Developer postOriginal post (opens in a new tab)

Professor Glickman on the Ladder with a Command Card

Hi everyone!

I’m here today to write to you about three things, all of which are connected. They are our PvP rating system (which we’ve often referred to as “true rank”), Ladders, and Matchmaking. The last two systems are built on top of the first, so I will focus mainly on that. Recently there have been many questions regarding how our rating system works, how the ladders work (“Why is person X (not) in Grandmaster?”, etcetera), and how matchmaking works. I should warn you that this blog post may get a bit technical here and there, you can blame me for being a programmer ;)

Skill Rating Explained
Whenever we have been asked what true rank is and how it’s calculated, we’ve described roughly as follows: you win against someone, you gain points, you lose against someone, you lose points. How much your lose or gain depends on your skill rating compared to your opponent's. If you’re, according to your skill rating, a weaker player, but win, you gain far more points than if you win against someone of a similar rating.

More recently, people have been asking for us to switch to Elo — a skill rating system designed for chess dating from the 60’s, and has generally proved to be simple, yet efficient. What the people asking were unaware of, is that we’ve been using Elo rating since the start, and even in our previous title “March of War”. However, what I found while investigating player reports on several issues, is that our implementation of Elo was faulty! Elo ratings usually start between 1400 and 1600 for new players, whereas ours started at 0. This in itself would not be a problem (since it’s only relative values that matter), but we should have been allowing values under 0, and we did not.

Recalculate All Ratings?
The perceived solution to this was simple. Our Ladder System relies on your skill ratings to determine which tier you belong in, so a faulty rating implementation, and the faulty distribution it resulted in, seemed to indicate a simple solution: Start new players at 1500, and recalculate all existing players’ ratings using this new starting value.

While the initial results seemed promising — the skill ratings certainly seemed more spread out than before — we ran into a problem: like before, the large majority of the player base is centered around one skill rating. Previously, this was 0, and now, it was 1500. This made it very hard to determine which skill rating belongs to which Ladder tier. Grandmaster traditionally in chess has been associated with 2500+ ratings, but beyond that it was hard to pick any other boundaries.

Enter Professor Glickman
Soon enough, I discovered we were not the only ones to have this problem: In fact, it seems mostly everyone using Elo finds the vast majority of players centered around one point: the epoch, regardless of what that value is.

Luckily, since we were not the only ones with this problem, there are people who developed alternatives to Elo. One of these persons is Professor Mark Glickman, who developed the widely used Glicko algorithm in the 90’s, and its successor Glicko2 after that. People using these new algorithms reported that:

“After changing to Glicko-2, players have been far more evenly spread-out within the rating ranges, making it far easier to find an opponent of appropriate strength. It is also far easier to maintain a rating.”

Sounds too good to be true? Let’s find out. I set about incorporating the new algorithm, and had it recalculate everyone’s ratings, with, once again, 1500 being the default rating.

That’s more like it. A nice and even distribution, centered around 1500, without being too densely populated there.

Ladder System
So what does this mean for the Ladders? One of the most frequent questions we’ve been getting is: who ends up in which tier? Why haven’t I been promoted though my score is so high? And so on. Previously the boundaries for tiers have been calculated dynamically, which made it hard to answer these questions conclusively. Even more so since most players’ ratings were so close together.

Now that we have the nice distribution you can see above, the Ladder tier boundaries will be fixed henceforth. Of course, these may change in the future, but if they do, we’ll announce it. These will be the boundaries, based on the data:

  • Grandmaster: 2100+
  • Diamond: 1900-2100
  • Platinum: 1700-1900
  • Gold: 1500-1700
  • Silver: 1300-1500
  • Bronze: 1100-1300

Like before, a few stipulations apply. First of all, during the first week of a Season, the Grandmaster league is closed, and people with 2100+ ratings will be placed in Diamond. After the first week, players will start being promoted to Grandmaster, until 100 people are in that league.

Secondly, if your rating is 1695, and you gain 6 points, you won’t immediately promote to Platinum. You have to cross a certain margin (which currently is 18 points) before you’re promoted. This avoids people being promoted & demoted constantly when their rating is near a boundary. The same applies for being demoted: Your skill would have to drop 18 points under the lower boundary before you’re demoted.

But… What’s my Skill Rating?
I’m glad you asked. Now that we have shown our hand, so to say, given how your ratings are calculated, I feel that it is no longer justified to keep this rating hidden. Because of that, in the near future your Skill Rating will be displayed on your Profile, for you and anyone else to see.

What this means for Matchmaking
Fundamentally, nothing significantly changes to how our Matchmaking works. We try to find players with a similar skill rating, and that’s the gist of it. However, now that our ratings are more spread out, and more reliable, this should automatically have a positive, and hopefully perceivable, effect on our Matchmaking. Time will tell!

What does this mean for me?
Hopefully, a better experience in our game! To be more specific though, the most noticeable thing is that a fairly significant number of players will see themselves being promoted & demoted in the Ladder until they reach the tier they belong in according to their new rating, and the Ladder tiers’ new boundaries.

Well, that's it! I hope this post was informative for you. I'll see you in Face Off!