…Allowing us to manually reset are MMR once in awhile. Perhaps once a month or maybe every other month.
With the new back-end servers we can do this, but I’m unsure if or when we’d be want to do this. I’ve mentioned it to the team on several occasions, but I don’t recall what the designers said, and I know it just isn’t as high a priority as other work right now.
I am finding myself in matches that are too easy to me because before the switch to the new leaderboards I multiclassed a lot in solo queue and accumulated losses on some of my off classes. I have no idea what else would be possible to help players to play at their skill levels if they’re stuck in a weird MMR spectrum.
Then again I have played against some players like Ostrich Eggs and others who have high MMR in the past month.
These kind of matches can be caused by a number of things, but rating is probably the least responsible. Nearly everytime I’ve looked up a bad match, rating was usually the strongest factor trying to prevent the match.
Justin: Do you think MMR sometimes is overvalued metric for skill level considering the game’s design?
No, I don’t think that.
I’m aware there are some critical combinations of things (team comp, # of people queued at your rating, rank, etc..) that can be very detrimental to match quality. Those things will drop quality regardless of rating accuracy or emphasis.
That’s not to say rating is perfect. I’ve got it on my todo list to explain in more detail in a forum post, but here is the gist of two problems that are now fixed.
- Early on after the Dec 2nd build there was a bug that caused deviation to stay at the max. This bug was caused by a mismatch between how I was calculating the amount of time that elapsed since the last time you played. One section of code worked in milliseconds and the other in microseconds. I had poor test coverage of this code which is why it was missed.
- Due to some improvements we made intended to more accurately reflect a player’s impact on a team when updating ratings, the rating changed much slower. Unfortunately deviation was still changing at the normal rate which mean people zero’d in sooner than they should have relative to where their rating was. I’m still not 100% certain how much an effect this had on match quality, but I’m optimistic the fix will improve things.
These two impacted some players (especially those that had desertions early on before #1 was fixed) more than others causing them to have wildly inaccurate ratings that took far too long to adjust to where they should be.
These problems are both fixed now, and we did the partial-reset so that anyone affected will stabilize in their proper location much sooner. We could have done a full reset, but it was a minority of players that were hugely affected.
Everyone will benefit from the fix to #1, but you may not notice the impact of the fix because it will still take time for the population to spread out and stabilize.
Notice that both these problems aren’t with Glicko2, but how I set up the servers to use it. Mae Culpa.
Even with these problems, for players unaffected by the bugs the ratings are still mostly accurate. E.g. the top players are still the top players.
Isle of Janthir: Flux, Latch, Aegir
(edited by Justin ODell.9517)



