Skill based MM in random games has been, and probably always will be, a hot topic. Alas, many people do not understand the intricacies and difficulties of such a system and thus have flawed assumptions about how this system could be implemented and work. This post may hopefully remedie this for some people.

1. Skill based MM - what gives?

Skill based MM was a topic long before the first computers were programmed to provide entertainment for the masses. Arpad Elos ranking system was implemented in 1960 to rank chess players. His system was relatively simple when compared to the modern systems, such as True Skill but it influenced all systems after it.

What Arpad Elo understood is very fundamental: Performance can only be inferred from wins, draws and losses.

Yes, you read right, Eff, WN8, PR and what not are NOT a viable way to determine game to game Performance because they mainly use other variables, like kills or damage done.

But, but, why? you might ask.

And that's, indeed, a good question! It is however answered easily if you think about the following counter-question: what is easier, killing 10 noobs or killing 10 pros? See? 10 kills and 5k damage means nothing if we can't determine the skill of the "victims". What Eff, WN8 and PR assume is that over many many battles everyone meets the same amount of pros and noobs and thus WN8 is an indication for Performance because of this (and it really is).

But this does not help us in the slightest to implement skill based MM because the very moment that we did, WN8 for pros would fall (they would meet only pros) and WN8 or noobs would rise (they would meet only noobs) till noobs and pros would be the same. Not pretty!

2. Fundamentals of Skill based MM

Ok, so how does it work for chess players or in other games? Basically you give everyone a set amount of points (say, 1600). Now every time someone wins he gains points. The amount gained depends on the points of the opponent. If said opponent has more points than you have, you gain a lot, if he has far less points you gain almost nothing and vice versa for loosing.

Modern systems introduced means to prevent (or lessen) inflation (the top players will always gain points) and provide a variable for the certainty of the rating to account for players that have few games or that have not played a long time.

These modern systems allow it to place people after very few games. Starcraft 2, for instance, lets you play 5 games and then ranks you fairly well.

3. Ramifications

Now what does that mean for WoT? For once all the systems are ment to rank exactly one player. Team versions exist and these are either presenting the team as one player (each team has it's own ranking - thats best) or calculates an average (works ok for smallish teams).

You see that in team battles. If you play these a lot you might notice that even as a very good team you often see bad teams (the players of the team have low WN7/8). This is explainable by the amount of games that have been played in team battles. Remember that every new player starts with an average score? Well, to never see "noobs" you need to be very good AND all noobs need to have done a good amount of team games for them to loose enough ranks so that the difference is large enough. As long as there are still noobs just starting team battles and as long as you have not being declared grandmaster of teambattles you will see noobs now and then, at least.

Not likely to happen any time soon!

But what's with randoms? Well, basically the same as with teams, only less reliable (15 vs 5-7). You need to introduce a new ranking method, let it run for a while and then simply add the skill based MM algorythm to the rest of the MM procedure.

A rough rule of thumb when doing this is that the time it needs to find a match will be doubled for each additional iteration.

So what do we have already: Battle Tier, Tank Tier, MM weight, special (allegedly to prevent being bottom tier too often). Thats 4 clauses. So adding a fight would roughly multiply the amount of time by eight.

Ok, for you? Well, then remember when I said that average works well for small teams. In WoT teams are 15 vs 15, which is freaking huge!

In LoL, which has a comparable amount of players, it takes about 5 minutes to get a skill based match, 5vs5 and nothing else (no tiers, MM weight, special).

So it is save to assume that we can at least double this time, more likely triple it.

So 10 to 15 minutes for a game to have a balanced team. Still ok?

Edit There have been suggestion that building the teams and then swapping players would "easily" provide some sort of skill based MM with "no" additional time.

If you think of it, then no, this is no solution. I actually logged quite a lot of games to validate the XVM-winchance (post is ont he german WoT board, can't link from here). I re-examined these games and the case that there are actually easy swaps (same tier and type/mm-weight of tanks, swap would even out team) is actually VERY rare!

Also this would mean that WG needs to fully implement skill-based MM just for this, seldom effective, feature.

Finally to compute for all the possible team combinations would certainly take some time - time that is most often WASTED.