Reasonable summary. Rigged backpressure. It arrests and attenuates mobility (hard to climb, hard to drop). Keeps games “close” for the sake of engagement (gambler’s ruin). The same washed content creators who were grandfathered in 3 years ago get to keep their ranks well above the clouds, away from actual climbing hell in 2020 4+ years no-reset esports integrity. Going 50/50 is a lot different than e.g. solo climb your signal through all the noise.
That’s just semantics. It will still classify +3/-3 s.d. climbers/derankers and knapsack according to recent trends.
Not in terms of collision probability. Part of what makes you top 10% correlates to being online a lot and playing lots more games. That means more alts/burner accounts from the upper % generating interactions with the middle/lower %. A single Masters player can show up and disrupt countless gold/plat matches, per role, per account - because frequency of contact.
yeah man, checking my corners leads to data points. using natural cover creates data. no dude, it doesn’t.
how? tell me, how would you define confidence? not even leading to datapoints, what is your definition of confidence? its abstract by nature my guy. its a feeling.
again, no it couldn’t. at most, data would show that the performance never dipped. thats not exactly empirical evidence of adaptation.
logic and observation based on a non quantifiable external stimuli. aka, something an algorithm cant detect.
yeah they kind of are. interpretation of body language or its in game equivalent is not something an algorithm can do. not without years of development and scientific expertise. these guys are devs, not scientists.
how would you go about programming contextual decision making? this isnt a game where there is a single “right” answer.
see now this is where your argument fails. we aren’t talking about ai. we are talking about bots. there is no intelligence, therefore it can not understand.
If that Masters player is consistently out performing people in their matches such that it could possibly disrupt the ladder, then the MMR system would take note of that.
Even if it doesn’t immediately move the player out of the rank, it would actively mitigate their impact by increasing their MMR, and increasing the expectancy that they would win a given game and perform well.
The system notices when there’s a large discrepancy in player skill, and as I’ve said before: there’s more data point on gold and plat players, so it’s way more confident on gold and plat rankings. Anyone not performing to that level will be marked, and their impact will be mitigated.
If that isn’t happening, then the player is sandbagging. If that’s the case, then they shouldn’t be negatively impacting the stats of players in the games that they’re smurfing to to begin with.
I would also contend that this isn’t explicitly true.
I, for one, don’t actually play the game that often in spite of being Masters/GM. I also know that many high ranked players I’ve interacted with frankly have no interest in smurfing to stomp on worse players. Generally it’s exclusively to play with friends, or the occasional unranked to GM by a streamer.
In the instance that they’re queueing with friends, they likely aren’t trying their hardest, and are rarely hard carrying. This is the case for me: my gameplay on my main is drastically different from my smurf account that I play with friends on, because I’m just not putting as much attention toward the game as playing like a “GM” really requires.
If the masters player is actually performing like a masters player, then the system will see that 1 out of 12 players severely outperformed. It’s not going to punish the other 11 people for a single outlier. It’s instead going to start tracking the outlier into a different bracket.
It does, because session based mmr adjustments/revisions do not take non-local information into account (afaik).
For instance:
It will see this over and over as the mmr goes up, the disruptions happen, and the user hops accounts. Repeat ad neuseum. From what we can see, the mmr alg does not adapt mmr with input from non-local ladder conditions, like active smurfs, where other accounts are placed, etc.
Which means the system is absorbing more disrupt than it is shedding. The existence of a top 10% individual (being more likely to have more accounts and play more often) implies superadditive disruption.
What would this have to do with instance of the game which features the smurf?
This isn’t true. Again, one of the entire reasons why the Glicko system was created was to prevent this very thing. Not only do the accounts have to exist they need to be actively used. I’m sorry, but top tier players simply don’t have enough time to play on their mains, in addition to consistently playing on various smurf accounts. It isn’t happening.
They would have to sacrifice time on one of those accounts, which is going to gradually decrease it’s significance in the MMR system as it isn’t being used.
Again: Play smurf, main is less impactful, ladder gradually shifts as a high-ranked data point disappears. Play main, smurf is less impactful, ladder doesn’t shift because a single momentary smurf account isn’t enough data to do anything with.
The system absolutely does consider where other accounts are placed. You can see this in the TrueSkill documentation. It places player’s skill level as a gaussian distribution against the environment (the ladder).
You can also see it in the original Elo system which seeds players with a default value then derives their performance in relation to other players with a lower K-factor (meaning greater confidence).
This is quite literally how you are ranked. Without considering these factors there would not be ranks. TrueSkill and Glicko take it even further by considering the relative weight of accounts more precisely than Elo does, by looking at factors such as time played in addition to absences (along with other factors), and creating a range (denoted by Sigma in both systems) rather than a point of rating like Elo does.
Let’s differentiate Diamond from Masters+ for a moment, because there’s an enormous skill gap.
Masters+ players make up roughly ~4% of the playerbase. Contrast this to the 25% of Plat.
In order for Masters smurf data to outweigh Plat player data, every single Masters+ player would have to play on a smurf account 6 times as much as every Plat player plays in general to outweigh the impact of Plat players, in addition to also playing on their own accounts with enough frequency to maintain the MMR system’s confidence in their own main account rankings, and maintain the upper ladder.
Consider the fact that high-ranked players tend have a higher percentage of players streaming, of which the vast majority of their time is spent on their main account, even if they do the occasional Unranked to GM (with the exception of Somjuu, who seems to only play in Plat anymore). Add in the fact that a large portion of the upper parts of the ladder also play in a large number of PUGs and scrims as there’s also a larger number of these players on Tier 3+ teams, and now you’ve seeded some of them directly out of the ladder for significant portions of time.
They simply could not have the significant enough impact required to manipulate the mid-lower ladder in any noteworthy way.
You are of the belief a top 10% player shows up in ~10% of matches. This is one point of contention.
On live, mmr adjustments aren’t taking into consideration the additional collision frequency of top tier players entering normal tier games. The disruption effects aren’t being properly accounted for.
We would have to see the numbers. The working assumption is that a top 5-10% shows up a lot more, and on more accounts, then Normie 40-60%. Yes there are many Normies and their own enthalpy is supposed to keep them nearly in place. Yes those overperformers are supposed to bubble out. But as you respawn more of these overperformers the impacts on normies become more lingering, as you are actually changing that landscape. It is superadditive population dynamics - because the encounter rate is constant-sum but the mobility is zero-sum. Glinko isn’t solving this iirc.
Why is that the assumption. What do you base this off of?
This does not matter. It’s unimportant. Accounts are not the sole creator of data points, individual match results and player performances over time are. 2 accounts that have played 1 game each are less valuable than 1 account that has played 2 games because the 2 accounts have significantly less confidence in their rating. The system inherently cares more about confident data than simply having data. This actually makes the account subadditive, not superadditive as you claim.
It doesn’t matter if a GM player has 3 accounts, or 5 accounts, or 50 accounts: they are still only a single person, and they are still limited by the actions that a single individual can take. I won’t be addressing this again.
And, again, the overperformers will be singled out by the system. That is the entire point.
The system will literally make their performance against a lower rated player less important for the lower rated player. If I make a new account, place into Plat, and win 10 games in a row by absolutely stomping everyone, the matchmaker will not create a significant negative impact on their accounts because of that. It would instead increase the rating of my account, and increase the confidence that I’m better than them, because I clearly don’t belong in those games. This is obviously an exaggerated case, but it would apply over large or extended samples as well. If a player continuously outperforms then it inherently increases the expectancy for them to win, and so enemies will be punished less for losing against said player. This is incredibly basic.
Please, explain why that would be? MMR and SR have been very clearly demonstrated to not be zero-sum before.
If you flip a coin a dozen times and the first 11 flips give you heads, does that mean you’re more likely get tails on the 12th? No. It’s a fallacy. The same is with the matchmaker and smurfs / throwers.
Look at the big picture. Track your games for a season. Competitive matchmaker only looks at your SR outside of placements.
Yeah you have, and I wasn’t interested the first time. I already stated I’m not interested in discussing the particulars of TrueSkill. You already acknowledged that one implementation of TrueSkill can vary from the next with different significantly different results. And yes we’ve been over this, so I’m not sure what you’re on about.
Finally, you’re defining a smurf much more narrowly than I am so I don’t agree with your math. I think we’re done here.
I don’t recall acknowledging anything of the sort, but regardless:
THIS IS THE ABSOLUTE BASIC OF WHAT TRUESKILL DOES, and it ISN’T EVEN A SOLE FACIT OF TRUESKILL. This is literally the basics of how Bayesian matchmaking systems (read, essentially all of them) work.
Feel free to ignore the literal leading research in matchmaking all you want, you do not know what you’re talking about. Keep trying to pretend to be some arbiter of stats all you want with your entry level programming knowledge, you’re clueless bud.
For everyone’s sake, just stop trying to argue. You can’t even participate in the basics of the conversation and continue to ignore simple facts of all matchmaking systems.
Looks like you’re getting upset – we are talking about a video game. Maybe take a breather and calm down a bit. And yeah I – literally – just said I’m done with you. But maybe take your advice and just stop trying to argue.
Eh, not really upset armchair psychologist. Just not sure how else I could possibly get it through to you that you haven’t the slightest idea what you’re talking about.
And I mean-
You quoted my response that wasn’t even directed at you, bud.
Doesn’t take a degree in armchair psychology to see when someone’s getting excited/bothered.
And you’re under the strange presumption that I consider your opinion relevant or important. I’m done bickering, I see you could continue on with this all day.
Then don’t reintroduce yourself to a conversation that hasn’t involved you just to try to get the last word in. You can’t just run from a lost argument by trying to grandstand.
That’s the theory but I have to agree with him. Only yesterday 3 of my friends and I were matched against 6 level 28/29 players who were all Plat with private profiles. It didn’t take a genius to figure its not their real rank yet the Matchmaker gave us a gold tank.
Yes, I would agree with that, but there are many other processes involved that have an influence on matchmaking.
The reporting system, something like an internal secret “social score”. Statistics for maps, role dependent statistics. And, of course, how your games performed in the last 72 hours in terms of quality.
On this basis, session times are also measured and predicted in order to optimize matchmaking. Cheater and Smurfs also play a very special role and are tracked across accounts with “anonymized” data.
It is easy to spot a smurf, and it is also possible to detect if a player cheats on another account. The same processes can also be applied to groups.