Part 1: Questionable Matchmaking Strategies

Competitive Discussion
02/11/2018 03:54 PMPosted by S23
The video I'm linking to is a Blizzard matchmaking programmer. Keep in mind this video was shot before Overwatch existed so Overwatch is not mentioned, but other games like StarCraft, WoW Arena and others are mentioned. It's reasonable to consider that Overwatch is using some of the same strategies that I'll be discussing in posts over time. Here's the first one:

https://youtu.be/y4TjH2SdDOQ?t=1974
There's an approach where it's [the matches] are still kind of 50/50, but you don't always give a 50/50 match. You give like a variety of matches. You give sometimes, you have some smart way, you say no for this match I'm going to give him actually a hard match, for this next match I'm going to give him an even match, for this next match I'm going to give him an easy match. That was motivated somewhat by my past experience where I used to run Wolfenstein Enemy Territory server back in the day in college or whatever and it got to the point where the matches were super even all the time and actually after a while it gets actually kind of tiring and so I kinda came up with the opinion that it's ok to allow a little bit of a balance here and there and have a mix of even hard and easy matches.


If this is implemented into Overwatch, what does it mean? Well, if you are in a match that is purposely created so that it is very difficult to win, what would you call that? Keep in mind they have the option of giving you a fair match, but his opinion is that gets tiring so instead they place you in a match with almost predetermined odds so that you get a mixed bag of results. This is by definition rigged matchmaking.

Here's the next excerpt: https://youtu.be/y4TjH2SdDOQ?t=386
The goal of matchmaking is for it to be fun for players, and secondarily it is for it to be fair. You might think that sounds weird because it needs to be fair or it's not fun, but it's really about player perception. Sometimes things that are perfectly fair to players, they don't think they're fun.


He goes on to say that he imagines what people think is fun, and then tries to create a matchmaker that provides that mathematically; this would be an artificial matchmaker. This idea of care bear matchmaking where everyone gets to win and we are trying to make everyone happy is a breeding ground for toxicity. We already know from the first link that they are able to place us in matches that are easy to win. Well if you are trying to make everyone have fun, and this is more important than being fair, then you would give players who aren't doing well easy to win games so they have "fun". You are often placed in matches that you were never meant to win, because "it's time to let someone else win for a little bit".

The matchmaker you are playing in is has good intentions, but is bad. I do not think Blizzard is out to get us, I think they literally believe they are doing what is most "fun" for us by rigging matches in a way to give us what they think we want. Remember the Mercy buff that made everyone so upset? When they made those changes they thought they were doing a good thing. It wasn't until the public responded that they finally changed it. We need to be more vocal about the matchmaker. Something is wrong with it and it needs to change.

This is supposed to be an e-sport game. A truly competitive game. Every person can agree that we want an honest matchmaker, we don't want something that sometimes gives you free wins and sometimes gives you forced losses to add variance. Blizzard, something is wrong, there are posts about it every day but instead of an honest matchmaker we are getting new skins and voice lines. You have actually had to implement a built in anti-toxicity reporting system because people are so pissed off at your matchmaking. Quit putting band-aides on the matter and go to the root cause - your matchmaking needs improvement.

If you're interested, watch the video in it's entirety, there's a lot more questionable strategies coming in this one, and another video that I will detail in part 2 and so forth.


Regardless of if it's true or not, it doesn't matter. You don't improve unless you have reason to improve i.e. going against harder people. The only real issue is the one tricks and hero mains that people have to try to build around.
Sometimes the matches are godawful and sometimes the matchmaker is better than I thought.
For example today I played on my low-mid diamond Doomfist/Pharah account and we had 2 other alts on our team. I really thought the game would be a total stomp in our favor.
But it seems like the game accounted for that and the other 3 nonalts in our team were total bots and way worse than low-mid diamond.
The game was essentially a 3v6 and we only won by a small margin.
02/12/2018 08:10 AMPosted by S23
...

There are streaks, yes. But they are not rigged. They are a natural consequence of how random numbers work. See https://us.battle.net/forums/en/overwatch/topic/20759379305 for a detailed analysis. Show me your SR history (by game) and I can run your data through my analysis as well.


lol, I just looked at this post, what is the goal of this? Are you claiming that your analysis can detect abnormalities? This just explains basic math principals and is done in an overly convoluted way. Explain to me clearly what your analysis does (other than say "there are no strange patterns here"), what do you think this proves?


Yes, my analysis can detect abnormalities. What the analysis shows is that the streaks that people see are consistent with a system without rigging: where win probability depends only on rating, with a 50% win probability occurring when a players MMR matches their true rating.

What my (and others) experience climbing shows is that the system responds to players who improve.

By the way, if you want to claim how much time you have put into the game as evidence, you need to post from your main.
I've played 98% of my matches this season soloqing and I was always near a 50% winrate. With 10+ heroes that have 50%+ winrates.
For every gaming session my stats were to the roof in oversumo.
Playing like my life depended on victory.
50/50 win/loss always.

I decided that I would not soloq again.

Teamed up with buddies who are at the same level as me and all flex players:
10 wins in a row.
Joking around while playing. Actual. Fun.

If you ask me, don't soloq. Team up with flex players. Simple as that. :)
02/11/2018 10:02 PMPosted by S23
02/11/2018 09:44 PMPosted by Kaawumba
...

Yeah, that's how it works. Except matchmaking is by MMR, not SR. https://twitter.com/playoverwatch/status/850435344457543680

Of course, since SR tracks MMR closely ( https://us.battle.net/forums/en/overwatch/topic/20758686566?page=1#post-6 ) this is a distinction without a difference, except when a diamond+ player is decayed due to inactivity.


It's your right to believe everything you're told, and there's nothing wrong with that. As for me, I use critical thinking, I listen to other peoples explanations and keep it in mind while I observe and let my observations speak to me and tell me what's happening. I don't know exactly how it works, but I am sure there is more to it than what we're being told. Whether accidental, or on purpose, there are rigged streaks (win and loss) in this game.

When you have time to put in as many hours into this game as I have, the streaks will stare you in the face and you'll have to make a choice, do you swallow your pride and go back on everything you've been defending, or do you convince yourself everything is working as you've been told.


"Open minded". Here friend, a video: https://www.youtube.com/watch?v=T69TOuqaqXI
<span class="truncated">...</span>

lol, I just looked at this post, what is the goal of this? Are you claiming that your analysis can detect abnormalities? This just explains basic math principals and is done in an overly convoluted way. Explain to me clearly what your analysis does (other than say "there are no strange patterns here"), what do you think this proves?


Yes, my analysis can detect abnormalities. What the analysis shows is that the streaks that people see are consistent with a system without rigging: where win probability depends only on rating, with a 50% win probability occurring when a players MMR matches their true rating.

What my (and others) experience climbing shows is that the system responds to players who improve.

By the way, if you want to claim how much time you have put into the game as evidence, you need to post from your main.


hah, I guess you thought nobody would question your "analysis". It's just a bunch of charts that have little meaning to anyone. It's literally just plotting wins and losses and drawing a line, that doesn't tell any type of story at all. You can't just look at someones data from playing comp and "detect" if there is something rigged. Especially if it is rigged in a way to simulate an organic matchmaker.

So let's test it. Show me a hypothetical example of a series of wins and losses e.g. "wlwllwllw" that your analysis would detect as abnormal.
02/11/2018 10:02 PMPosted by S23
I don't know exactly how it works, but I am sure there is more to it than what we're being told.


If you're getting forced losses, someone else is getting forced wins. You don't understand anything. 2 bottles of snake oil please!
02/12/2018 12:30 PMPosted by Tactician
02/11/2018 10:02 PMPosted by S23
I don't know exactly how it works, but I am sure there is more to it than what we're being told.


If you're getting forced losses, someone else is getting forced wins. You don't understand anything. 2 bottles of snake oil please!


Of course, that's been stated 100's of times on this forum. Forced losses and forced wins are a thing, how do you not know that? Show me a log of your games and I'll show you loss streaks once you reach a new tier.
02/12/2018 12:45 PMPosted by S23
02/12/2018 12:30 PMPosted by Tactician
...

If you're getting forced losses, someone else is getting forced wins. You don't understand anything. 2 bottles of snake oil please!


Of course, that's been stated 100's of times on this forum. Forced losses and forced wins are a thing, how do you not know that? Show me a log of your games and I'll show you loss streaks once you reach a new tier.


Unless you're matched against the same opponent for an entire loss streak there's no requirement that another player gets a win streak to mirror your loss streak.

The system could be saddling a small number of players with loss streaks in order to distribute wins across many players.

There's an inherent problem with match making systems for games like this, in that everyone wants to win more than they lose, but that's impossible, someone would have to lose more. If you consider that problem you'll see that keeping the maximum number of people happy is a balancing act that requires orchestration and manipulation the likes of which could never be made transparent.
<span class="truncated">...</span>

Yes, my analysis can detect abnormalities. What the analysis shows is that the streaks that people see are consistent with a system without rigging: where win probability depends only on rating, with a 50% win probability occurring when a players MMR matches their true rating.

What my (and others) experience climbing shows is that the system responds to players who improve.

By the way, if you want to claim how much time you have put into the game as evidence, you need to post from your main.


hah, I guess you thought nobody would question your "analysis". It's just a bunch of charts that have little meaning to anyone. It's literally just plotting wins and losses and drawing a line, that doesn't tell any type of story at all. You can't just look at someones data from playing comp and "detect" if there is something rigged. Especially if it is rigged in a way to simulate an organic matchmaker.

So let's test it. Show me a hypothetical example of a series of wins and losses e.g. "wlwllwllw" that your analysis would detect as abnormal.


wwwlllwwwlllwww..... would throw off the autocorrelation function, because an unrigged system would not only have streaks of length three.

A guaranteed loss after a specified number of wins would throw off the autocorrelation function, because there would be no streaks above a specific length.

Guaranteed 50% matchmaking would have people's SRs wandering all over the place. You would have top 500 players who end up in bronze due to random chance. People would not be tied to specific MMR/SR.

And so forth.
02/12/2018 01:17 PMPosted by Kaawumba
...

hah, I guess you thought nobody would question your "analysis". It's just a bunch of charts that have little meaning to anyone. It's literally just plotting wins and losses and drawing a line, that doesn't tell any type of story at all. You can't just look at someones data from playing comp and "detect" if there is something rigged. Especially if it is rigged in a way to simulate an organic matchmaker.

So let's test it. Show me a hypothetical example of a series of wins and losses e.g. "wlwllwllw" that your analysis would detect as abnormal.


wwwlllwwwlllwww..... would throw off the autocorrelation function, because an unrigged system would not only have streaks of length three.

A guaranteed loss after a specified number of wins would throw off the autocorrelation function, because there would be no streaks above a specific length.

Guaranteed 50% matchmaking would have people's SRs wandering all over the place. You would have top 500 players who end up in bronze due to random chance. People would not be tied to specific MMR/SR.

And so forth.


Ok, so the only thing your analysis would detect is if there was always a loss after max specified wins. So of course your analysis didn't find anything, nobody has claimed something like that exists nor has anyone given you a max number of wins which would then guarantee a loss.

I think you created this "analysis" hoping nobody would question it, they would just see numbers and charts and say "oh well I guess someone proved matchmaking isn't broken". You and I both know your analysis proves nothing, it's an overview of how random numbers work presented in a complicated way so that people won't question it. It proves nothing.
02/12/2018 01:53 PMPosted by S23
Ok, so the only thing your analysis would detect is if there was always a loss after max specified wins.


Did you see the "and so forth" at the end? Do you know what it means?

How do you think the system is rigged? Be specific.
02/12/2018 01:57 PMPosted by Kaawumba
02/12/2018 01:53 PMPosted by S23
Ok, so the only thing your analysis would detect is if there was always a loss after max specified wins.


Did you see the "and so forth" at the end? Do you know what it means?

How do you think the system is rigged? Be specific.


The biggest problem with your charts, is that it doesnt take into account alot of variables, including match quality, and how hard a game had to be carried for a win.

Further more, you havent yet been able to simulate the same kind of matching that we see in game. An actual 50/50 change eg: basic balanced teams would lead to a far better balance of skill and matches in tiers. You also didnt take into account the use of a bell curve in match making. They didnt say its a natural bell curve distribution, they said they use a bell curve, meaning there is some level of artificial implementation.
02/12/2018 01:57 PMPosted by Kaawumba
02/12/2018 01:53 PMPosted by S23
Ok, so the only thing your analysis would detect is if there was always a loss after max specified wins.


Did you see the "and so forth" at the end? Do you know what it means?

How do you think the system is rigged? Be specific.


I know what "and so forth" means. I don't know why you placed this at the end of your response, as a sentence of its own and without any context to refer to. Maybe you're using it some obscure way like your charts.

Specifically I think the matchmaker is doing more than just grabbing 12 people of similar MMR and ping and placing them into a match together. Anything beyond that is rigged.

One of the largest clues is the impending loss streak that usually (but not always) happens when you reach a season high, career high, or new tier. It's not just losing the matches. It's how they are lost. They were impossible to win.

whoever programmed this has made a massive oversight because we are all describing the same thing and it is not natural.

If you ask me, I have a suspicion that the loss streaks are triggered somehow due to long playing sessions and playing flex. Changing heros during the match seems to do something very bad, especially if you make multiple switches.
02/12/2018 02:14 PMPosted by IceQueen
The biggest problem with your charts, is that it doesnt take into account alot of variables, including match quality, and how hard a game had to be carried for a win.


I boil it down to a probability: 50% change to win at this SR, 10% change to win at that SR, 90% change to win at this other SR. yes, it is a simplification.

02/12/2018 02:14 PMPosted by IceQueen
They didnt say its a natural bell curve distribution, they said they use a bell curve, meaning there is some level of artificial implementation.


Do you have a quote on this?
02/12/2018 12:45 PMPosted by S23
Of course, that's been stated 100's of times on this forum. Forced losses and forced wins are a thing, how do you not know that? Show me a log of your games and I'll show you loss streaks once you reach a new tier.


make it 3! i want 3 bottles of snake oil!!
02/12/2018 03:16 PMPosted by S23
One of the largest clues is the impending loss streak that usually (but not always) happens when you reach a season high, career high, or new tier. It's not just losing the matches. It's how they are lost. They were impossible to win.


If extra streaks were being added due to rigging, the autocorrelation function would have statistically significantly high values for longer streaks. The shape of the autocorrelation function would be different.

02/12/2018 03:16 PMPosted by S23
If you ask me, I have a suspicion that the loss streaks are triggered somehow due to long playing sessions and playing flex. Changing heros during the match seems to do something very bad, especially if you make multiple switches.


This causes you trouble because it is dubious play. Playing tired will make you more likely to lose. Excessive switching throws away ult charge and puts you on heroes that you are not as good at.

02/12/2018 05:03 PMPosted by S23
Ok sure. I'll provide the proof next time you reach a new career high while running solo queue. The loss streak you receive will be your proof.

Now prove to me that won't happen.


Real data explicitly contradicts this.
https://docs.google.com/spreadsheets/d/19qAqcIcKfIulz_YN960HgV65c2GfphYdndc6lZli2CA/edit?usp=sharing

This data is all solo queue. I have 7 blocks of career highs. Only one or two of these are followed by significant loss streaks.
https://yourlogicalfallacyis.com/false-cause
The problem is not that the matchmaker gives you hard and easy matches. Even if it didn't go out of its way.. that would happen anyway.

The problem is that you can win a really hard match at 2500, then lose a bunch of even matches and wind up at 2200.. then lose an easy match and drop down to 2150 in the blink of an eye.

It's like the game totally forgot that just an hour or two ago you were hundreds of points higher, and the only relevant data it has to go on is the last match you played.

You will win and lose matches, and some of them will be really unbalanced. But the game reacts like a child playing with a thermostat. It goes hard one way or the other and never gives any thought to your past performances.

Elo ratings were originally designed to work this way because not everyone had a laptop and a smart phone, and besides, it takes a lot less time to play a chess game than an OW match. Blizzard could be doing more with the information it has about all of the game's players.
02/12/2018 05:49 PMPosted by ShiroKabocha
The problem is not that the matchmaker gives you hard and easy matches. Even if it didn't go out of its way.. that would happen anyway.

The problem is that you can win a really hard match at 2500, then lose a bunch of even matches and wind up at 2200.. then lose an easy match and drop down to 2150 in the blink of an eye.

It's like the game totally forgot that just an hour or two ago you were hundreds of points higher, and the only relevant data it has to go on is the last match you played.

You will win and lose matches, and some of them will be really unbalanced. But the game reacts like a child playing with a thermostat. It goes hard one way or the other and never gives any thought to your past performances.

Elo ratings were originally designed to work this way because not everyone had a laptop and a smart phone, and besides, it takes a lot less time to play a chess game than an OW match. Blizzard could be doing more with the information it has about all of the game's players.


No one in Chess would be silly enough to adjust individual ELO after a team game of Bughouse, but somehow people here are under the illusion Blizzard can. Lol.

Join the Conversation

Return to Forum