On December 6 Valve published an article about ranked Matchmaking on Dota2 official blog. This article is also useful for Counter-Strike:Global Offensive. If the firm from Seattle puts its desks on wheels, it's not only for office races but also to share their projects among workers. Here, the modifications made on the two main games from Valve mix up and give an insight on what could be done for the MatchMaking on CS:GO.

The purpose of the paper is to extract information on the functioning of MatchMaking on Counter-Strike:Global Offensive from Dota2’s explanation. It is clear that the two games published by the same company benefit from each other's developments, provided they remain two different titles in two different styles with different numbers of players. They come with their specificities, in particular about MatchMaking. Wherever possible we will try to best explain but some points could escape.

Note that we directly asked about similarities between MatchMaking systems to Valve and we confirmed that the general principles are the same. Behind the algorithms that are governing these principles are themselves very specific to each game.

First, we have to point out that developers implemented on Dota2 an existing system on CS:GO : the Ranked Matchmaking. Indeed, if you won ten matchmakings on Dota2 it did not give you a specific rank or at least, you could not see it. It seems to be not so important but it has an impact on the way players approach a game. Ask yourself if you would be equally involved in MatchMaking games without the fear to down in the rankings. However, player's level is still determined to try to balance games, we will see later how. Let's get into the chase.

 

What is a good match for Valve?

Obviously these criteria are probably never fully achieved. For each of them, the system assigns a score and if all are satisfactory, the match is created. The compiled list gives information on a criterion which is more important than another; they are not classified in a specific order. At the end of the article we will try to speculate on what Valve could use for CSGO, in addition of these elements.

About this list : MM will not particularly try to make you lose more than win, or end series of victories. If this seems to be the case it is in fact a consequence of the first criterion for a 50% chance to win for each team. If you win a lot of games your Elo will rise and the average level of the opposing team too until you lose a matchmaking game.

The number of games won does not appear in the list of criteria above, experience is preferred because Valve noticed that comparable levels of players had very different expectations Matchmaking, depending on the number of games played. On Dota 2, the developers believe that the difference in experience between the players who took part in a range of 40-120 parts is the same as the players in the range from 120-280 parts. To resume, it seems logical that the more you played matchmakings, less the next part will enrich your experience or more you have experience, the less you learn every game.

MatchMaking Rating

Firstly, you should notice that the system will assign two ratings (MMR, matchmaking ranking, sort of Elo), which differ if you play alone or with friends. On CSGO an intermediate value is determined and there is a rank to a certain MMR level.

The MMR is based on wins and losses but also on individual performance, including the first games where your level is uncertain. Perhaps some have tried to start a second account MatchMaking while having a stable ranking on the first. If you do not completely miss the first ten games won by your team, you get very quickly to the rank of your first account. We know that experienced players are more likely to reach the heads and especially the shooting distances are greater than the less experienced players. Presumably these coupled with other information such as money and weapons used data provides a few games a reasonable idea of MMR. Provided it is possible in rare occasions to see his MMR (and perhaps his grade) decrease after a win and vice versa.

The impact of players from the same lobby

Naturally as soon as players gather in rooms and in various configurations things become more complicated. In this case the system has to manage a balanced game with different levels. How can the system manage it?

The MMR of each player in a room is analysed, here we are talking about the MMR playing with friends. Depending on the level gap between players in the same room and the number of players in this room, MMR points are added to obtain an adjusted MMR. Here is a typical case proposed by Valve, on Dota2 the Radiant confronts Dire.

Two teams established by the MatchMaking players ranked in decreasing MMR adjusted

Several things to note:

- At an equivalent level, coming from a lobby gives a boost to the adjusted MMR facing a solo player. Concretely if you play against five players from the same lobby and that your team is comprised of solo players, individual MMR of each player is lower than yours. The ability to communicate is therefore taken into account.

- The difference in skill between players in the lobby D also leads to a boost (+9 * 2) since Valve has determined that play and communicate with a wide top skilled player has a positive direct impact on the level of the weaker.

- The skill gap in the lobby F is relatively low, the boost too (+1).

- Players A, E & B arrive alone, their MMR correspond approximately to the average adjusted MMR, 2700. That is why they have been placed in these teams.

- Preferably, they will be placed with both players from other rooms.

- In the same way, a room that borders another room is ideal, here the lobby D frames  lobby F. The idea is to have rooms with an average level close to the average adjusted MMR.    

Hypotheses about Global Offensive

Once the match has been completed, still need to know if it has been interesting and fun to play. For Dota2, Valve looks on the gold difference won by teams from the last time this difference cross zero and the end of the game then calculates the integral of the function on this interval. Intuitively, the higher the value is, the more interesting and thight was the game.

By calculating the surface of the red area, Valve determine if the game was balanced or not

On CSGO, one might assume that the final score would be sufficient to determine whether the part was interesting to play. With the unbalanced maps, is losing a game 12-3 / 2-13 considered "fun"? Or Valve gives a satisfaction rating side by side based on the average scores on a large sample. It must be assumed that all players have taken into account the principle that most maps are unbalanced and are for example satisfied with a 6-9 in CT on Dust2. The same way as Dota2 teams can expect a difficult early game and a late game more enjoyable based on the picks & bans.

What else could be used specifically to balance the parties? Knowing that the maps are not balanced, maybe the choice of the map and especially the choice to start on one side or the other comes in. For example the eBot can give the probability of winning a game according to the map and starting side. The impact of the first side is obviously very important, especially in the minds of players. These values could be very different depending on the skill involved.

Another big question is: how convicted accounts are handled ? Is the game deleted from the ranking calculation ? There are no cases of players ranking up without playing and not much while losing. Maybe the rectification is only visible after a win? So for the end user it does not look too much inconsistent. But it would mean you can have a fake rank displayed (fake because the game still need your real rank to set up balanced games).

Here are some ways to understand a little more about who you are fighting and why; and possibly estimate the level of players you’re facing. Once again, keep in mind the note on top: different types and four times more unique players in November on Dota 2.

Many thanks to LaMangousteFurieuse for the translation