Everyone across all ratings is pretty much complaining about winning 5 games and getting 0 rating or going 3-3 as a healer for obvious reasons.
Another company called Valve came up with a really great solution for their game called Dota2.
MATCHMAKING WITH CONFIDENCE
Matchmaking Rating (MMR) is really just made of two numbers: Rank and Rank Confidence. Your Rank is an estimate of your skill, and your Rank Confidence is our confidence in that estimate. When you play a match, we adjust your Rank based on whether you win or lose. Your Rank Confidence also goes up, because now we have more data.
UNDESIRED CLUMPING: THE WORST KIND OF CLUMPING
Dota has traditionally used a modified version of the Elo algorithm to implement Rank and Rank Confidence changes. But over time we’ve noticed a few problems with our approach. For one example, the distribution of MMRs has gradually shifted lower, causing an undesired clumping in the 0-1000 MMR range. Another problem was that players who took a break and returned were having a hard time getting back to an accurate MMR, because their previous skill level was no longer representative.
NEW AND IMPROVED CLUMPING
To address this, we’re switching to a different algorithm called Glicko, which should let you quickly and accurately get matched with other players at a similar skill level. On the math side, Glicko lets us better factor a player’s Rank Confidence into our calculations, resulting in more accurate decisions about Rank gains and losses.
The switch to our new matchmaking system will feel a little like the start of a new matchmaking season:
All players will be placed back into a short calibration mode, initially seeded by their previous rank.
Calibration is no longer a fixed number of matches. Instead, a player is considered calibrated whenever their Rank Confidence is above a certain threshold.
Upon calibration, it is likely that you will end up with a different medal than you had before. However, even if your medal changes significantly, you should expect to be matched with players of a similar skill level.
Matches will no longer have a fixed MMR gain/loss. It will be variable based on a number of factors, including the Ranks and Rank Confidence of the participants. However, we will cap the gain/loss per match to prevent particularly negative outcomes.
A player’s Rank Confidence will slowly lower over time if the player does not play matches.
We’ve been running both matchmaking systems simultaneously behind the scenes for a while now to help us build confidence in these changes. We understand that accurate MMR is only one of many factors that impact match quality, and like all changes in this space we’re interested in hearing your feedback.
Given how patches, buffs, and nerfs happen in WoW, I’d be pretty hesitant to implement anything that had rating or MMR decay. Maybe if it was over the course of months.
This is already happening to a small degree, though I do think it could be better. The issue right now is if you go 0-6 your first match, you’re kinda screwed. Likewise, if you go 6-0, you’re probably going to be placed in a bracket that’s too high.
If they did a sampling of your first 3-5 matches WHILE still rewarding rating, I think that would be best.
That’s already happening, though I do think it can be improved for healers and slightly touched on for DPS. Mostly in factoring out the 0-6 matches… and maybe even some of the 6-0 matches as well. A good 6-0 match can really protect both your CR and MMR for quite a few matches. The opposite also being true.
Like if a healer goes 3:3 5 matches in a row and then get slammed with a loss because of a counter comp, that’s an issue currently. Also, going 3:3 shouldn’t have such a heavy impact on your MMR. Going 3:3 should be as neutral as possible, but currently, 3:3 outcomes can result in large MMR drops.
As with above, I don’t like the idea of MMR decay. Maybe from season to season only… I mean, if we are staying in a +/- 200 MMR range, I’d say that’ okay.
Imo they have lots of good changes for the MMR and Blizz could be inspired to test more, in order to find the sweet spot.
It’s hard to say what could work good in WoW without experimenting a bit, I am all in for small changes frequently vs the big changes every 3-4 weeks.
Shuffle MMR should start at 0 at the beginning of the season, and they should not apply artificial inflation to it like they do with regular arenas. It inflates enough naturally just via participation to not need the increased inflation added, and starting everyone at 0 instead of 1500 mmr would slow down the beginning of season spikes for the players who are inevitably going to be at the top of the ladder anyway.
Yeah the crowd that loses down from 1500mmr before stabilizing would just stay at 0 and/or progress upwards to 1200 instead of losing down to it and CR catching up. Would also keep more players in the inbetween ranks lower mmr/cr for longer. The players who are going to go 6/6 for the placement matches until they all get in their own bracket together will just happen at a lower mmr/cr instead of jumping from almost the halfway point of MMR cap.
A thing to take into account with shuffle v regular arenas too, is that shuffle ratings move faster in both directions because you interact with the rating system in 6 game exposure units against the same players instead of match per match against different opponents. This makes the curve a lot less linear for the top of the ladder as they climb because each set of 6 wins gives a lot more points.
1500 is basically the 0 point on the graph, and with 1500 as a starting point for MMR, you technically do have “negative” mmr if you’re thinking about it in the same relative terms. That same line on a graph with a lower limit of 0 will squish the first 2/3 of the curve itself and move it down the y axis. If you look at regular arena as a linear curve of x + 1500 and shuffle as an x^1->6 curve, this at least makes the first 2/3 of the shuffle curve a bit closer to the movement of regular arenas curve.
