Director’s Take – Making a Great Match
Two weeks ago, we talked about balance and matchmaking in our blog update. Matchmaking has continued to be a big topic in the community and it's something we've been focused on internally. If it's important to you then it’s important for us to talk about, so I'd like to update you all on what we're seeing and doing.
One note before we begin - I think we're at risk of sounding like a broken record with our communication on matchmaking. "We're working on it and it's getting better" isn't the best message, especially when players are still experiencing bad matches. So, I wanted to share some data to help you better understand the state of matchmaking and the impact of some of our recent changes. Let’s get into it!
Over the past week we’ve made changes to the system that have brought down the skill gap between players in a match significantly for high and low MMR matches. If you look at the Ranked Skill Difference chart below you can see two spikes that negatively impacted match quality in Season 3 and the recent changes that have started bringing this back down. We’re almost back to where we were when Season 3 started.
Looking at the high MMR group (GM players and above) roughly half of our matches have an MMR range of about 4-5 divisions, while 25% are at a difference of about 5-6 divisions and the worst 1% of matches are at roughly 10 divisions. Grouping behavior drives a large part of this gap – remember, in Masters, players can group with other players who are up to 5 divisions apart.
Competitive Role Queue Intra-match Player Delta Percentiles
█ = 50th Percentile █ = 75th Percentile █ = 95th Percentile █ = 99th Percentile
Okay, so that’s where we were as of March 6. Let's talk a bit about how our matchmaker was working at this point and the changes we’re making to improve it. I'll try not to get too into the weeds!
When our matchmaker tries to make a match, it looks for players as close to each other's MMR as possible. As it continues to search it will expand to players with larger differences in skill. Prior to Season 3, we had the ability to control the rate of expansion independently for each queue. In Season 3, we changed how our matchmaker expands over time
, but removed some of our control over per queue behavior. This meant that if we wanted to reduce the size of skill differences in Competitive, we had to do the same thing in Unranked and Arcade. If we wanted to get shorter queue times for Unranked by expanding skill difference requirements, we could do that, but then Competitive would have wider skill differences too.
In the patch we released on Tuesday, we implemented the first of a series of systemic changes (more are coming in Season 4) that give us back the flexibility to tune these modes separately and gives us new ways to tune the system. Competitive can once again have its own set of parameters for how wide the skill gap can be that are separate from our other modes. This week we will be tuning these values to tighten the skill gap in Competitive as much as possible while keeping an eye on queue times for that mode.
Additionally, in the mid-season patch we added the capability to sort parties together that have a similar player delta. We’re now able to get similar parties together in a way that minimizes role delta, which has resulted in significantly lower party queue times. It also concentrates wide parties together, which has made a dent in our player MMR delta.
Hopefully, that provides some much-needed context for all of you and helps you to see the amount of work going into the system. However, your experiences are what really matter. We'll be collecting a lot of data this week and listening to all of you as well. I'll be providing some short updates soon, most likely through Twitter.
That's it! Hopefully we'll be moving on to other topics for future updates. See you all in game!