Source: wargaming.livejournal.com/845.html

Hello everyone,

well, for once, this is something noone else seems to have, if I see it correctly :) The link above leads to the new Wargaming LJ blog (in Russian of course). How do I know it’s legit? Because Cannoneer (WG RU developer) sent me a link (so credits go to him).

So, you’ve seen the MM video, right? If you haven’t, you should, because this post refers to it. In order to repeat what was in the video though:

- there is a possibility of the MM to create non-standard teams. The amount of such battles should not exceed 5 percent from the sum of all battles, but in concrete situatios this number can vary in accordance with the theory of probability. For example, if you played 10 battles with IS-3 in a row and in 3 of these battles you encountered a breach of the “first 5 slots” rule (tiers of top 5 vehicles were not equal), it’s not a MM error and it most likely won’t repeat itself in next 10 battles.

- the MM tries to align the team composition with the current state of queue, so that in cases, where the team composition is non-standard, the template for selection of tanks into teams can be significantly different from the default template (used when the server is started) (SS: if you remember the video, it is stated there that “weird” battles happen for example shortly after server restart, but in fact before the templates, based on the queue are formed, the server uses a “default” template for team building). Statistics of played battles does influence the default template from the first minutes and as soon as in 30 minutes, the template corresponds to the current composition of the queue and “keeps track” of the changes in it. For example, on the test server the queue might be filled with vehicles of one type the player joing the queue with and it’s possible to notice that there are more and more of these “popular” tanks and their waiting time gets lower and lower.

If the vehicle stands in queue for more than one minute, the matchmaker tries harder to push it into battle, reducing the team composition requirements. But there are 30 vehicles entering the battle (2 teams with 15 players), so even if you wait less than one minute, you can still get into the battle, in which the team is composed in order to push the long waiting player in.

There is one more reason for the appearance of non-standard teams: platoons. If the platoon consists of vehicles of one type (for example three tier 7 heavies), the MM easily picks a team for such a platoon. And if the platoon is a “trollplatoon” consisting of one tier 10 and one tier 2 tank? The MM can correctly handle even this situation, but the composition of both teams will likely be non-standard. And there is one more type of platoons, for which the MM can make non-standard teams: platoons that have artillery or scouts in them. These classes of vehicles have their own MM rules and platoons with them should not break the MM rules.

While collecting the team for the battle, the MM fills empty slots in it while moving gradually from slot 1 to slot 15, skipping those that have already been filled. At the same time it keeps an eye on the queue and from the queue the vehicle, whose MM weight is closest to the one assigned to the empty slot is picked. But even if the MM weight of the vehicle differs from the one of the slot, the slot is filled with it and the MM weights of unoccupied slots are adjusted so that the sum of the team MM weight doesn’t differ by more than 10 percent (20 percent in case there is a player in the queue, waiting for more than three minutes). This explains the cases where the teams have a significantly weaker tank on the last slot of the team: the MM corrected the MM weights of the slots several times and the last unoccupied slot proved to be the hardest to fill.

Slots of both commands are filled in sync, from slot 1 to slot 15. “Rearranging” of 30 already picked vehicles in order to ensure the equality of compositions does not happen, so even if one team has a group of vehicles of the same type and the other doesn’t, when the other MM conditions are met, such teams are sent to battle. When filling the empty slots in teams starting from 6th slot, the MM does not crosscheck the type and tier of the vehicle against the slot with the same number on other team, so it makes no sense to compare “pair to pair” both teams in battle from 6th place onwards. But even in first 5 slots the rule of “equal pairing” can be cancelled, if there are long-waiting vehicles in the queue.

If the amoung of players playing on the server is low or if there is an overflow of one type of vehicles in the queue and the battle waiting time increases, in such a cases following exceptions in MM rules are made:

- there can be battles with more than 5 arty per team (if an arty player waited for more than three minutes for battle)
- teams that are not full can be released into battle (7-14 vehicles), both teams have to have the same amount of vehicles though (this happens when any player’s waiting time is longer than three minutes)
- light tanks with increased MM weight (scouts) can assume the slots for medium and heavy tanks on the top of the team
- a notable disbalance of tech class and tiers might appear within the sum of MM weights of the team (this concerns arty)
- starting from battletier 7, a medium tank, TD or arty can appear on the top position of the team
- if the waiting time is longer than 5 minutes, the tank drops back to the hangar from the queue

All cases written above are however an exception.