All of the above aside…I maintain the complete absence of change to remove the completely unnecessary and not beneficial mmr…or even address it, or even respond to a forum post like this… basically says everything that needs to be said. Blizzard has 0 plans to improve in this area.
1 Like
I agree with your sentiment, but of course I feel there is much more to be said. I hope that every Overwatch player who cares about fair competition will continue to discuss this topic and demand change ad nauseam. The present and future of online gaming is at stake.
1 Like
i still don’t get why they hide information ( something prob for money from investors)
about
- present active players online
- popularity between tiers ( 1% gm 5% masters 10% diamond etc)
- how much players left competitive matches and stick to arcade,qp or workshop maps
- base statistics of KDA per team
1 Like
But you didn’t quote the entire thing?
You twisted what you did quote to fit what you wanted to portray in this thread, while leaving out other key points.
On this point, even though it wasn’t directed at me, I do agree, the problem is, you have 0 evidence outside of player/personal experience to support the claims you’ve made, unless you can provide details directly from Blizzard, this thread is worthless, and should be/is looked at as “he said she said” childish argument.
Not every player is demanding change to a system that is working as intended, only those “hard stuck” “rigged propaganda” players want it, other players are having minimal to no issue with the current system (myself included), the fact players are still climbing and falling, based on their level of skill, proves the system is working, as intended.
Sure it sucks having to grind, or be at your skill max (like I am) and never see change, but those who truly believe the system is against them, should take a look at their own gameplay, and really pick it apart, I do not believe in forced wins/losses, in every single match I’ve played, if ONE person did “x” it could have changed the entire pace of the match. The major issue being: people DO NOT look at the win condition, and instead blame the matchmaker (yes sometimes it can be out of your control, smurfs, throwers cheaters, etc), but this isn’t on the system, it’s on the players.
If they release how the system works, we (as a community) would face players farming necessary stats to climb, and end up exactly where we are now.
3 Likes
Hey, I made a tool that simulates semi-randomized matchmaking that effects 50/50 outcomes using a knapsack or “best-fit” style algo.
docs.google.com/spreadsheets/d/1XtVVYyEx0__EwdrGJ0WGbZr_rUI3Up94Wacu-Fi2kRc
Toy around with it here, and let me know what I can add to make it more detailed. 
The knapsack works by taking turns picking players for each team, with each counterpick being based on the average MMR of the opposite team.
So to imagine it, let’s say you are 2000 SR and 1st pick, it will then try to find a player as close to 2000 SR as possible as 2nd pick for the other team, which ends up being a player at 1800.
For 3rd pick, it will try to find a player as close to 1800 as possible, in order to bring the averages of the teams closer. In this case it finds another player at 1800, bringing the team average to 1900. 4th pick will then try to find a player as close to 1900 for the opposite team, and so forth.
average(2000, 1800) = (2000 + 1800) / 2 = 1900
The graph is a great visualization of this! (It’s much harder to explain in fewer words)
Currently, only relative MMR is used as the core of the matchmaker. SR does not matter for calculation purposes, and it is locked to player picks within +/- 500 SR of the Target SR.
Relative MMR is represented as values upwards or downwards from zero. You can consider it your base level, for example using your own SR.
Dean eM3res Died Haechi AmenPhoenix Wanted
Also, the reason the resulting match SR may be different from your input SR would be the equivalent of you being placed into a match where you outclass or underclass the average. The match overall will still have fair picks.
Hmm, I can also just force the first player pick to be the target SR (or your SR). Right now it’s simulating picking all players from a virtual player pool. 
Let me know if you’d like this!
3 Likes
well we are not asking to let us know how matchmaking works. but the usefull information about population or scoreboard
people are toxic anyway
1 Like
That’s not how quotations work. Anyway, you are moving the goal post. You accused me of failing to quote an entire sentence or paragraph. Now you’re accusing me of ‘not quoting an entire forum post?’
You have done something fantastic! This model you’ve created is a really cool representation of how match making works roughly, based on what we know about it. Very well designed and clear, easy to use, and it demonstrates the principals of mathematical statistics and handicapping that are involved in this case.
Given that we do not have full knowledge of Activision/Blizzard’s design for the matchmaker, I have some questions and maybe you have considered some of these things. But I do think we can say that your model is analogous to Activision/Blizzard’s patented Matchmaker, and perhaps we can strengthen the analogy further.
How did you decide on the scale of values which represent MMR? I see they are all fractional numbers with values that don’t exceed 0.2. Do they add up to something? And I see that they can be positive or negative. Do you have reason to believe the Overwatch application software processes the numbers in this way, can you explain?
Do you think that all of the performance measurements are truly simplified into a single number like this? I somewhat doubt it, based on what I have read in Activision/Blizzard’s patents of invention. They have laid the groundwork for analyzing player performance with high fidelity and granularity, building sophisticated profiles not only for individual players but for groups.
And the things they are gauging, such as players’ ability to play individual characters and classes, their ability to synergize and exercise tactics…I think that the calculations that go on at the moment of matchmaking may be a lot more complicated than what we see in your model. Is it possible to represent those factors in this format?
On the other hand, a more sophisticated model may not be necessary. And it’s possible that the Battlenet/Overwatch applications do work with a single MMR number in the back end. I just think it’s unlikely, or that it is more of a reference number for developers and not used in computation as much as its constituent statistics, of which there may be many.
Assuming that the presence of skill in one area is likely correlated with skill in other areas (I believe it is, as does Activision/Blizzard), then I think this is a very good representation for players to look at, to understand what Matchmaking is and how it works.
I might want to feature your work in the original post of my thread. Do I have permission to reference your model in its current state?
2 Likes
this is really a good representation.
since amount of people playing in each role also doesn’t correspond. Easy numbers. If there are 2 people cueing for tanks 5 for DPS, and 3 for Support, would the averages work out and the numbers be 'similar"
2 Likes
The win condition differs though from hero to hero. For example, Sombra’s win condition and how they register her SR rating is garbage. I put money on it that it plays heavy on damage done and how many hacks you do, but it can’t register the amount of times I’ve sprayed a tiny bit at another hero for them to be distracted and get shot by another dps. Is the system tracking, as a Mei how many times a hero is trapped by my ice wall and then gets killed by my team or that my ice block helped shield members of my team from shatter, no it isn’t because it’s not reflected. How about as an Ana tracking how many sleeps saved a life, not just how many sleeps did you make or what sleeps stopped Ultimates or playing any hero with a movement ability to bait out an ability. Are they tracking abilities dodged since some heroes win condition is that? OW has a half-arsed algo written by someone who thinks it’s perfect.
The best quote I can show that the MM is bs is from a young player asking “what tank should I play” in plat with a brand spanking new account and then playing at a silver/bronze level. It’s further shown how broken and idiotic it is by lazily putting every one in gold/plat on entry. OW is only about the money not the quality. That’s why so many lifelong devs are hopping off Blizzard’s train because Activision wrecked them.
3 Likes
This is what I find so galling about the double standard of SR/MMR, and “performance-based skill rating adjustment.” It is impossible for the application to appreciate such nuances in strategy, the way that only a human being can (at time of writing). So PBSR adjustment will never be appropriate. It can never compensate for algorithmic handicapping, which has such a profound effect on match results.
2 Likes
Ah, yes! The simulation is based on an exponential curve between -0.250 and +0.250. This creates a “bowl” shape of possible values which are more frequent near zero and exponentially less frequent towards the extremes. This produces a bell curve of “best” candidates.
The reason I decided to scale it at +/- 0.250 is simply because I believed it to be a fair value, but it can be any value and then rescaled to fit relative SR within +/- 500 SR.
Additionally, the value of +/- 0.250 prevents the odds from ever going beyond 25/75, or 1 in 3, which is extremely unlikely!
The numbers don’t “add up” per se, but they are based on a relative moving average of raw MMR skill based on the opposite team’s relative skill. I added more about this in my post above!
The reason I used a knapsack technique is it’s what came to mind when thinking about how to best-fit as many players of relative skill into a potato bag when those players all have strange and unique sizes/shapes. It’s the first algorithm that comes to mind whenever a predetermined outcome isn’t known, but a good solution must exist.
I then coupled that idea with that of Team Captains picking teammates for their team, schoolground/playground style. Each Captain takes turns picking the best possible teammates for their team, until teams are full or they run out of teammates.
In my particular example, I actually do this with the opposite intention. Instead of picking the best possible teammate (or most skilled), instead I look for a teammate that will bring the average skill (MMR) of each team as close as possible. (I also give a brief example of this in my other post! 
)
My example only takes one number into account, which is only the MMR value.
I’m more than sure Overwatch uses dozens of factors in its calculation, however I wouldn’t be surprised if those multitudes of factors are resolved into a “number” or number-object, such as the established MMR value.
I think MMR is Overwatch’s best solution to determine raw player skill and enumerate a value which can then get plugged into an algorithm, such as the one I made.
If there are additional factors to consider, I might suggest that Blizzard probably calculates a different value of MMR out of several possible matchmaking scenarios and deliberately “waits” until a “best-fit” occurs (some win % within some threshold near 50%) or until a predetermined amount of time elapses and several possible matchups exist. The latter here being when the system matches you into several instances of possible games that are considered fair but outside of the “best-fit” threshold.
These are just my thoughts though! 
This basically means they own a monopoly on your data, and use that data to modify your player experience. Also, that is a mountain of data I don’t have access to! 
It’s not difficult to do, but it requires an enormous dataset of player data which can then be cross-compared against other players in the dataset, to then “granulize” an MMR value.
That generated MMR value is then plugged into the same model/tool I made.
As far as I know, there aren’t multiple MMR values. But there are multitudes of hidden player data that may influence an MMR value. As for what those hidden player values are, I have no idea.
It could be as generic as hours played on a character, or as specific as the “value” of sleeps you drop as Ana or number of players hit by Sigma’s grav slam, or as abstract as players you’ve recently played with or what skins you have selected.
I also think it works similarly to how property is valuated in real estate. Many factors are measured, summarized, compiled, and evaluated based on a multitude of factors, like regulation compliance, quality, quantity, age, risk, damages and violations. All of these things are deliberated upon by a firm, and then a single house price is generated based on the qualities of the property.
This is why MMR is such an important number. It’s basically your virtual appraisal/worth inside the game.
That’s totally fine! I hope it’s helpful!
The current model doesn’t use an actual dataset, and it doesn’t calculate for roles just yet. It works similar to the original Open-Queue system.
Also factoring in how many players are waiting for a game doesn’t factor into this model, as it asks a fundamentally different question.
If I were to implement a party selection into the sim, such as 2 players queuing at once for open queue and simulating subsequent counter-picks, it would only demonstrate the skill levels of your potential teammates. I do think there is genuine value in seeing how groups can affect the matchmaker, but that would best be demonstrated in a different model/tool.
Going further, adding ranks into the sim is another step that’s pretty complex, just thinking about it, though it’s not too bad to do either, as you’re basically running 3 knapsacks inside one bigger knapsack, then trying to create a best-fit based on a multitude of possible rank-pairs. I’m not sure what visualization would satisfy the demonstration!
Just for the sake of this tool, let’s assume this is what you’d see when you solo-queue in open-queue, and this is the resulting matchup!
3 Likes
Thanks for understanding what I meant. As I read the quote back… that sentence is terrible.
2 Likes
Cool, this reminds me of the graphs that I was seeing during my research, in Microsoft’s patent for a ‘Bayesian skill scoring framework.’
I figured, thanks for confirming!
I wonder what are the implications of the designers choice of scale (just thinking out loud).
That’s interesting. Do you suppose Activision/Blizzard’s Matchmaker works the same way?
Thank you for sharing your thoughts, this is exactly what I wanted! You seem to really know what you are talking about. I am not an expert when it comes to mathematical statistics and/or computer programming. May I ask your professional/educational background?
I am a technical writer myself. Technical writers try to understand systems and then write about them for a general audience, as I am doing here. My work constantly puts me in the position of making assumptions (without which, research and review are paralyzed). But I always look for subject-matter experts to confirm or refute my assumptions.
In the course of this thread, many posters have posited assumptions about the nature of MMR as a number or ‘number-object.’ Your perspective on this carries the most weight, in my opinion, so I’ll keep it in mind. We are still in the territory of educated guesses.
For sure. I don’t think we need access to the data. We already know its major parameters from Activision/Blizzard’s patents and description of invention for the Matchmaker. And we know those parameters include expected things like damage/healing output, kill count, etc., but also highly unexpected and problematic things such as gender, income and location.
The model you provided is very good, I am just fantasizing about ways to make it better. But as I think about how it might be more elaborate, I also think that it is better to keep simple, so as not to be accused of making unwarranted assumptions about Activision/Blizzard’s Matchmaker.
From my reading, primary factors include things like role/class preference, damage/healing output, synergy with other role types…the list of potential parameters goes on for miles. It includes things like whether or not the player uses VOIP and how much, how much they communicate via chat/emotes. It could include things that are hard to measure, such as efficiency of movement. And it certainly includes all of the statistics that we are awarded medals on.
Fascinating! I think you have a very good take on this, and you are clearly a student or master of the underlying subjects. Thanks for chiming in and contributing your work. I’ll reference your model and provide a link in the original post, when I find the right place for it.
2 Likes
Disgusting. And I’ve known this for a while, the game would be so good without this. It is absolutely disheartening and fuels me to quit when the game punishes for playing good by lowering their chances to win. But the thing is. I really doubt blizz gives a single ****.
2 Likes
Sorry it has taken me a month to respond to this, because it’s a good question and it helps me see where I have been losing readers/viewers like you. Your hypothetical scenario is based on an assumption that I am not making, and which is not true; the assumption being that the Matchmaker tries to break the winning/losing streaks of individual players and groups.
The Matchmaker certainly does not do this, and neither is it the goal of handicapping. The very purpose of handicapping is to create a match whose outcome is uncertain, with 50% odds of victory for either team. Whether the outcome of the match is part of a streak for any participating player is purely incidental. However, the Matchmaker creates those 50% odds by gauging the performance of players from match to match, and appointing teams accordingly.
So to answer your question, MMR/handicapping applies to every player in every match. Handicapping can be positive or negative, and whether it is positive or negative for you depends on how skilled you are, relative to others at your SR level. I hope that helps you see my position. Your questions imply a sense of unfairness. I share your sentiment.
To your point about career stagnation: I found that my Overwatch career stagnated the same way. More than stagnating, my SR ranged from Platinum to Bronze several times across my career, seemingly at random. And all the while I have witnessed lesser-experienced and lesser-skilled players in practically all of my teams, doing foolish things that I stopped doing in my first month of playing and rarely matching my stats for medals.
Every time somebody comes along and agrees with my argument, it is like a breath of fresh air. Cheers pal
4 Likes
Given the differences in the game between player pool, ping, hero selections, map, etc. -
What is the fundamental difference in the handicapping (that is both positive and negative for players) and the simple reality that 12 different people in any match - under any system - will always have relative skill and performance differences between them as a byproduct of each player being a different entity?
That is to say - conceptually, your idea of handicapping, at least on that particular point, seems to be a natural byproduct of team-based, online games, rather than a nefarious, systematic manipulation of a player base. There will always be a best and worst performer/player within a match, and that is in no way an indicator that the system is rigged.
The best aspects of your argument to me have always revolved around the lack of forthcoming information from Blizz and the fact that the matchmaker is A) hiding the figure that is actually used to put you in matches, B) possibly using metrics that many wouldn’t feel are valuable to a competitive environment, and C) does a less-than-ideal job of actually accounting for player skill when assigning MMR. The part where you lose me is when you seem to suggest a 50% chance of winning is somehow less fair than any other system that could produce anything other than 50-50, as well. To clarify - I realize you’re arguing that the 50% chance is purely artificial, but I don’t see how a system of pure SR (plus group size) and no other data produces a less volatile or more enjoyable experience for the player base.
That said, keep on keepin’ on, Cuthbert. I don’t think we will ever see eye-to-eye on it completely (y’know… since we’re still not on the same page after years of this being a topic), but I admire your passion and dedication. I hope you have a wonderful day.
Edit: Additionally, I’d still be a willing guinea pig to see how your suggested system plays in practice, FWIW. So, sign me up for the experiment if it ever becomes a thing.
2 Likes
I really like this doc and feel like it might be really close to how Blizzard actually does matchmaking simply because of the scale that must deal with and the simplicity of the algorithm. My question is, how did you generate the individual MMRs? When I select a target MMR I’ve noticed it is possible to get teams that are quite far from the target, sometimes >200SR away. Are you assuming that the MatchMaker™ calculates the average of the next available 12 players as the target MMR? My initial attempt assumed that the target was the SR of one of the players but after many simulations it was clear that wasn’t the case, so I wasn’t sure how that part of the simulation was constructed.
Another point, your simulation supports the idea that when you are going up in SR it is almost guaranteed that your team will be filled with players of lower SR/MMR than you as it would be hard to create a scenario where you are always lower than the target SR/MMR for a match and therefor always the player getting carried. Have you considered doing an analysis over many runs (the sequence being important) of a player who is A) going up 50SR a match vs B) going down 30-40SR per match. It’s not so much that it’s up / down but near the edges of the skill range (under 1000SR for example) the skill slope of players gets much steeper simply because there are so few of them in that range.
3 Likes
any news if this is getting addressed for OW2?
2 Likes
Hey Sheevah, nice to see you again. I’ve done a lot more meditation and research on this subject since we spoke last. I’m afraid time has only entrenched me further in my position, but I appreciate your questions and your open mindedness as always. Have you seen my new video based on this thread? It’s posted at the top.
Getting to your question, you are right that there will always be differences in player skill between any 12 players. The purpose of a competitive match, in a team-based game like Overwatch, is to prove differences of skill by the victory or defeat of either team. The winning team presumably has the best players, and Overwatch players rightly assume that their record of win/loss should determine their careers.
But the Matchmaker’s algorithmic handicapping of each match turns the entire premise upside-down, and defies players’ assumptions of the system. Yes, differences in player skill are natural. But the applications’ state of knowledge about those differences (MMR) is not natural, it is highly artificial. And the purpose for such statistical analysis (ensuring 50% odds for each competing team) is not one that most players have considered.
It may interest you to know that Activision Publishing Incorporated, the game holding company of Overwatch, has entitled themselves to use gender, income, and residential location as segregating factors, for the purpose of matchmaking. These are ‘details of invention’ in their 2015 Matchmaker patent.
I am not actually saying this. I think that MMR does a virtually perfect job of determining players’ skill, relative to other players at their SR level. But I think the purpose of determining relative skill is, as you say, nefarious. It’s being done to effect handicapping without players’ knowledge or consent. I have argued and proven in my polls that the vast majority of players do not consent to being analyzed and handicapped this way.
Since they have made no comment on thousands’ of players complaint against the system, we can only assume Activision/Blizzard plans to carry their Matchmaker invention forward unchanged.
4 Likes
Because the MMR is specifically designed to help the player. It’s a slower moving number that tracks your skill in the background. It STOPS you from falling too far if you have a bad day, and if the game recognises your MMR is 3500 (been at this number for a long time, lets say a year) but your SR is now 3200, it helps you get back to where you were… It also attempts to prevent people from de-ranking to smurf. They get pushed back up the ladder quicker due to their hidden MMR.
Everyone here is complaining about losing streaks. Imagine if you rolled the dice and lost 10 games in a row (just based probability, it happens to everyone). You’d drop ~350SR and the only way back would be to win 10 in a row. MMR provides some elasticity to your rank so everyone isn’t yoyoing.
SC2 has used MMR since it’s release, no one complained about it that I saw, I got myself to low masters. The difference? It’s you and one other person and if you lose, it’s your fault. I fixed my stupid mistakes and ‘tried hard’ and went from gold to high diamond in about a month.
The fact is this is a team game and people aren’t symbiotic. MMR isn’t the problem. It’s because it’s a team game and your win/loss depends on 11 other people. It isn’t rigged. If every match is mixed with 11 other random people, there is always going to be someone performing well/badly. There will probably be someone AFK texting in spawn or doing damage and not healing.
It would be impossible to place these people on your team intentionally (because people play at a slightly different level/emotion/attentiveness every game). And even if that WAS possible… How would these same people be on the same team as all the other people complaining about rigging at the same time? People throwing and cheating are in the vast minority. How do we share them out?
You fix ONE mistake, just ONE, for example, not approaching a McCree with a stun as Tracer, and you will instantly get better at the game and convert more into wins. Stay alive as Rein? You convert into wins, don’t feed as Hog and give your enemy Hanzo endless ults, convert into wins.
All this to say, it isn’t rigged there are just way too many variables to actually create a system that always produces a ‘fair match’, but that’s based on the human beings input, not the numbers the system matches up. Matching peoples’ MMR gives the ‘closest’ match it can, the rest is up to you. If someone of a high skill level can buy a new account and be GM within 6 hours, the system isn’t broken.
P.S. this message isn’t a rant at you specifically, just seemed like the most appropriate message to respond to with my thoughts.
2 Likes