with 18-20s or an 80s playing with 40s

No os values in image

[languid spire]

their 6th one's os is higher than my 4th one's

[scarlet plank]

K its running

[pale sleet]

omg too much os game broken

[scarlet plank]

Keep in mind this is going to work very similar for All Welcome type lobbies were we see similar OS gaps, the stars show the players that swapped

[scarlet plank]

ok nvm I had to remove the no anchor graph as it broke the AI's brain cause it did a total recalc....Ill work on being able to show the graphs with no anchors properly

[steady horizon]

How do you feel about op small team lobbies? Is 8v8 the only mode that is currently interesting to you?

[lapis wing]

it should have just broke our party

[lapis wing]

28.63 is duplicated and im gone from the sort

[scarlet plank]

yeah that was probably my fault I input the numbers manually

[scarlet plank]

I swear the only difference between the AI's and high school interns is the fact that "something" actually gets produced by the AI lol I corrected the mean calcs as well.

[proud wigeon]

This shouldn't be a problem that is solved by making the OS matchmaking system less mathematically fair, OS is not a progression system it is an estimator of skill relative to the playerbase and is not meant to be infinitely gained.

The issue with this approach is that if someone comes along who is truly good enough to hit these extremely high os numbers playing every game and never dodging (hence having a properly accurate OS number) they would then stomp every game because they are being considered to be orders of magnitude worse than they are for matchmaking purposes.

You shouldn't be winning more because your OS number is higher than theirs, you should be winning more because you are better. (At a large scale ignoring game-to-game variance)

If a player is unable to carry the teams they are getting placed on, then their OS is likely inflated from where it should sit. (I would like to clarify I'm not calling anyone bad at the game or anything)

At ~2.8 Uncertainty every 5 OS (~4.8) loosely correlates to winning 50% more in a head to head.

Statistically at 80 OS Emre should win 95.5% head-to-head against the next highest rated player in the playerbase at 68 OS (I'm going off bar-stats, may not be accurate on next highest), The balance system is doing its job and will always give Emre the worse team to try and balance out this fact.

[lapis wing]

how did the default balance not come up with the default sort ? avoids? looks like parties are respected too

[dark bane]

yes the system is fair, it's just obnoxious as every effort you make to improve and succeed is met with worse and worse teammates. it's like some toxic employer lol

The only healthy solution is to shed some os by ah, doing sub-optimal strats for 50 games

in 1v1, as you get higher and higher your WR actually goes up. because there's fewer people higher os than lower os for you to play. Which is more pleasant as you can enjoy a natural 55-70% WR that corresponds with your skill.

there isn't really a solution for this, though having a larger population makes the problem only affect the very top group of players.

If you want to have your success reflect your improvement you should just sweat to say 50 then chill back to 40. Then when you try different style you can compare your success with equal team balance as when you first started at 40.

basically super soft smurfing xD i dunno.

Playing with very high os = trying to win >50% of games with very bad team balance
Playing at medium os = trying to win 60-70% of games with okay team balance

[scarlet plank]

The data I am showing is from actual games with the algo already done via the actual game so yeah all avoids/parties/vibes are respected intialy, I am not running this data through the balancer, then applying !splitpros. As I lack the needed skill with elixer to do that accuratly. I have converted the process to python as I do not trust the AI to do it for me in these cases and started generating seeded simulated data and found several failure points. It was easy to improve the situation with !splitnoobs -> !splitpros for these heavily imbalanced lobbies, and include an arbitrary rule that forces the teams to essentially 'look better to humans' by smoothing out players but as I have tried to start moving towards something that would be more production quality I am rapidly reaching my skill ceiling here. Part of my goal is to use already approved code and methods to build off so I am not trying to re-invent the wheel. I have a few more ideas like trying to use lambda to be stricter vs my hoarding calls so it does a better job of prioritizing pure balance. It works well in highly imbalance cases but in ones that are already approaching 50% its not really improving much in my test cases (which is seed generated data) and I want to make sure its not making things worse for lobbies where it really shouldn't be used so it can actually exist in the wild. Its a lot harder to do that on the high end of the OS spectrum rather than just reducing people to zero OS as !splitnoobs can.

Plus you have to also keep in mind the default balance INCLUDES uncertinty, I am tossing that out the window with !splitpros as this is based off !splitnoobs

Imo the current version does make a worthwhile difference in our real world data, but if we are being honest as we should in this case, the data is the very definition of 'cherry picked' and even though it is those 'cherry picked' situations we are targeting, we can't have it just breaking the game every time someone makes a !splitpros call when they shouldn't

[lapis wing]

thats wild as fuck maybe we should cap the win prediction deviation that avoid can generate

[scarlet plank]

Yeah without touching the current balance algo as Balance:Parties Im in a situation where I can increase lambda to be stricter against hoarding (IE look better for a human who doesn't really know what they are looking at) or decrease it to prioritze 50/50 and I can force it to just throw out any case where it create a larger deviation than already exsisted. but I cant do everything at once, that was initially why I wanted a 50/50 algo as well.

[lapis wing]

the in game sort gave us 75% that game by respecting parties and avoids
then you bring an algorithmn that still respected parties, but traded avoids for splitpros and suddently we are at like 52%?

[scarlet plank]

I didn't respect parties, if it looks like it did that was a coincidence

[lapis wing]

yea I mean by coincidence it did respect parties, so parties were not the cause of bad balance

[scarlet plank]

Its going to be a combination of all of it in most of these cases where things are so wrong that !splitpros looks like a magical fix all. I can't see people's avoids either so I can't fully determine the cause of the default balance, I can only assume it was done correctly

[lapis wing]

seems like if it coincidentally gave out a sort that kept parties, then only avoids skewed the win predict from 52% toward 75%

[scarlet plank]

Probably but I can not see that kind of thing so its impossible to know, but its my assumption as well.

The overall idea is hey "this lobby is unplayable" so players can choose to !splitpros or they can select the pure 50/50. Splitpros will try to smooth the lobby to some degree without stacking it throwing everything but OS out and not stacking the strongest players team, and 50/50 would be an algo that respects uncertainty.

Trying to do both of those things with one command is not really compatible

[lapis wing]

yea I get that. as far as im concerned splitpros doesn't achieve what its meant to achieve, if overstacking pros on one side was negatively affecting the game outcome despite a fair win predict, then it just means that pros are underrated

and they are only underrated because of splitpro in the first place, so let them overstack one side, stomp, and gain os to reflect their actual contribution.

for example there is this bacondude guy who is underated at like 38 os. if he plays exclusively in a lobby where he forcefully gets split against another underrated ~38 os guy, like yitiaoxianyu, then neither of them will climb, both will remain underrated forever

[scarlet plank]

The real purpose behind !splitpros is to assume everyone below the next lowest 10's bracket of OS in a match is essentially 'a noob' It tries to turn situations where anyone valued in the "second highest OS 10's digit" are the only players worth counting in the match so that high OS outliers do not encounter situations where they would essentially be 1v4 or 1v5. You would always get at least 1 more player of the next highest OS bracket than the default balancer would give you in those specfic situations thus evening the teams player for player, while refusing to ever give the highest OS player MORE of those bracket 2 players than the lower OS team, which the current balancer absolutely will.

[lapis wing]

I dont think you can look at what the game quality would be for a specific balance and draw conclusion on how the balancer should be. some players are underrated and some are overrated. if we had a fair 50% chance and every game's os gain loss was the same, there would be a couple weeks of pure stomps until outliers (overrated and underrated players) gradually shift toward their real os

if emre is getting stomped 1v4 thats because the balancer is doing its job, slowly

there are probably 1 to 3 overrated player on his team or 1 to 3 underrated player on enemy team and this stomp is necessary to bring everyone where they deserve to be

[scarlet plank]

Well recall how we got here to begin with -> large numbers of complaints about players being unable to play together, being forced to play with people they didn't want. Changing those things made the majority of BaR players happy, increased the number of over all players, improved the life of everyone in clans ALL at the cost of actual balance. It really isn't all about Emre here but it does directly apply to his games often. That's exactly why i agree with you that we need a 50/50 algo for more competitive play.

If we think of it like widgets=0 youy can imagine where the higher OS players eventually move to the 50/50 algo as their default, which will certainly cause issues when boosted players from small bubbles try to enter but if they remain in that enviroment it will eventualy bring them back in line.

[lapis wing]

yes I remember but I dont think the implementation we did was precise at all. im sure I could scour my stream and see a few lobbies where my party was broken down, manually put the numbers in to find that a 50 - 51% win predict was possible while respecting party simply by 'overstacking pros on one side', but instead we got split and ended up with still above 10% deviation from 50%

it seems like we put way too much weight on splitting the higher os and respecting avoid then what is healthy for the integrity of OS

[scarlet plank]

I would think that in almost every case that sohuld be true. 50/50 is going to be very close to 50/50 nearly every time if you want it to be.

If you look at some of the data where I do the 50/50 compare to split pros its "close" but that is due to the lobby, there are other times where it will move the difference from like 2.0 OS to something like 25os.

[lapis wing]

and btw force splitting higher os has very negative downside too, especially when like you say boosted players from small bubbles try to enter, no offense to anyone but I've seen a handful of glitterplayer go on losing streaks as they try to enter the rotato bubble. if youre lucky enough to be the other forcefully split higher os playing in the same timezone, you get a free os train up as they derank, since the balance will force you against them every game

[scarlet plank]

ex:

This result should be thrown out with extreme prejudice if your goal is a true 50/50 but I would suppose that when taken at face value without the diff compares many people would select the bottom options as "more balanced" if they didn't know the actual values

[lapis wing]

i think it just shows most people would misunderstand the skill disparity between 82 and 46 os player. if it was 1v1, the 46os player would win 1 in 33 games

[scarlet plank]

Right this goes back to my intial assumption here: #1419465382053412885 message

and somewhat satisfys Chrono's idea here #1419465382053412885 message it also pushes towards 50/50 compared to the standard algo when things are really out of whack.

That still doesn't make it better than 50/50 but in cases where we don't have MM and we have whatever the whackjobs in any given lobby are going to be playing with, it should improve those edge cases

[lapis wing]

but os gains/loss are not linearly proportional with win predict so that would inflate superhigh os players

[scarlet plank]

100% it will when used frequently, but again if you wanna play every game with !splitnoobs, then your higher OS really doesn't matter for the overall balance as you are anchor'd and not considered when the the non-anchor players are balanced.

So you can keep gaining 1million OS, eventually the number will be hidden anyway infavor of an actual ranked system, and it wont matter in the mean time cause you're not even part of the equation for the rest of the team

OR the culture can shift to 50/50 for high OS and things will sus out in a few months or even rapidly after a seasonal uncertainty resest in a primarily 50/50 enviroment

[lapis wing]

why just for high os? if noob lobbies are allowed 90-10 balance, then the innacuracy of their os will trickle up in the high os lobbies anyway at some point right?

[scarlet plank]

Yes but due to the general controls on lobbies already present in BaR, think max 25os lobbies, the only time this is benefitting those of mid range os is when someone is like 28-35 OS and playing in All Welcome full of sub 20's as an outlier. These cases are already inflating people as is to some degree so we really are not creating any new problems, and with the generally limited access to lobbies like that, its at least a slower process, and tbh boosting a few more scrubs into that low 30's mid range doesn't seem like a bad idea as people in those ranges are already begging for more lobbies to play in. They will just become lobsters in a slightly larger pond so to say.

I assume those mid range lobbies will also tend to adopt the 50/50 algo as they have adopted widgets = 0 and should create something of a gate keeper for 40+ os, the changes to apply_change_ratio/3 we use to smooth out gains/losses per game from uncertainty in the current algo have already inflated general OS values across the board anyway and drawing the player base into a wider range of values between 0-100 could very well improve overall game quality

[lapis wing]

you hear that @languid spire ? get to work, 100 os is the next step

[scarlet plank]

I gotta go for now, but I do really appreciate the questions here as we have gained a coherent examination of the problems and current state of things

[lapis wing]

I do think noob lobbies being allowed wider deviations from 50% is ultimately unhealthy for everyone.

when an inflated or deflated noob's match outcome diverges from the prediction, up to 15 other players rating gets pulled in the wrong direction.
since noobs mix with mid lobbies and mid lobbies mix with high os lobbies, a single isolated noob anomaly can quickly cascade upward and make the balance of lobbies across all spectrums feel less accurate.

[scarlet plank]

https://github.com/26Projects/teiserver/blob/master/lib/teiserver/battle/balance/split_pros.ex

K it's created on my branch at least... still need to do some actual environment testing then it will at least be in good enough shape to be worth some of the balance teams time to look at

[outer hornet]

What is a pro

[scarlet plank]

Anyone significantly beyond the existing lobbies average balance

Technically the top 2 os players in the lobbies are "the pros" in that system. If they are not significantly beyond the rest of players then this won't do anything but push the lobby closer to a 50/50 split based purely on OS

[thin path]

yeah os kinda ruins the game when it comes to playing with friends the higher you are the worse it'll be I am not a high os player idk if you guys consider 41 to be but the issue is there and I see often users like emre, cheetah, chevrolet and other actual high ones get on teams that are for sure going to lose and the majority of the time that's what happens.

also a few hours ago I read something kuchy said about the team deviation and how the team with lower chance of winning is always toxic the thread is there and is quite visable

[scarlet plank]

#1419465382053412885 message This example and the one below probably sum it up best

This is built on top of !splitnoobs so we are using approved methods already part of the current balance system, many of the reduction in OS spreads are more a result of not respecting avoids and parties rahter than the the command itself, and it has protection for cases in which is really shouldnt have been used.

[lucid meadow]

I love games with 70 - 30% win probability balance

The algo should take the next step and make it 100-0% for true forced 50-50 win-loss

[untold wing]

I mean clearly the system is working because emre keeps winning... but hot dang 80 os is crazy. Also, I have seen emre loose a fair lane vs a 40 os get completely wiped and get carried by his team. Which is also crazy. This guy is like a main character in a tv show.

[languid spire]

op %13

[scarlet plank]

Gotta stop playing in noob lobbies man 😅

Im about to get on a plane so won't be messing with anything until next week.

[languid spire]

omg im playing usually +30 min lobbies

[ashen osprey]

noob is a relative term @languid spire - surely in those max 20 os lobbies there's people calling eachother noob

[languid spire]

holy molyyyyy

[cedar void]

Game is trying to make you lose, but the fact you won indicates it's failing

Really the people hurt by this are your opponents, who lost more OS becuase they lost an expected win.

[languid spire]

Why does the game want me to lose, game isbitch?

1 more

[solemn heath]

holy did you win that they got the exodia for balance

[languid spire]

lose

[lucid meadow]

Smh this clanker algorithm

[proper spear]

Im gonna say it: its your own fault for tardwrangling poops by spamming straits
We warned you that farming OS there wont bring you any joy and you ignored us.
Now deal with the consequences

[dark bane]

Pretty sure Emre got most of his os in rotato. Unlike most os farmer he actually won on strait too, kinda insane

Watching the games it was like he had a crystal ball and knew which team would win

[solemn heath]

Emre too very op

[placid sequoia]

plot armor is OP

[scarlet plank]

I guess the thing Im wondering most about is the conflict between 1. The system is working correctly trying to make games harder and harder and thus contain the ever growing exception - or - 2. Once you get significantly beyond X os over or under the generally accepted min or max it doesn't matter anyway. Both of these really can't be true at the same time, and there is no real room for the gray area one might try to claim in between in implementation without substantially arbitrary choices.

Personally Im interested in trying to spread the range of OS out more as I believe it will be good for the game overall. I feel like we have arrived at a pretty comfortable rate of change this latest season with the OS change for wins/losses per game, so I don't think attempting to increase the variance in OS spread via that method is a good idea.

[serene belfry]

I think as long as he plays in min os lobbies there is a theoretical maximum to how difficult his games can get. In the worst case he has to carry 7 30os against players 2-9 in the ranking.
Not sure if he can gain os forever but as long as he plays min os lobbies he should theoretically be able to maintain a higher than 50% winrate right? Would that count as the system failing? In any case, Emre to the moon!

[scarlet plank]

That is essentialy the question I am asking...do we consider that failing due to the arbitrary contraints of "min OS" a true match making system would have started giving him weaker and weaker opponenets, say 10-15os "certain" players to prevent someone from ever getting so far out of line as the top 1% (or whatever was chosen) of the player base. This is how you get matches in other games with 'pro players' + total noobs getting thrown together into a 'pro looby' so do we want to allow people to inflate in the context of Min OS or not.... that is kind of the question we need answered before moving forward with any real change to the system.

[thin path]

fix os please

[blazing stag]

I have read most of the messages in this chat. But one thing that remains unclear, to me, from this discussion is. Is there a reason to not try to balance every individual game to be a 50/50 win/loss prediction and balance teamgames that route? Am I missing something fundamental? As people seem to be talking about the game trying to drag emres winrate down to 50/50. Or is the point that the game does this, as a result of a "flawed" metric it is using to balance (OS), in this line of thought?

[lapis wing]

I think the answer is that bar doesn't have a competitive mode right now and some people would prefer the balance prioritize things other than a fair 50/50. such as respecting party and avoids, splitting noobs apart, splitting captains apart.

my opinion is that for party and avoid the balancer should only respect them up to a cap of like 55-45
and splitting noobs and captain shouldn't be a thing

[cedar void]

There's also player enjoyment to consider. Casual players enjoy a game more when it's no a super close 50:50 every single match. That can be incredibly tiring and leave no room to experiment.

Many games purposefully occasionally give you an 'easier' match to lessen the pressure and allow you to relax a bit.

It also let's people 'climb' easier by giving them the corresponding difficult matches and a chance to win extra OS with good performance.

[hasty estuary]

In any big competitive game your WR should be 50%. Why should it be different for BAR?

I don't quite understand the problem here.

The fact Emre's OS keeps going up is an indication that regardless of the fact they're on "bad" teams they're overcoming that disadvantage anyway. Why on earth should we change the balancer in light of that? So they can win more games? That sounds unpleasant for the enemy team. Sure, from Emre's perspective they're doing everything right while every other front is collapsing, but from the enemy team's perspective they're winning everywhere but getting run over anyway (surface-level - idk what Emre does to win, presumably lots of helping the team etc).

If the balancer was adjusted it seems like it would just make it unpleasant for the other team instead.

The goal should be for 50% Win chance for either team because anything else will probably feel bad for the disfavoured team and not be that much fun for the winning team either.

[scarlet plank]

Its important to note that even if we did the 50/50 strictly, it would not be 50/50 often. it would be CLOSE to 50/50 often. Crunching all 62,000 possibilities everytime someone puts !balance into the chat or clicks the button in a match would have performance costs.

I do not personally know of any popular game with competitive modes that actually balance on a strict 50/50. What I did with !splitpros is simply a killswitch to push things closer to that 50/50 point (when compared to the current algorithm) while ensuring that the highest OS player does not get stacked. There are MANY ways to get within 1-2% of a 50/50 OS split in any given lobby containing 16 players and they are not all equal in outcome.

Consider what actualy happens where there are 4,500+ different ways that you can arrange 16 players to be within a 50/55% split that you have to find out of a set of 62,000+possible combinations and then check to see which of those 4500+ is actually desirable, and do it over and over while little Timmy is smashing the balance button

[blazing stag]

why? This is not an argument, just a statement.

[serene belfry]

It is also not a true statement. Chronopolize was talking about 1v1, the best 1v1 player will by definition never have a 50% winrate. (Example: AoE 2 Hera currently enjoys about an 80% winrate in the 1v1 ladder and that is not a small game.)

[blazing stag]

The fact Emre's OS keeps going up is an indication that regardless of the fact they're on "bad" teams they're overcoming that disadvantage anyway. Why on earth should we change the balancer in light of that? So they can win more games? That sounds unpleasant for the enemy team. Sure, from Emre's perspective they're doing everything right while every other front is collapsing, but from the enemy team's perspective they're winning everywhere but getting run over anyway (surface-level - idk what Emre does to win, presumably lots of helping the team etc).

If the balancer was adjusted it seems like it would just make it unpleasant for the other team instead.

the part you are overlooking, in my opinion, is that (apparently, if I understand correctly) with the current balance it is essentially a death sentence most games to have a super high OS player on your team. as the game is trying to punish him into losing, making your team substantially worse. essentially creating games where according to the game you are not winning. ( at least some of them, I have not done a statistical analysis of Emres games and his win chances). now we have the inverse effect, all lanes are "statistically" getting steamrolled except of the high OS lane, with these 7 players also not enjoying getting steamrolled most likely.

I agree that it does not matter that Emre specifically is getting fucked by the balance (no offense). If 1 player get's screwed because they are abnormal, AND it is better for the health of the game then I see that as a "non-issue".
However I see unbalanced games as not being good for the health of a game. If your team

The goal should be for 50% Win chance for either team because anything else will probably feel bad for the disfavoured team and not be that much fun for the winning team either.

This is different from what you seem to be arguing for in the rest of the message, as well as the message before. So I am a bit confused.
The point/complaint is that current the game doesn't try to “strictly” force the matches to be (as close as reasonably possible to) 50/50 win chance. Is this (he matches to be as close as reasonably possible to 50/50 win chance) also what you are advocating for?

slowmode is very annoying.

[blazing stag]

my opinion is that for party and avoid the balancer should only respect them up to a cap of like 55-45
and splitting noobs and captain shouldn't be a thing
seems reasonable.

[blazing stag]

I´d argue that getting perfect balance shouldn't be expect, if that is indeed too computationally expensive. However I would argue some of the examples shown are not acceptable (e.g. the 95 vs 5 example). And I think kuchy's suggestion is reasonable. although I would intuitively make the bounds bigger. 40 vs 60 does not feel outrageous to me. However that is just a guess not backed by data or experience.

[scarlet plank]

Im not sure I fully understand what you are saying but my point was a technical one in regards to how the computer and code actualy operate when assembling a team. The odds of having multiple near 50% configurations of 16 players is high. You might not necessarily get the "best one" first. Dealing with all of that takes time and cuases the client to hang up if it gets excessive.

[outer hornet]

Would like to see stat of how often the win predictor was right

[scarlet plank]

What would you consider right? IE, if a team has at least a 51% chance to win, and they win the system is corect?

The general accruacy of the prediction is pretty much the number itself. The system will tend to accurately predict the outcome of a team with a 60% chance to win 60% of the time

The odds of 1% and 99% predictions being correct will still fall into a system of normal distribution (and any time the real data says we are wrong we just say: Not enough data )

[outer hornet]

Os is super unbalanced janky number to base accurate win % off. A 30 os player could vary greatly in skill level depending which silo they play in. I highly doubt if system says 60 / 40 it’s actually that depending on context of players, lobby, and map

[scarlet plank]

Yeah sure but if we are trying to be specific, you can't just then declare the system is broken because 'silos' then use congregated data across all those silos to point and say SEE it's broken, you have to look at each silo

[outer hornet]

Didn’t say it’s broken. Just saying people giving way too much power to these %s

[scarlet plank]

Well I do keep proposing we get rid of OS, and just start rating everyone in CCRafus/m scores.... we could require Blood alchohol content reports to give you a sliding scale OS value hehe.... I mean realistically what are we supposed to do but track win/loss over time

[outer hornet]

Only have ranked on in controlled environments

[scarlet plank]

Ok fair idea, sounds logical enough.... but what is a controlled system here. It's still going to be subject to nearly all of the same conditions of the win/loss outcome and doesnt account for me playing like butt today vs popping off yesterday..to keep going we have lots of example of Match made games and people find those systems inadaquet as well

IF we get BaR specific then turning off "ranked" means that the outcomes are not going to be accounted for at all, and we will be doing rnd(16) for team rosters I dont see that as an improvement

[lucid meadow]

Ye do all rnd so can abuse party harder

[outer hornet]

People shouldn’t be allowed to play 1 spot 1 map for 1000 games. If they ever do anything else that os is meaningless. Not to mention you can play custom meme modes with ranked on

[scarlet plank]

So if we started force specing people in lobbies if the map is supreme, and they have played supreme the last 12 games? (Im liking this idea more and more now) lol ...Im not trying to be rude here Im just trying to walk though the perspective

I've sat around thinking about it quite a bit and after looking at all kinds of stats that we track in the game and trying to find the ones that are the most predictive of outcomes... you just keep running into problem after problem trying to deviate from anything beyond win/loss

Match making IS coming... its even kind of working, it will address alot of the siloing issues, but its going to bring a list of its own problems as well

[outer hornet]

Yes so you know that it’s all unreliable without controlled match making. I never said count anything beyond win loss

[scarlet plank]

Yeah i am exploring the idea here - if we limit the types of matches people can play, the maps they can play, the modes they can play, then that is essentialy a new que. This walks us into seasonal ranks with pre-curated map lists and settings, then you have to ask...what about the ISthmus/ATG guys. Do we have an eternal que for those maps? Do we reset it with each new map verison? Do we expand it to map specific rating? Does that mean we que up a 40/50os player as a 16os player on a map they have never played

it gets really messy fast even when you feel like it should be simplifiing things

[outer hornet]

If you want to play 1 map only then you go into unranked custom lobbies simple

[scarlet plank]

You would still want those lobbies balanced though yeah? If we have no ranks or ratings, then we are literally rnd(16) for teams

[outer hornet]

If you turned ranked off it still balances the teams

[scarlet plank]

How so?

[outer hornet]

Idek why you’re trying so hard to argue with me lmao. Everything you’re saying is agreeing with me that os is completely fucked therefore the win % doesn’t mean much

[scarlet plank]

Mainly cause I am genuinely looking for ideas here and you hadn't quite reached the point where you seemed annoyed with it yet lol... If we don't track win/loss and people dramatically improve then the numbers by there names that will be used in a real world BaR scenario of unranked play become even more completely fucked as you put it

[outer hornet]

System shouldn’t be based around unranked fun games. You can grind norms in league and your rank won’t go up.

[lapis wing]

os is completely fucked because of win%, not in spite of it

win% matter because it aggravates the inaccuracy of os and makes it cascade on itself

[hasty estuary]

A death sentence and yet their OS is still going up.

Something is clearly not squaring here. If they were losing games their OS would be going down, but that's not happening. So something is not right about the story being presented. As someone else mentioned, if they're dodging games they think they'll lose based on the teams that would only make the problem they'•re having worse, since their OS will be higher than it should be.

Mechanically for a matchmaker the ideal is generally a 50% win chance on either side. You can compensate for difference in win chance by adjusting the OS reward on the other side but for players you want a game they have even odds of winning because that's presumably more interesting.

[delicate light]

parties, avoid system, unpunished dodging (how about a 2 min cooldown to join queue) mean that theres no point in blaming the OS balancing algorithm until this is dealt with.

[lucid meadow]

Ye put 2min cd on rejoin queue in an already dead game

[tepid mango]

League of Legends has very fucky matchmaking and it is abhorrent, streaks of 10 wins and 10 losses one after another; to some extent makes it feel like agency is out of my hands and rather in the hands of the matchmaking deciding which team gets more consistent players.