Finding the perfect match

by Justin O'Dell on 20th November, 2014

Hello, world! My name is Justin O’Dell, and I’m one of the server programmers on the Structured Player vs. Player (sPvP) team.

I’m here to talk to you about the upcoming changes to matchmaking and ladder ranking.

To start off, the first thing I should mention is that we’re separating matchmaking from ladder ranking. Previously, your ladder ranking was based on your matchmaking rating. Removing this link gives us far greater control over how the ladder progresses over the course of a season, as well as the hidden variability and volatility seen with the old ladder. More importantly, this change gives us much greater flexibility in both areas, without which many of the following changes would simply not be possible.

Matchmaking

Matchmaking is one of my favorite topics, and I’m proud to say that, with your help, we’ve made some really significant additions that will greatly improve the sPvP experience for everyone.

To accomplish this, we decided to move to a score-based search method that takes into consideration several metrics that we believe are important for a match to be fun. We chose a score-based system over other methods because we found that it’s a good compromise between the often competing goals of match quality and short queue times.

Ratings

At the heart of matchmaking is your Glicko2 matchmaking rating (MMR). This rating helps us match you with other players near the same skill level. Now, in addition to your core rating, we will also keep track of ratings for each profession you play.

By using more than one rating, we hope to encourage you to experiment with other professions you may not play regularly. Are you a really good mesmer? We’ll match you with other great players. Are you an engineer newbie? We’ll find you easier matches while you learn the ropes!

Rank

While we firmly believe that a player’s rank doesn’t necessarily match that player’s skill level, it is a pretty good indicator of experience. With that premise in mind, we will now prefer to place you with others near your rank.

Party Size

Since playing with your friends gives you an advantage, we’ll try to keep things fair by matching you with other parties of the same size. Playing with a full team? We’ll prefer matches with other full teams. Playing alone? We’ll prefer to match you with other lone wolves. Even if we can’t find another party of the same size within your skill level, we’ll work to get it as close as possible.

And More

In addition, we’re also taking into consideration team composition, ladder position, and dishonor. Tired of seeing matches with five ham-bow warriors? So are we! We’ll also spice up competition by matching you up with your closest ladder rivals.

Ladder Ranking

Imagine a world where everyone took the saying “It’s not whether you win or lose, it’s how you play the game” seriously. Wouldn’t that be great?

Ladder ranking is now based on points which are awarded by how well you play the game and how challenging that game was. To this end, we’ve come up with a scoring matrix with two dimensions: odds of victory and final score.

Final Score
0-199 200-299 300-399 400-499 500
Odds of Victory 0-19% -1 0 +1 +2 +3
20-39% -1 -1 0 +1 +2
40-59% -1 -1 -1 0 +1
60-79% -2 -2 -1 0 +1
80-100% -3 -2 -1 0 +1

[1] Winning will always award a minimum of one point.
[2] Deserting will always award the maximum point loss, as if the player’s team had a final score of zero.

Here are some of the benefits:

  • Players are rewarded for playing well, even if they lose the game.
  • Players who play well are rewarded more often than those who have a few lucky matches.
  • Players are given a chance at success and a reason to keep fighting, even if a comeback may not be possible.
  • Players are compensated for uneven matches by increasing rewards with risk.

Now I can hear you asking, “Final score is easy, but how can you predict which team is going to win?”

After crunching some numbers and running simulations, we’ve found the two biggest factors in deciding outcomes are the difference in skill level between the two teams (based on MMR) and the difference in the teams’ maximum party sizes. So we’ve come up with an algorithm that turns those two metrics into odds of winning.

As John mentioned, we are running a test season in December, so we’re looking for your input. After this test, we’ll take another look at the data and continue to experiment, so feel free to stop by the forums and share your thoughts with us!

One More Thing…

To better coordinate with the community, I’ll be maintaining a page on the official Guild Wars 2 wiki that includes most of the technical details for matchmaking, ladders, dishonor, and server configuration, including all the knobs and levers currently in use. This should come in handy, as we now have the ability to tweak matchmaking on the fly. Keep an eye on the wiki page for more details as we make changes in search of a matchmaking sweet spot.

Also, be sure to join me on the Structured Player vs. Player forum for more in-depth discussions and answers to (almost) all of your questions.

Good luck, and have fun!