How Match Making Works

[Omcrazy.4756]

So, I hesitate to post this question because I expect it will devolve into criticisms of spvp and matchmaking in general. But I am really curious how the match making works for spvp ranked. Specifically, what MMR range does it start searching and how quickly does it broaden that search?

 

There seems to be a wide range of skill in the teams I play on or play against. Some players will just absolutely dominate me and I'll wonder how I am being placed against them (not in a QQ manner, just "wow, they are good") and some I will completely abuse, even 2v1 and I know they are playing out of their league. Now, a lot of this could simply be luck where players are getting placed too high or too low as happens in any pvp game with ranks. Some could simply be players who deserve to be in whatever bracket on one class but not others (happens with me when I switch to Engi).

 

But I also got to thinking about queue times. Queue times are faster than just about any other MMO pvp I've played and pretty fast even compared to other non-MMO pvp games. I generally take 2-3min to get a match, 5min happens rarely. The short queue time and apparent disparity in skill levels makes me wonder if the match making system is using a really wide range or widening it quickly in an effort to shorten how long players have to wait for a match.

 

Just wondering if there have been any official response regarding how the match making works as far as MMR ranges goes. If I wait in a queue long enough will my wimpy G1 get matched vs Plat3s or Legends?

 

 

[ibkillin.9610]

Basically the community is very small now so Anet's current MMR system is beyond flawed. However, at this point I'm not really sure they can do anything to fix it. What they really need to do is get rid of the duo queue PERIOD. Have it so ranked is SOLO and if you want 5 man you do the AT's. That would at least put some small band aid on this because currently it's way too easy to abuse the system which many ppl do , but I will choose not to name them since we all know who they are anyways.

[Exedore.6320]

# Details

[Wiki Article](https://wiki.guildwars2.com/wiki/PvP_Matchmaking_Algorithm)

Some of the information or parameters may be slightly out of date, but the overall concept is still valid. I'll try to summarize (my memory may be a bit fuzzy since I last digested it several months ago):

* Rating is Glicko2 and from what I can tell, evaluated based on player vs. composite team. That means your rating gain/loss is based on your current rating vs. the average rating of all 5 opponents. Win/loss is the only factor which adjusts rating; score difference, end-of-match awards, etc. do not affect rating.

* Keep in mind that rating is an approximation of your skill. For a rating system, it's perfectly acceptable for you to go up or down 100 points or so (using the scale of 0 to 2400) day to day.

* The matchmaker uses a variety of criteria to find both teammates and opponents. Rating is the dominating factor, but things like party size and profession uniqueness also come into play. For a party, from what I can tell, the average rating of players is used to find a match.

* The rating difference allowed for a match expands over time. The values on the wiki indicate an initial difference of 25 rating for the first 5 minutes, then it expands up to 1200 after 10 minutes in queue. That's 120 rating every 30 seconds. Keep in mind that opponents' rating difference from being in the queue a long time can overlap yours, but yours won't overlap theirs.

* The matchmaker takes the person at the front of the queue, looks for parties (a party can be 1 player) in their rating range to build a potential list. If that list big enough, the matchmaker starts assembling teams. If not, that person moves to the back of the queue.

* When building a team in an unbalanced case (one team has more players), the matchmaker tends to pick the party whose rating is most similar to the average rating of the more populous team. This generally means that the two highest rated players in the match are opponents.

 

# Problems/Criticisms

* **Population** Any matchmaking system struggles with a low population. This is especially true at the ends of the rating distribution (which is a bull curve). The choices for the algorithm design are: wait forever until you find a match; eventually give these players a match, no matter how bad it is. ANet's design is the latter.

* **Glicko2 isn't good for teams** This claim is largely unfounded. No rating system is perfect, but I haven't seen any mathematical proof that Glicko2 is woefully inefficient or that another system is superior (note that TrueSkill's comparisons were against Elo, not Glicko).

* **Win/Loss alone is not a good indication of individual skill** Other factors are arguably important such as closeness of match score and individual performance (damage, kills, etc.). Other games have tried these factors, but ended up using win/loss as the dominant factor. Some games kept other items as small adjustments, but they simply weren't reliable enough to be a significant judge of skill. For example, if you're slamming your head against a defensive build on a point they own all game, you may have high damage, but you didn't really contribute to the game. As a counter-example of score differential, consider a losing team killing the lord on Legacy of the Foefire when they're going to lose anyway. Instead of a 500-250 loss, they have a 500-400 loss; the winning team probably didn't care to defend lord since they knew they'd win on the point tick anyway.

* **Party vs. Solo queue** Lots of complaints that parties have an overwhelmingly unfair advantages over a group of solo players. However, solo players have almost all the same tools and options as a party, and most parties don't use those tools well. No ability to queue as a group hurts a team-based, social game.