How does matchmaking in quickplay work?

Can someone explain whats happening here?

Through 50 games, the matches are relatively close, i’m able to call objectives, people listen and respond, and in the end I have approximately 80% win rate.

Through the next 50 games, I lose 80% of the matches, and most games my team has literally no chance. the players on my team are that bad. they dont seem to understand the basics of their own abilities, yet alone how to use them together with the team.

How is the matchmaking working, and does it get better?

1 Like

It works like this:

One time you have good players on your team until you win too many matches. Then Blizzard on purpose puts bad players on your team so you lose 5-10 matches in a row. Then repeat.

9 Likes

there has to be more to it. in the first 50 games, it wasnt like there were potato heads on the other team. both teams seemed to have at least adequate understanding of the basics.

now… even in wins, it feels like dumb luck. there are people wandering around the map walking into towers, dying to creep waves. standing in the middle of the other team not doing anything. and no one talks at all. they say absolutely nothing, unless you ask them what they are doing… then they report your for being toxic.

2 Likes

Matching works by grouping select clusters of mmr within a certain range of each other, and having an approximate average of all players on both sides being similar. ie a side might match 1000,1050,1100,1150, & 1200 mmr against a team of 5 1100’s.

However, those are approximate values as the ‘exact’ number isn’t known (esp for QM,) and there can be a difference between a number trending upward compared to one trending downward. Lotta conspiracy rants see that as “forcing” loses, but since the system uses MMR averaging, there isn’t a difference between a 1200 and a 1200.

While all sorts of people rally to tin-foil theories that suit claims the game “forces” them to lose, the general issue is that said players are clueless, unobservant, and don’t know how to have a bigger impact on what is actually happening, and since they don’t know, it’s easier for them to assume the universe is out to get them personally, so they rally around theories that claim something is out to get them.

Functionally, the matching doesn’t know who does what, and the metrics of a “good player” aren’t so binary, so the system just tries to make guesses based on average performances. Heroes Profile, a third-party site that tracks game stats from uploaded replays (called parsing) indicate you are playing on a low-level account.

From what you may have noticed, outside of grouped players, you’ve likely also encountered other accounts that are lower leveled AND are also played with heroes that have lower levels.

With low level accounts, the game doesn’t know where to seed a player (what mmr they “really” play) and QM gives an extra handicap for the first few levels a player plays on a hero since they’ll usually play worse with lower leveled heroes than they do higher ones.

So functionally, between lower account and low level heroes, there’s a higher range of uncertainty on what the game can anticipate on the MMR value for players, so it’s prone to seemingly dramatic shifts because the game can’t magically know if players are legit new, smurfs, boosted, etc etc esp on lower leveled accounts.

What tends to commonly be the case is a sort of ‘rubberbanding’ on matching certainty. In the 1000 -1200 mmr example, a player with a few games will be on the lower match end of the mmr average and have “better” players in their bloc. As they win games, they are in the same bloc, but now they’re at the higher end, so the average mmr has ‘worse’ players for the same average. ie, when a player was 1000, they may have played with a 1200 in their team, and now that they player is a 1200, they then have a 1000 in their match because they’re still in the same mmr “cluster” with where the matches average the matched mmr.

If the player wins, the confidence the game has in the rating shifts, and they progress upward (in ranked, that’s usually seen with promotions form say bronze to silver, but that’s more decorum than not) If the player loses a couple, then the confidence of the MMR settles and sees the lower value as being closer to the ‘true’ average of the player.

Because there’s a lot of uncertainty at certain ranges of play, it’s not uncommon to see some games flip/flop because there’s no real investment in making a new account and getting a hundred or so levels fairly quickly.

As such, it is harder to discern when players are legit new to the game, rely on friends to take charge, are just goofing around, etc etc.

A number of game metrics try to suggest it takes more than 100 games to actual seed a player, but some of those estimates don’t include as diverse a hero/role system for HotS. And while it should hold that when players have good basics, it shouldn’t matter too much what hero they play, functionally, more of the playerbase is pretty bad at the game, and so long as they have something else to blame for their conduct and losses, they don’t have much incentive to actually improve themselves by knowing, seeing, and doing better.

As some point things do get “better” but it’s not a linear distinction, especially for Quick Match, because players can level up heroes in AI or ARAM modes and essentially undercut any sort of indication that they’d ever actually be ‘better’ at the game regardless of their account level.

Some players have 3000+ levels, and don’t know the distinct objective for each map. So a key part of some of the matching does fall into “luck” especially for players that predominately pick assassin roles. Since there’s too many QM players in non “essential” roles, game experience can be all over the place in having mass assassin/bruiser games where people demand tanks/healers, but never actually play them themselves, and the matching can’t magically produce what doesn’t exist.

So tl;dr, based on what I saw, some of your earlier matches had some benefit of the doubt for newbie queue, and your later series of games may be out of that and you’re in the rut of “well, they’re not “new”” in your matching.

The level on the account should play absolutely NO PART in matchmaking. If that is the case, its absolutely absurd.

What is the benefit of the doubt you are talking about? Its a new account so lets put him in games with people who know what game they are playing?

Ok… thanks for that benefit, i won literally 80% of those games.

My reward? Is it getting put into games with more experienced players?
Nope, apparently my reward is getting put into games with the “well, they’re not “new”” people. People who play like they’ve just eaten a bottle of elmers glue for dinner.

1 Like

No. You have good runs and bad runs.
Plus, you might change yourself, get bold or scared.
Plus, players around might change. Tassadar washes the floor, people observe they can’t win, they shift towards other heroes and now Tassadar is crap.

There was a period I had 85% winrate for weeks with Tassadar and Qhira. With my injured wrist in a cast. Both heroes have been buffed since, but my winrate is down with both heroes to 40-45% - with the same build! I regularly swing on Li Li between long-time 55-60% winrate and 20-25% short time.

This doesn’t make sense to me.
Why would the MM keep short term and long term MMR, and challenge / bump you intentionally if they deviate?
The entire point of known MMR systems is that your rating takes your opponent and the result into consideration. If you win, it gets more difficult. If you lose, it gets easier.

The main issue is that skill has 5-6 major dimensions and it’s completely possible that a player will perform far below expected because they don’t fit together with / as a team. Easiest examples are tanks who get no followup, or players in my matches who simply attack move enemy core and just go until dead, respawn, repeat. And with them feeding, your rank drops, meaning you are bad if you cannot support this playstyle.

Plus the effect of players constantly being misplaced a little. A player might get lucky and keep playing other lucky players with an inflated rank, then suddenly meet someone after some bad luck. Since rating is relative, if you take two inflated players and put them against each other, it’s a balanced match. Copies of Fan can get legit stuck in bronze (when he meets other smurfs during B2GM) and I can get legit stuck in GM, constantly meeting another me. That’s why often a team of 25% winrate folks on a 10 loss streak each smash in SL. Meaning, if you are matched with someone with a full red match history, you are actually likelier to win. 6D chess I tell you.

By the way, this also goes backwards. There was a point where I had a bit of luck and I’ve been placed into QM fairly often with gold and plat players. The whole thing was fantastic. I was healing, they didn’t die, I was tanking, they would follow up, I was assassin, they were setting up, poking, finishing. I had a blast. Then the game decided to match me with different players, and it was the opposite. I unload with Jaina but it’s not enough. I work hard with Ming but there is never an opportunity to finish, to reset. I heal 150k more than the opponent healer but we lose anyway. Maybe I’m the type of player whose winrate goes down as the MMR goes down. I’m gradually approaching bronze one step at a time, down from Gold 1, and it’s getting more and more hopeless. I know, I’m a sinner, I didn’t spend cash.

And thus the seeds of clown fiesta are sown.

(Guess I should re-edit and detailify this, but meh.)

What’s always amusing about people ranting that the game “forces” them to lose is that you are always playing with other people. People are fickle, and a computer cannot figure out if they are going to win or lose. It can average MMR, and that’s about it…

The reality of the situation is that many players tilt or tend to play poorly if behind. Until you get to higher level, most people will look “good” when they are winning, and look “terrible” when they are losing. It is the same player, and they didn’t suddenly get bad.

People are terrible at playing from behind in this game despite it being one of the easiest scaling video games I’ve seen with respect to comebacks.

I bet you could take the same 10 people, scramble the order of who is on each team, have them all play each other for 100+ games, and watching replays you would consider the same player both good and bad depending on the game.

Don’t get too hung up on the game “giving you bad allies”. The game isn’t doing that, people are just playing poorly. When you look at their accounts, they still have ~50% win rate and there is a good chance they even won half of their last 10 games.

On rare occasion you’ll run into 4-5 stacks of smurfs, or get AFKs. Those are not common-place, they are just the games you remember. If you chart your games on a map this is like 2% of your matches (unless you are playing on lower population servers).

If you dislike the inconsistency of QM, just migrate to Storm League when you get a chance. SL isn’t perfect, but it does smooth some of this out.

2 Likes

The functional issue is that initial seeding has to start somewhere and can’t magically be calibrated from the onset. So your assertion, and all the flawed reasoning that went into it, is bunk. Level plays at least a small part somewhere in matching.

I already pointed out that you have matched that have similarly low level accounts, so you’re not expressly being put in games of people that “know what game they’re playing” by the same metrics. However, since experience is subjective, then account level doesn’t matter for most of the game experience, but since you’re skimping out on details anyway, it’s not really worth my time to try to further explain something you’re not interesting in reading otherwise you’d actually read through this and not draw bad conclusions from what you don’t want to process.

Because smurfing is a thing.

Functionally, the goal of a matchmaker is to try to find the rating in a particular frame of time. The complication of that is that skills should change over time, but when it comes to finding the initial estimation, the game can’t, doesn’t, won’t, know the difference between a legit newbie and an experienced player doing a climb again, or if it’s not the same player on the account at all.

So “known mmr” systems have “confidence” in their mmr appraisal and when players have long winning streaks or long losing streaks, it challenges the probability the matching uses so games like LoL, HotS, and others have other tuning constraints on how mmr progresses for their games. Some try to speed up smurf detection, others try to slow the process down, but a number of games have “rubber banding” effects because these aren’t 1v1 games, so the direct influence of Elo changes doesn’t really work for how matches ascertain “probability” in winning one-off games against various rating averages.

The ‘real’ effect of Elo would be played out in more than one game, however, since Dota-like games have long-enough matches as is, people don’t play the same matchups to verify the actual point exchange that would be used in something like Chess. So other elements exist to try to compensate for the differences in players matched, games played, etc etc

It’s less that the system “challenges” the player, but rather the consequence of how matching metrics average numbers together. People think of the rating as “skill” and the game does not rate that at all, it’s just probability of a series of games won compared to other ratings.

The realization of that is better seen in fighting games since their Meta rating are based on matchups and indicate probability seen through games played. Ie Tier A character may be expected to win 4 out of 7 matches against Tier B character; that doesn’t mean that that has to be the case, but that is the expectation. Same deal for MMR, it has an expectation of games won (or lost) in a series, and if the expectations exceed the probability, then the matching for the value is going to change.

Granted, phrasing it that way may not make more sense to you, but there are a number of similarities to Fighting game genre to Dota-like games than other genres.

The bottom line is that since these games aren’t matching 1v1, they have to make some adjustments in how their system predicts “fair” games, and evaluates MMR in the aftermath.

Mentor system.

1 Like

All I know is, I queued as Hammer yesterday and:

1st game against: Zeratul , Hanzo and Gul Dan
2nd game against: Gazlowe ,Nazeebo and Chromie.
3rd game, I came to my senses and stopped playing QM again.

1 Like

the initial seeding should put you right at the top of the bell curve. if you lose you should move to the left. if you win you should move to the right.

there are metrics to give an approximation of game impact that should modulate this. it really should be that simple.

the people I am put on teams with are now simply an entirely different caliber of player than the people I was playing with at the beginning of my heroes of the storm experience. I guess we’ll see how the storm league works…

you are saying the forced 50% is a conspiracy; if the system is anywhere like overwatch it is not. some years back i tracked everyones mmr on my team vs the other team for ~300 games, (not the largest sample i know) during win streaks the difference between #1 and #5 on my team got larger and larger, while the other teams got tighter and tighter. and it appeared to hit a threshold where you could break through and get a sort of level up. if I find the excel i will attach, the data is interesting.

Yes, just had a game where enemy had 2 good bruisers: xul and qhira. We had all melee assassins and an abathur. I realize that it is work to balance quick game start vs fair matchups but that is just no fun.

Game was a very probable loss, especially when the illidan turned out to be one of those whiny feeders who then blames their team.

I would much rather wait a teeny bit longer for a game and have a hope in hell of winning.

1 Like

you are saying the forced 50% is a conspiracy

I’ve seen too many 42% lifetime and 57% lifetime Ranked profiles to believe that anything else but player performance decides your win ratio in the long term.

And those accounts had thousands of Ranked games. (Those +55% are very often grand masters, though)

Also in lower MMR (such as below bronze 1) a huge amount of accounts are even under 40% lifetime in SL.

1 Like

As you win the matchmaker, at least in overwatch, puts you with worse allies, you will trend towards 50% there is no way around it. instead of just moving you up, its trying to keep you relatively in place and gives you lower mmr allies. The only way to to start getting more evenly matched teams is to go on a lose streak, or to bust through the artificially created barrier.

This is entirely different than overwatch, and frankly seems completely borked–at least in quickmatch. Storm league remains to be seen.

1 Like

I hate MMR averaging in QM as much as you (like putting a 1350 MMR with a 2750 MMR in an average ~2k rated match), but I don’t think the company is doing it out of maliciousness.

It’s because there simply aren’t enough high MMR players or players on win streaks.
If you waited until it found you 9 other players at 2750 MMR, you’d wait an hour or more.

If you want to avoid this and you’re fairly high MMR I’d probably start playing SL only. Even with the current low population, they’re doing a fairly good job when it comes to matchmaking and ranks.

(A GM might still occassionally play with a D1, but that’s not too bad)

2 Likes

Considering all the questions regarding QM matchmaking, how roles are matched, etc., there should be a pinned topic answering these questions. I’d love a day to go by where we don’t have six new threads asking the same question.

1 Like

It matches players who are on a wining streak with players who are on a losing streak. Check the match history of every player at the end and observe.

It doesn’t though. I’ve checked many times and yes, some players on the team will be on loss streaks, but other players will be on win streaks. It matches for role, hero level and then it tries to average the MMR for both teams.

If your complaint was that MMR averaging is a flaw in Matchmaking, I wouldn’t argue against that. But there is no “forced 50%” through forcing players on win streaks to play with “bad” players.

Remember too if HOTS had a larger player pool, you could have far more precise matchmaking.

It attempts to put you into a game with 9 other players who are within a certain MMR range. And arrange those 10 total players into two teams who the game estimates both have as close as possible a 50% chance of winning based on MMR.

As the queue expands, the range of MMR examples, mirrored heroes are allowed, winrate estimate might be allowed to slip further from 50% target for each team.

The big issue is one that was noted in a different thread:

The game tracks your MMR, and applies small modifiers depending on a few factors:

  1. If a hero is low level (I believe under level 5) your MMR is modified down a small amount.
  2. If your hero is a high level (15+?) your MMR is modified up a small amount.
  3. If your are in a group, your MMR is modified up, the larger the group, the larger the modification.

The problem here, is, well… sure, my Zeratul is level 16, but I’m not good at him still.

So you have these mismatches.

And doing “MMR per hero” doesn’t work, because getting a solid number is going to take dozens of games.

1 Like

Hammer sucks so idk what you were expecting