So, lets talk about the broken matchmaker

General Discussion
1

Or in particular why people think it is broken, and what they would rather it did.

Like, specifics. Weirdly enough, we may be in a position to influence the matchmaker design soon.

Lets see if we can even get a vague agreement on what you want it to be like.

Matchmakers have tradeoffs.

If you tighten it up, you get people complaining it is forcing a 50% win, as they predictably get outside of their actual ability to play, up or down.

If it is too loose, you will get people saying it is “too stompy”.

We get people complaining about both, (sometimes it is the SAME people).

So in which direction should we go in?

2

Give me a matchmaker that allows players to play with those at their skill-level, and not the boosted losers who queue with their friends non-stop. That’ll fix half the problems overnight!

2 Likes
3

Well, nerf Antinade and Ranged Burst Damage.

Then they can make it so there are a lot less real or perceived “Throw Pick Tanks”. (And “Throw Pick Supports”, for that matter.)

Since the can cause not only real TeamBalance issues, but also just cause a lot of unnecessary tilt. And just the tilt by itself, can cause an even match, to become an uneven match.

4

So you want a “soloQ only” mode with separate MMR rankings? That is doable.

openQ would likely to sacrificed for it though. Good tradeoff or not?

1 Like
5

Solid tradeoff.

Blizzard should do this or learn to reward those who solo-queue consistently against multi-queues. I shouldn’t win/lose the same elo as someone in a duo/trio/quad when I’m playing solo.

1 Like
6

I personally don’t have that much of a problem with the matchmaking. I like the group change they said they were planning for season 9. I’d kinda like that in QP too and not just comp.

7

Honestly, If you look at what the matchmaker is designed to do it isn’t broken. It is working just as intended.

I largely believe the problem sits with people’s expectations of what they want it to do instead.

Can we talk about changes to its design? Sure, but several things are going to be non-negotiable from the dev teams side (IMO). Those being:

  • Quickly finding matches. Blizzard needs fast matchmaking in order for the the free to play battle pass and store monetization layout to work. Not to mention both they and us have seen what long queue times do to the game.
  • Allowing players of high disparity in skill to queue together. Blizzard knows that friends need a place to get the core experience of overwatch together. Friends keep each other playing.
8

strokes chin

  • Having something which keeps track of soloQ MMR separately from group MMR is doable.

  • Having it raise up the score of games which are group vs group, or soloQ only is doable. Make those games more likely.

Basically try to fit in 2 and 3 groups together more often, AND have them use a separate “grouped MMR” rating for those players when it does.

Doable. Absolutely doable.

oh, I’m out of the loop there, what is it?

Indeed that is tricky to have. But… as Pupil says, they can have seperate MMR. After all, playing in a premade is it’s own skillset.

9

Educating people why most of their “matchmaking woes” are really just issues borne from other sources, specifically ones that matchmaker can’t control.

But that’s not happening, so alas, a man can only dream

3 Likes
10

It’s something for competitive only. They are planning on removing restrictions for grouping but if the disparity in skill is high the group will be marked as a “Wide group” and look for another of similar composition.

1 Like
11

Just post the code/algs and we’ll take a look.

Nothing rigged, nothing to hide.
Also: well-designed systems are tamper-resilient so smurfing/spoofing wouldn’t be a major issue if they hired professionals.

Cheers.

12

That isn’t the matchmaker though. Banning of smurfs etc is a separate system. But I can see why people would lump it in together.

I agree with that.

Interesting.

I’m guessing so they can have more narrow groups outside of it.

13

Separate MMR would directly impact queue times though if it strictly looked for those and ignored the expressed purpose of finding fast games.

14

It doesn’t weirdly enough.

If you are in a group, it uses the group MMR, if you are not in a group it uses the solo MMR. That is done before the matchmaker is fired up. It just gets different values depending on if you are in a group or not. The matchmaker code literally won’t even see a difference.

It is just being handed different values before you queue.

Now, having it favor group vs group and solo vs solo would, but the MMR stuff wouldn’t at all.

15

How would it not impact queue times if it was looking to match you with a group MMR? Wouldn’t the values dictate who it is looking for? Or would it still behave the same and still balance it between speed in balance With a leaning toward speed as more time passes in queue?

16

It wouldn’t be any different in code for the matchmaker at all.

You would still hand it the mmr for the roles they were queued for, and who they were grouped with.

The difference is, the MMR would be from a “this is how good they are as DPS when they are in a group based on past games when they were in a group” vs “this is how good they are as DPS when they are solo based on past games when they were in soloing”

The matchmaker still only sees DPS MMR and doesn’t care what column it comes from.

It literally doesn’t know anything different has been implemented, and it queues exactly the same as it did before.

Like you could hand something over saying

[
  {
    Player: 12,
    DPS: 0.12,
    Support: 0.13,
  }, {
    Player: 24,
    DPS: 0.2,
    Tank: 0.21,
  }, {
    Player: 15,
    Support: 0.09
  }, {
...
]

and the values come from a different place.

The matchmaker itself doesn’t know or care that DPS has a different value because they were grouped.

17

Not necessarily. Both systems are coupled and share data/info. You can false flag smurf behaviour (using poor classifiers and other metrics that carry over into assessing skill level).

So we finally agree on something? wow it took 7 years.

18

They both use the same data source. That isn’t the same as being coupled though. You can wildly change one without having to change the other at all.

The person building the matchmaking solver doesn’t have to care what the smurfing detection code has in it.

The smurfing detection code isn’t calling the matchmaker either.

A change in one doesn’t force changes in the other. So I wouldn’t call them coupled at all. They have different concerns, different code bases, and don’t need to call either other at all.

They don’t even need to be working for the other to operate.

We agree on plenty of things. Just we don’t agree on what the matchmaker should take into account. Or at least what the engine should be ABLE to take into account.

I’m like, “Rusalka 3” should let the game creators have freedom in how they want to go about doing stuff.

If they want to match people who leave games more often with other leavers, then they should be able to do that. That is something a matchmaker should have the capacity to do.

19

We should first try and remove the challenges that force people to flex queue and then see how it goes. The amount of times I had a super stompy game and when I checked the profile, it’s just someone on their off role for the challenges.

After that, we can talk about soloq and stacking. Imho, the best way to solve this issue is that if you are a soloQ player, and for whatever reason you got put up against grouped players, then you won’t loose as much SR as you would normally and similarly the other party won’t gain as much SR as they would have. This is only for cases where soloq players are put up against grouped up people.

Using this system, maybe blizz can even bring back stacking in GM

20

I’m a big fan of fixing things outside of the matchmaker which cause worse matches :slight_smile: