In that video I’ve linked you to multiple times, Jeff explained that there are difficulties for new accounts. If they have no information for a role you have never played, they will use the MMR from a role you have played to seed you initially at a rank closer to the data they do have for you. It’s a very clear and statistically accurate thing they’re doing. He also has explained clearly that they start it with a very high amount of uncertainty. When an MMR has high uncertainty, it can move faster than they allow MMR to move for situations where there is plenty of recent relevant information.
Seeding 4000SR equivalent MMR from Damage role to Tank role makes sense that it would place you in 3500-3900SR. There are only 5 placement games now, so they don’t have a ton of data to change based upon. Just because they use the information they have available to take a good guess at what an unknown SR / MMR should be doesn’t mean that it’s evidence that there aren’t separate values for each role stored. I mean, we obviously have separate SR, why would they keep only 1 MMR?
You keep refusing to look at the information that plainly explains the anomalies you claim are evidence of misinformation, but the developers have been honest and open about things if you only go and look.
Edit: See later response where this was a misinterpretation of what Jeff said.
I don’t think that Blizzard is perfect, they’ve made many decisions I disagree with. However in this situation, you are assuming nefarious purposes when there are none.
If all 3 roles are placed at the same time, this will probably be true, but the scenario Lelouch is talking about where 1 role has a very established SR, the other roles will be impacted by that information in the case where there is no data on that account for the other 2 roles. This is consistent with Jeff’s statements in the Seagull interview.
Jeff did say that they’ll use as much information as they can when they’re having to guess because there’s zero history. So in scenarios that there is completely zero history for Comp, then the first placement game for each role will effectively solidify some MMR / SR number for that role. Edit: This is based on Season 18 specific information at introduction of Role Queue.
I think it’s consistent with my interpretation that if someone went and played 200 games in Damage, getting themselves to 4000SR, that this information would be used if they then played the first ever Support placement match on this account. I think they would take the MMR from Damage and use it as an input (not an MMR with high certainty) for the very first Support match, but that would be it. It wouldn’t be causing them to be tied together, it would just be an initial hint about the MMR for that first placement match in support.
Unfortunately some of this is speculation and interpretation on my part, but I feel that it is an informed guess based on the information available.
Doesnt matter what your interpretation is. Its not consistent with reality and facts. Go and look at samito new account, he climbed to GM and then on same account he started in exactly same place as his DPS, 2350± on other roles.
I have watched SVB, Samito, Yeattle, Dafran and many other,s its the same story again and on their streams its clearly visibile that roles do have completely separated SR/MMR.
Ok, I had to waste my free time to prove something which is just super obvious to anyone who smurf, but you gyus keep arguing about it. Where is Lelouch with evidence? Well Here I am with evidence:
THis is stream of Samito doing unranked to GM on new account. This is third part of his tank journey on tank role where he is at mid Masters:
And This is done on same account after, when he goes DPS role unranked to gm playing it for the first time and guess what, he is starrting again IN GOLD! Because thats how it is, every role is separated.
There were some cases of players placing too high after being OTP on one role, but all of them are pre role Q cases.
That’s fine, I don’t have anything but what Jeff said, but it’s completely possible that Jeff was only talking about Season 18 when Role Queue was just introduced.
Technically Samito did the DPS first I think, the tank is second, but yes, shows clearly gold for the first placement as tank after he was already 4000SR as Damage.
“So this is a weird answer, it’s yes and no. In a lot of ways this is an MMR reset but not in how the players are viewing an MMR reset. I think when players hear MMR reset they think everybody goes to zero [MMR] and we just start over as if it’s the first time you’ve ever played the game. We have been… we have a lot of information about you as a player. We have your previous … all role MMR that we know about, and literally for months and for patches we’ve been tracking role MMR. So we have a lot of data to seed people with. Now of course there are complications, there’s people who haven’t played the game in two years, so everybody’s in sort of a different state, but we try as best as possible to take the knowledge we have and seed you a little bit. The most disastrous moments for any competitive game is an MMR reset. It is absolutely terrible…”
I appear to have been misinterpreting his statement as saying that they seed MMR a little bit using any information they can in all situations. So, Samito placed diamond in both roles, within about 100 SR of each other, which does support your statement that there is no seeding.
That entire conversation was strictly about initial positioning on the introduction of role queue for already placed players. “this” in context is clearly the introduction of Role Queue.
It’s been a long time since role queue was introduced so I don’t have any particular links handy, but I’m pretty sure it was also expressed that after the introduction with its weird guessing-game algorithm, it would go forward as 3 completely independent MMRs.
Yes, I agree. I was providing detail for what my misconceptions were based upon. I interpreted it as them always seeding MMR with any information they had when it was unknown, but evidence has made it clear that this was incorrect.
I was never arguing that MMR was shared across roles. I don’t think there’s been any ambiguity about the fact that there are now 4 main competitive MMR values for every account: Open Queue, RQ Tank, RQ Damage, and RQ Support. He also made it clear that every game mode has their own MMR, even those that don’t have SR. So your QP MMR is different from your QP Classic MMR etc etc etc.
This is of course yet another argument in favor of keeping MMR… For all the modes that don’t have an SR.
It’s SO much simpler than that. They put the best six players on one team, the worst six on the other, and poof! They get exactly what they want – complete and utter stomp EVERY SINGLE GAME.
Why not have Top 500 play low bronze people every game if that’s their intention?
OK, maybe not that obvious, why not GM and high bronze?
OK, OK… Why not Masters and Silver?
Diamond and Gold?
High Plat and low Plat?
Because that is an imaginary world and nothing to do with reality.
If your imagination had any basis in reality, then clearly matches on the borders would be obvious. If the average SR for a match was 2000, and they put the best 6 on one team, then one team would have all gold people and the other team would have all silver. Show me that screenshot, and we can start the genuine outrage.
But since that’s not how it happens, and human emotions, statistical mathematics, and confirmation bias demonstrate why your claim is false, then there’s really nothing else to discuss.
I’ll expand slightly just for the people who are reasonable:
Human emotions impact the outcome of games significantly. People who let the circumstances derail them and distract them from the work of the match are the core reason for “stomps”. When people feel like things are going well, they don’t let themselves get distracted by the little things their teammates are doing, or the way that enemy keeps beating them in duels. When they feel like things are going badly, they get pissed off at their teammates for letting them lose those duels, or for picking the wrong hero or for whatever excuse they pick (just read the forums).
Winning is just as much about keeping things together emotionally after losing a team fight as it is mechanical skill. Even OWL players are susceptible to the roll. So of course 6 random people stuck together may or may not have what it takes to play like a team better than the other team.
Not everyone deserves their current MMR / SR for the same reasons. For some people, their emotional volatility is the main thing that keeps them from climbing. Getting several of those people into a match on the same team is a recipe for disaster. For some people, core mechanics are why they can’t climb. For still others, awareness and game sense is the main reason. If aim was the only thing that mattered in this game, the matches would be more consistent. Since it’s not, stop trying to blame the matchmaker for the chaos.
On top of all of the natural variance of how well 6 random people gel with each other, there’s still the aspect of players whose SR is not yet accurate.
People need more emotional fortitude to climb. Just getting yourself so that you do not tilt at losses is a major step to getting yourself moving in the right direction.
I’ve said far more than the post I was responding to deserves. Sorry for my voluminous response.
Thanks for gathering this Zax. Developer statements left this point unclear, but the gathered data is not. I’m busy at the moment, but I’ll update the main post when I have some more time.
To be clear, my current understanding is that if an account played open queue before role queue launched, they start role queue with a measured MMR in all roles.
If an account did not play open queue before role queue launched, then each role starts at approximately 2350, regardless of what has been going on in the other roles.
Yeah, its honestly weird sometimes, Role Q beta worked like kind of soft reset for some accounts. I noticed that on older account i didnt play long time. When I was doing placement after role q beta was introduced, I had huge swings and after placement bonus. But my initial postition wasnt 2350. It looked like how new accounts work from 2350 but i had higher starting position.
But yeah many people were overplaced a lot because system was using their previous shared MMR as starting point for all roles.
(I’m Diamond by the way) I always get tilted that I get matched against low to mid Plats and I stomp them or I get matched with Masters and High Diamonds and I get stomped or Carried. It’s just not fun. Don’t know why the match making doesn’t put me in Low to Mid Diamond
I just meant to say that different things for different people are the most important things to work on in order to climb.
There’s the MMR the game has estimated for you, and then there’s the basically impossible to know for certain “true MMR” for people. I definitely understand that the game is not perfect at estimating that MMR, but there’s a clear spectrum of skills across the player base, and it is not a stretch to say that it is a real and worthwhile process to compare the relative skills of players. In order to find matches that are as fair as possible, it is very valuable to have a numerical estimation of someone’s skills relative to everyone else.
Of course higher or lower MMR doesn’t make someone worth more or less AS A HUMAN, but wow, why would anyone try to make that argument? It’s just about trying to find games where you’re playing against people of roughly the same skill as you.
And back to what I was saying, there are different strengths or weaknesses that people have. Sometimes people can climb by reducing weaknesses, and sometimes people can climb by strengthening things they’re already strong at. There’s no such thing as perfection, only improvement and achieving the goals you set out for yourself.
Why not have Top 500 play low bronze people every game if that’s their intention?
Fine. Do that. It’s no different than what they’re doing now anyway. I’m a <500 Bronze player and they’re putting me against high silvers and low golds. Not to mention all the smurfs who aren’t REALLY playing at bronze level or silver level. Just based on math alone they are 3X and even 4X better than me, so why not top 500 too? I’m getting creamed either way.
Thank you kindly for compiling all these sources of information into such a coherent guide. And more, for continuing to update!
When I eventually found your S24 thread, it helped change my mind, having hypothesised (in a big Reddit post) about apparent matchmaker bias coming from MMR being off-set. I don’t think I can tag members here (?) so I’ll quote my new thread:
I borrowed the markup code for collapsible text sections from you too, which has helped make the lengthy breakdown of my statistical match analysis more approachable. Hopefully.
(1) I think I’ve been able to quantify evidence of apparent matchmaker bias there, against one role in particular. (Or at least show some very unlikely series of events that fits within my long term subjective experience of bias.)
(2) Also, I saw a weird blip, half way through, at the start of season 27, where the apparent bias went away: my 45% win rate flipped to 100% for 10 straight matches (across many different days). Anyone else see anything like that? It seemed it might have followed a period of several matches where my (medal) stats were particularly high (but mostly not winning). Like, perhaps it was a mechanism I triggered…? Or randomly allocated (or ~3% random series of events).
That also meant I got to verify that win streak bonuses are a thing (something I never expected to see for myself). I got up to 33 SR for the last 2 wins and 28 for a couple before that (which wasn’t quite unprecedented elsewhere).
(3) Call me ridiculous and desperate, but I was also testing, right from the start of the analysis, to see if QP stats/MMR might influence matchmaking bias. And while there’s no way to prove events weren’t circumstantial, throwing 10 (or so) QP games on support appeared to get that role unstuck in comp. With significantly more favourable (subjective) match balance ratings, that fell away, back to being more negative. Perhaps as the (un)biasing effect was diluted by comp matches (or, arguably, as the SR climbed slightly, but didn’t reach my probable skill limit).
(4) I saw some weird behaviour with my QP MMR (that I was estimating, at great length), after I un-threw my games on that role. Like, it was not at all just like climbing the MMR back up. The first game I stomped, with PotG, then the next was unwinnable and all over the place for a whole tranche of games after that. See graph [Fig.19] in the write-up.
So I wonder if that was a reflection of the MMR uncertainty value (you talk about above) suddenly going through the roof…? Or if it’s maybe a separate (secret) system for dealing with account sharing…?
You state that MMR changes more quickly when MMR uncertainty is higher. Which makes sense on a new account with accelerated SR changes. Maybe in QP, after a long time away. But would we really expect to see this in comp MMR? Assuming that SR really is just a projection of MMR, that would make for big gains/losses after, say, a year of not playing. Does that happen?
(5) I saw a whole bunch of other minor findings and trivialities. Like flex/ticket queues weren’t worst for match outcomes (maybe slightly positive, but on limited data points).
⠀
⠀
Also, can I ask, is there any solid information about how the matchmaker tries to make pseudo-comps, by not having two players who mostly play the same character, on the same team (and now role)? It seems to work out an unlikely number of times, that I get a Rein player team mate, since I switched to maining Zarya (from main tank), for one example.
I’m sure Your Overwatch has talked about devs alluding to this, in the past… Maybe. And to counter-balancing uneven group sizes with (solo queue) players of a higher MMR (or other qualities). Though I couldn’t re-find that video, when I went looking for it last year.
Anyway, thanks again for the great post. And I understand if you want to stick to official sources and avoid speculating with me.
