On one hand, I didn’t know that the scaling was in sigmas. On the other hand, the scaling is irrelevant.
On one hand, they never before said that MMR is unbounded. On the other hand, I would have been very surprised if it was.
On one hand, it should finally convince everyone that MMR is a single number (per role, going forward) and SR is not used in matchmaking. On the other hand, the conspiracy theorists don’t seem to care about actual information.
So, in total, it added some clarity, but nothing particularly essential.
Seeing decay is being ditched, there is little reason now to seperate mmr and sr.
You may not recall but I did the assumption that if mmr is completely separate to sr and never tracks/uses/acknowledges sr you still can have “handicapped” games where two accounts at the same sr can have drastically different game experiences by design of the matchmaker. This is caused by mmr diverging vastly from sr, and if the mmr assumption has changed quickly due to overestimation of the players skill (rolled game either direction) it can lead to the “forced loss” experience.
As I have established the divergence occurs, I believe this to be the case.
So my hope is that SR is now just a pure conversion of mmr, and such variance therefore cannot occur. Match quality should improve drastically, and conspiracy posts of streaks will reduce massively.
I’ll try to repeat the experiment again once the changes go live, but could take months to hit a narrow SR window while one account is “streaking.” (And on same class of course)
I didn’t comment in that thread, because I believe that extrapolating SR/MMR details from wait times is fool’s errand, as so little about wait times has been described, and I can easily come up with alternative explanations (which may or may not be true) for any behavior you come across.
In the case of your test, they may have a special rule that people from the same IP, but not grouped together, are not put in the same match to reduce win trading. One person goes into the normal queue (for a match that has almost formed) while the other has to wait for a completely new match.
Another possibility is that 8 SR actually was significant in this case. One account had a predicted wait time just under the warn threshold, and the other account had a predicted wait time just over the warn threshold.
I make no statement that either of these possibilities is true, but it is more likely than your explanation, as it does not directly contradict Jeff.
Match quality should improve, and streaks reduce in length, but purely because role MMR will be a better predictor of how players will perform in a match.
I don’t think so. If you’re a +3 you’re T500 and if you’re a -3 you’re <500 SR. Never the two shall meet. The +3 will be matched with other people very close to +3 and will not be expected to carry, just to play like a +3.
The assumption is that I did not consider these things. I did, most likely more than anyone else since I was the one seeing what happened and bothered to watch if it was repeatable. It boils down to whether you believe me on my responses below.
Played plenty of games with and against family at same ranking with no warning, in all modes including comp, from these two machines, while not grouped. Actually the SR range was larger between accounts. IP does not appear to be taken into consideration.
This is actually impossible, as both accounts sat queued for 30 seconds before I confirmed the queue warning photo was legable. If anything, I have added a person to queue, so the warning should have gone. Also see above, where wider SR gaps have led to no warning queue message for both but played in same match.
If we just assess the situation, all I have shown is that the mmr between those accounts is large. The only quote this “contradicts” is that SR chases mmr closely. We have no given factor for closely.
If it wasn’t predictable, and the third time, I might agree. It is the only measure we have.
This one, I don’t know. You’re making assumptions about how matches are formed and how the message is generated that can’t be verified.
But, I do have a MMR - SR is large test. Did you write down the SR for these two accounts for the previous and following ten games? If yes, it is pretty easy to see the divergence for decayed and leaver accounts. See 3000+ Skill Rating Data and Analysis (now including DCs). Basically if MMR (rescaled to SR units) - SR > 50, and SR change per game is plotted, it shows a very simple and repeatable pattern.
Consistent is key if you plan on forming a group/team, and basing that off of the much more volatile number is not the way to go.
Think of it this way: Which is more reliable in a survey or experiment? N = 1 or N = 10? Sure neither of them are a lot, but if you find an outlier, that’s all the 1 shows, whereas you have something to level it out with the 10.
I mean, you’re right. That’s not what Overwatch is about.
People have been forming teams since the beginning of competitive, both professional and casual. I promise you most of them aren’t trying to crack the system, and people aren’t going to suddenly start out of nowhere.
Even if I’d recorded the SR changes, you yourself state that decay of 50SR takes 12 games to fix. Add in PBSR and that variance would be undetectable for larger variances. There is a reason why you did it specifically SR3000+ after all.
I was thinking today, after i won a game where i just kind of guarded the point as Roadhog. I just thought his presence there would be good, and i didn’t chase the enemies around for stats. So, i would not be surprised if my MMR after that game was a bit lower. We won though.
And i guess that might be the danger of showing people’s MMR. People might start judging others, or themselves, based on some extremely specific number that may or may not really MEAN you are better or worse than the others as a player.
MMR is supposed to be stats based? Well, stats don’t mean everything in OW, ALL the time. Sometimes. Just, not every game you need to be a stat monster.
So, i don’t know. On one hand, i hate that we have a hidden rating that determines the matchmaking. It’d be nice to SEE what your “actual” rating is.
On the other hand, i’d like to think that SR is close enough. Seems more based on wins and losses, and since you can win by playing smart, and not putting up massive stats all the time, maybe that’s good.
The theory that MMR is based on stats and SR is based on wins is a myth. They are both calculated very similarly, and typically are very close to each other (after rescaling MMR to be in units of SR).
yes however, they have left it to chance where forming teams yourself is not feasible. If they had all this QoL for the LFG, then they would need to make the MMR formula public. QoL for the LFG
There isn’t really much to “take into account”. If a streak is causing uncertainty to increase, then SR gains will start to go up after the streak has gone on long enough. If you have recorded your SR for enough games, it will be apparent that that is going on.
As you say, leaving and decay do not affect uncertainty.