MMR based team balancing is not for Overwatch

Competitive Discussion
21

SR decay. (20 chars)

22

That’s not what ‘hardstuck’ means. If you’re hardstuck you are where you belong and can’t climb higher for a long time.

If you’re 1 whole rank below where you belong you will climb and are thus not hardstuck.

The MMR system simply ranks how good you are. Who to put you with and against is decided by the matchmaker. The matchmaker can’t tell if someone is a smurf. It also can’t tell that you are 1 rank below where you should be. The matchmaker does not work the way you think it does.

Generally, it seems you don’t understand how the system works. Your grievances are real - some matches feel bad. However you’re incorrectly allocating blame.

23

Again, someone with a poor understanding of how the system works comes to a conclusion with little evidence to support it.

Knowing how MMR works has almost nothing to do with balancing teams (which is what this thread is about - “MMR based team balancing is not for Overwatch”).

Again, the matchmaker uses MMR but is something separate.

Given this seems to be a widespread misunderstanding I’m going to go in to a bit more detail with an example:

The matchmaker sees:

  1. MMR (we don’t really know but it might look something like 2436 +/- 47)
  2. Latency
  3. Group sizes

It likely uses more than this but these have explicitly been said to be used by devs.

What has explicitly been said to NOT be used:

  1. Win/loss history (streaks make no difference)
  2. SR
  3. Level (XP)

So, the matchmaker might see players as something that look like:

Player A)

MMR: 2347 +/- 67
Latency: 42 ms
Group size: 1

Player B)

MMR: 2322 +/- 24
Latency: 49 ms
Group size: 1

Player B)

MMR: 2353+/- 97
Latency: 13ms
Group size: 1

Who’s a smurf? If might be Player B because their MMR uncertainty is high. This might suggest that their MMR is too low and they’re actually a much more skilled player.

Or it could just be a new account and the player is improving very quickly.

Or it could be a tank player who suddenly decided they want to play DPS and their MMR is falling rapidly.

Together these things mean

  1. Whether you’re climbing or falling is not used
  2. Whether you’re experienced or new isn’t used
  3. It does not “balance” you out by putting a thrower with you - it can’t see their win/loss history
  4. It doesn’t think “you’re almost at a new tier, here, have a thrower”

Other than making teams with vastly different average MMR, ping, or group sizes - which very rarely happens - there isn’t really any way it could ‘rig’ your matches.

Which brings us to

Replace SR with MMR, include a few factors like ping and group size and that’s actually pretty close to what we have.

24

A bell curve may not be the best fit but I really doubt a power series is even close (the exponent would have to be negative for this to make any sense). There is simply no way that more people have the very lowest skill level than slightly above the very worst.

25

i thik it does more accurately describe the ow skill distribution.

OWL and top500 players are exponentially better than low GM players, to the point you could probably place multiple full ranks between them. it explains is why top500 is full of players with both main and alt accounts in there.
while bronze upto diamond there really isnt much difference between skill.

power law curve would put majority of players in bronze and silver with gradual increase and then at diamond the skill distribution would skyrocket.

if you built off this kind of distribution it would require a much different matchmaker but it could produce more consistent quality matches.

the problems we see with matchmaking could be down to using the bell curve assumption as the basis. which by nature wants to draw everyone into the middle average like gravity.

26

You’re talking about a non linear scale for SR. A power series is still a dumb idea.