Matchmaking SR and individual performance

Competitive Discussion
1

Does solo matchmaking adjust SR based on individual performance? If not, why, and why not do it at least in bronze/silver/gold?

DISCLAIMER: my point is mainly about solo matchmaking in lower tiers, meaning below bronze/silver/gold.

In most real team sports, a coach will pick players based on their individual skill. There are no team sports (or close to none) where matchmaking will assemble a team of 5, based on individual team-matchmaking scores. For example in football, teams are evaluated, players are not ranked based on the performance on their teams.

In matchmaking, an algorithm will pick players based on an individual score measuring the outcomes of the previous matches they played.

There are several argument against evaluating individual performance with an algorithm:

  • an algorithm cannot really evaluate individual skill properly
  • players can easily try to “please” the algorithm by gaming the algorithm
  • statistics like kills/death/per-hero stat/payload presence/assists etc do not measure the ability of a player to play well… or does it at least a little bit?

A chain is only as strong as its weakest link

Thomas Reid

I think this quote rings very true regarding team matchmaking: if individual player skill is not correlated with a match outcome, overwatch becomes a game of carrying, or it becomes random and only based on a hero meta, not on player effort and skill. It does not make sense to attach a score to an individual player, if that score only measures team performance.

Some time ago, I argued that individual skill should account for at minimum, maybe 10% to 30% of the SR adjustment, meaning, when measuring hero-relative, team-relative stats, for example:

  • winning or losing a game is a team-wide EARN or LOSS of 20 SR points, for all team mates, depending on outcome.
  • depending on your individual performance, individual players get a bonus between 0 and 10 or more, distributed across the 5 players. The worst player gets a bonus of 0, the best 10, and in between, 2.5, 5 and 7.5.

That system would still encourage players to play as a team and not as an individual, because no matter what happens, all players of the winning team earn or lose points, since the outcome score matters more than individual performance.

But with that system, players are still confident that they can be rewarded if they play well as an individual players, so they stop inspecting their teammates.

Overwatch is still not very transparent about how it calculates its SR scores either.

I don’t understand why people disagree on why individual performance should be part of the score even in a small amount, and so far, I can only side with players who feel frustrated about the game.

Maybe the meme “being stuck in silver” has some truth to it: very good players can obviously carry a bronze/silver team to victory, but it’s not true if the player is just “good” instead of “very good”: a player can improve, but he cannot “carry” enough to make a difference.

I like overwatch as a game, but the more I play, the more I get a sense of frustration when playing competitive.

1 Like
2

Some individual performance ±% is needed, or penalty for dying too much (actually the penalty should be turned to +% for other team members)

3

No.

The SR varies with win and loss and typoe of win/loss (if the ratings are unbalanced you’ll get “consolation”, “uphill battle”, “expected” or “reversal”).

You have to win, no matter how, no matter your stats, just win.

Matchmaking uses only MMR and some unknown rules to set up a match between 10 players. Its known it tends to mix up ranks instead of putting 10 players of the same average MMR.

MMR is only affected by winrate. No matter your stats, your hero, the opposing MMR. Its just win or lose.

Only gamedesigner or some dev could answer “why” they decided to set their MMR like this.

But it obviously creates huge problems like if you’re a leaver, its a loss. And this loss is worth the same as an uphill battle for the MMR, which is not fair.

4

No, the visible ranks have nothing to do with matchmaking.
Besides that, your sr pretty much mirrors your mmr always as visual decay or the “card system” is no longer a thing.

OW1 had it, and they removed it in ow2.
Here’s Gavin explaining why:
https://www.youtube.com/watch?v=sD6FvhdP2Rk&t=2961s

5

Yep but iirc, even devs talk about “diamond” mmr and stuff like that and one of the dev told (iirc on Eskay stream) that if there are 9 masters and 1 diamond, the matchmaking would rather try to wait for more diamonds players and mix masters and diamond instead of putting 9 masters togethers in one lobby.

Gavin explains why they left stats datas.

But here’s the problem with their actual version of MMR : it doesn’t matter agaisnt who you win, a win is just a win and a loss is just a loss.
So lets say, you gain a “uphill battle”, meaning the matchmaker did favored one team and you beat basically, better players than you; And then you join a game, you have a leaver or its a balanced game : it cancels your previous gain.

Which imo, doesn’t make sense. Cause it means you could climb easier if you’re lucky enough to stack “expected” victories or just if you’re lucky enough to face leavers.
Depending on the ranks, like the lower. the more leavers you have, it means the greater impact of luck goes onto your MMR. Imo, its not fair.

6

It means it is mixing mmr’s not ranks.
They pretty much go hand in hand anyways so the distinction doesnt really matter.
Just for reference, few tweets about sr==mmr:

Well, if you get uphill battle and win it, you get a bonus. So that does what you wanted, doesnt it?

if you would get the SR modifiers retroactively it could be abusable imo.
If you feel lke youre going to lose - just let the enemy stomp for less sr loss etc.

Other than that, everyone has equal chance of having leavers unless youre grouped.

7

As the OP speaks about individual performance and we see that neither MMR or SR is about individual performance ; I guess one could argue its not about SR or rank displayed but more about a way to tell if you’re good at the game or not.

Ofc a good player should perform that well that they’d win more often ; but for low ranks, it can leave the players pretty clueless about the outcome of a game and they can start thinking the game doesn’t reflect their performance.

8

Too much weight is given to a win or a loss. It’s really as simple as that to begin with.

I go again to a real life parallel. Shohei Ohtani. Awesome player yah? But if you ranked him according to his team’s (Angels) wins and losses, he’s middling at best.

It’s contribution to a team effort, whether that is a win or a loss, that should weigh heavily into the ranking of a player. That contribution needs to be measured beyond the simple Boolean of win or loss.

Until that happens, enjoy where you’re at.

1 Like
9

If you measure stats for all players for each hero in a division (kill, damage, assists, deaths, hero-specifics stats displayed ingame, payload time, and many more things), and then you calculate a performance score, based on how other players are performing for that hero, you do have some sort of insight that can help compare players within a team (only within a team, of course).

It is not perfect though, and obviously, that performance might need some balancing for each hero to be more fair, and overwatch developers would have to analyze that data to make adjustments with weights for each stat parameter, but since it’s hero relative, it’s already neutral-enough, in a way.

Of course, that doesn’t change that a win or loss should still matter more in SR calculation than a performance measurement.

My main point is that performance should help to not punish players who lose a match, while they clearly carried they team.

Overall, I don’t care about giving a bonus to the winners, so if all winners get the same SR, that’s fine.

But I see many situations where a player plays very well compared to his teammates, that team loses, and in that situation, that would be fair to reduce the SR loss for the best player, and increase it for the worst player.

A good player fighting tooth and nail cannot compensate a teammate that doesn’t make an effort to win, which is why it matters to redistribute SR losses in a losing team.

It’s easy to argue that it’s totally possible to “soft-throw”, meaning you can cause a team to lose without obvious sabotaging.

Of course, measuring individual performance is not perfect so that performance score should not matter very much, so as long as it’s only there to help a losing team and not a winning team, I think it’s a good compromise.

10

Thats pretty much what they did in ow1 and deemed it didnt work.
They said anyhow that it would be their “holy grail”, if they could make a system they would believe in.
But in ow, it is not straightforward at all.

11

Do you have a link? How much bonus/penalty did players receive? What were the argument for “not working”?

With all other adjustments they did in comp, this system is good enough if it’s only applied to the losing team in small amounts.

I think it’s still relevant to punish bad players of a losing, than good players in a winning team, if it’s done in small amounts. I don’t think they used that system.

12

https://www.youtube.com/watch?v=sD6FvhdP2Rk&t=2961s

besides all the “context” and team comps and different game modes, some heroes are inherently more “problematic” than others to valuate also.

A widowmaker is maybe quite straightforward.
But then say Lifeweaver? how do you reliably track good usage of platform and lifegrips against others?

13

I need to say it, Regression Models baby :sunglasses:, as i can extract from the recent talk about competitive with Gavin, they didn’t use such approach.