Bavor

Is the population of the NA server causing unusual things to happen with matchmaker?

27 posts in this topic

In the past I've done an analysis of matchmaker and XVM chance to win for solo pub battles:

http://forum.worldoftanks.com/index.php?/topic/420470-analysis-of-2115-solo-pub-battles-and-the-xvm-chance-to-win/

 

For the past ~6 months, the NA servers have had an average of under 17,000 players according to this:

https://stats.wotapi.ru/stats/wot/na/total

NA east has been averaging 12,500 players and NA West has been averaging 4,700 players.

I have run into issues in the past year where I've noticed some unusual things with matchmaker and I wonder if its due to the low server population.  When I looked at matchmaker, XVM chance to win, and other data in the past, the NA server had a higher average player population.  That was back in 2013 and 2014.  The screenshots I've found from that time period all show 25,000-38,000 players online(both NA servers) with the occasional 40,000+ players online screenshot.  Some of the screenshots of the garage on weekdays at 2:00-5:00 PM during 2013-2014 have similar player population on NA East to what 7:00-10:00 PM has now on NA East.  I even found a screenshots from 1 PM on a weekday that shows 21,000+ players online.  The screenshots from 8:00-10:00 PM during 2013-2014 usually show 35,000+ players online with about 10,000 on NA West.  I found some screenshots where US West had 10,000-15,000 players online.  Now I rarely see more than 5,000 players on NA West.

In summary:

  • Weekday Noon to 5PM(east coast time) NA server population in 2013-2014 is the same as 8PM-11PM(east coast time) NA server population now.
  • NA West server population is currently very low.
  • At certain hours of the day the NA East server currently has a lower population than NA West had during the same time of day in 2013-2014.

 

I'm wondering if the low server populations cause unusual things to happen that don't happen on other servers with higher populations.  I've noticed a few unusual things while collecting data about battles.

First unusual thing:

I've heard people say on TeamSpeak and official WoT forum that they think that matchmaker makes your last 10-15% of an XP grind harder for the next tank by placing you bottom tier more often.  I wanted to see if it was true.  So, I looked at the last ~15% of XP in my grind through several tier 8 tanks.  This is what I found:

Caernarvon:

  • Tier 10 battles in the last 15% of XP: 47.62%
  • Tier 9 battles in the last 15% of XP: 23.81%
  • Tier 8 battles in the last 15% of XP: 28.57%

T-28 Proto:

  • Tier 10 battles in the last 15% of XP: 44.4%
  • Tier 9 battles in the last 15% of XP: 27.78%
  • Tier 8 battles in the last 15% of XP: 27.78%

Inden Pz.: 

  • Tier 10 battles in the last 15% of XP: 46.7%
  • Tier 9 battles in the last 15% of XP: 26.67%
  • Tier 8 battles in the last 15% of XP: 26.67%

All of those were between 18 and 30 battles to complete the last 15% of the XP required for the next tank.

So, is this a result of clan wars 2.0 being dead and more people playing tier 10 tanks in pub battles so that tier 8 tanks are getting more tier 10 battles or is this a result of matchmaker making the end of your XP grind harder like some players claim?  In all cases the XVM chance to win for the last 15% of the XP grind was 48-53% so its not that matchmaker is placing the player on awful teams for the last part of the grind.  Also there is a rumor that tanks get better matchmaking for the first 10 or so battles in that tank.  However, I haven't studied that to see if its true.  If that's true, then maybe the grind for a tank is evened out, as far as what tier it gets on average, over the entire grind.  To test how valid the theory is that you get easier matchmaking for the first 10 battles and harder matchmaking at the end, I'd need to record every battle for the grind through several tanks.

 

Second unusual thing:

In my previous analysis of matchmaker and XVM chance to win, I recorded the XVM chance to win and if the battle was won or lost.  I rarely saw a streak of low or high XVM chance to win battles.  I tried starting another analysis several times in the past 9 months and ran into something else unusual.  I'd get streaks of 125 or more battles over the course of a week where my average XVM chance to win was 40% or lower and I'd get streaks of sub 35% chance to win.  It would happen when playing a variety of tanks from tier 6 through 10.  This made me think there was something wrong/broken with XVM and I'd delete the data and start over.  I personally use XVM chance to win as a general measure of the skill difference between the teams and not the actual chance to win.

Some of the other players who are tracking battle stats said they have seen similar things happening and think Wargaming is experimenting and collecting data for skill based matchmaking.  I'm wondering if the streaks of low skill teams over many battles is happening because many of the above average players on the NA server have either left the game or play less often. Due to the low server population and the better players leaving the game that increases the percentage of below average players in battles.  So, all it takes in a few good players on one team or a good platoon on one team to give that team a huge advantage according to the XVM chance to win formula.

On a side note, I've noticed a lot more sub 500 WN8 players in tier 10 battles.  That leads me to think that the streaks of low skill teams are due to higher skill players leaving the game and/or playing less often.

 

Third unusual thing:

Last month, I noticed the same players on my team several battles in a row.  So, I decided to take screenshots at the start of battle for a few nights of playing WoT and I noticed when I play a multiple battles in the same tier tanks(such as playing a variety of tier 8 premium tanks), I get quite few of the same players in my battles.  Sometimes its not unusual to have the same 5 or 6 out of the 30 players in my battles for 6 or more battles in a row if I play the same tier for multiple battles in a row.  This would even happen between 7 PM and 11 PM.  This does not include platooned players.  It use to be unusual to get the same players on your team for many battles in a row.  Lately, its a lot more common.  Other players collecting data on battles that I've talked with have said its Wargaming experimenting with skill based matchmaking.  I think its just a result of the low server population.

 

This has me wondering:

  • Is the low server population of the NA servers causing the unusual things to happen with matchmaking?  
  • If Wargaming was experimenting with skill based matchmaking, would it actually be easy to detect?
  • Is data collected on the NA server as valid as data collected on other servers due to lower player population causing unusual things to happen?
  • Does the low player population on the NA server along with the death of tier 10 clan wars cause tier 8 tanks to get more tier 10 battles now?

 

In the past, I noticed that the XVM chance to win formula was very inaccurate outside of the 40-60% range and wanted to develop a new formula that was simpler and more accurate.  I had noticed that just using the average skill difference between the two teams as the basis for a formula was almost as accurate as the XVM chance to win formula(using a small sample size) and was more accurate outside of the XVM 40-60% chance to win.  However I don't think I can collect enough accurate data on the NA server to develop the formula due to the low player population.

Share this post


Link to post
Share on other sites

I didn't track it but I can be sure I have never gotten 100+ game streaks of less than %40 chance to win.  Mine has been all over.  I have heard people say they lost 15+ games in a row, never had anything like that at all.  

 

I never heard of the last %15 of the grind, but that seems true from your numbers.  

 

I thought you get mercy matchmaker for the first 20 games.

Share this post


Link to post
Share on other sites
 

I never heard of the last %15 of the grind, but that seems true from your numbers.  

I thought you get mercy matchmaker for the first 20 games.

You'd have to track the full grind to be sure that your numbers aren't also true for before the last 15%.

Share this post


Link to post
Share on other sites

Last 15% of a 150k grind is just 22,500 xp, or about 20 battles.  That sample size is ridiculously low for trying to prove a malicious programming feature to make you free xp tanks.  

I'd want to see the percentages for the entire grind as well as the battles AFTER the grind over multiple tanks and accounts.  

Also, this is the ONLY actual data I've seen for server populations.  Thanks for providing the link.
However, the link only shows the past 6 months, even if you select all data.  The play is almost completely flat since school started, and there is substantially more activity (+20%) during summer vacation.  could you provide the screenshots you found form last year?

Share this post


Link to post
Share on other sites

I concur with those that state measuring the full grind...but the variables involved in that type of analysis would be overwhelming in itself and you'd come out looking like this...658e40b32264af34e3b37d18156aa0be.jpg

Share this post


Link to post
Share on other sites
 
  • If Wargaming was experimenting with skill based matchmaking, would it actually be easy to detect?

Yes, as long as you enable replay collection and it applied to a significant proportion of battles. However, if they were experimenting with skill-based MM, it's pretty implausible that they'd do so on NA.

Otherwise #1 is cherry-picked and statistically insignificant. #2 sounds like an XVM bug. #3 is reasonable for low-pop servers.

Share this post


Link to post
Share on other sites
 

I thought you get mercy matchmaker for the first 20 games.

I haven't taken samples etc, but I have noticed for quite some time that your first match is always top tier, and the second match gives you the worst possible mm.

Share this post


Link to post
Share on other sites

That link to sever population data is fascinating
A few other observations from looking at the peaks:

  • Thursday is the least played day, Sunday is the most
  • Looks like for the average week, there is about 11% higher peak activity on Sunday than Thursday
  • The effects of a patch drop: on wed Nov 18th, there was about 21% lower peak than normal wednesdays.  

Why does WG drop patches on wed, if thursday is the low point?  

 

Share this post


Link to post
Share on other sites
 

Your sample, n=?

Sample size for what exactly?

For the first part with the grind of the last ~15% XP for the tier 8 tanks it was:

Caernarvon: 21 Battles

T28 proto: 18 Battles

Inden Pz.: 30 Battles

raVgfEx.jpg

 

 

I didn't track it but I can be sure I have never gotten 100+ game streaks of less than %40 chance to win.  Mine has been all over.  I have heard people say they lost 15+ games in a row, never had anything like that at all.  

 

I never heard of the last %15 of the grind, but that seems true from your numbers.  

 

I thought you get mercy matchmaker for the first 20 games.

The 125+ game streak of low XVM chance to win wasn't every battle being low.  It was the average XVM chance to win for 125+ battles.  It was a mix of everything from sub 20% to a few in the low 70% range.  However, the average was below 40%.  

 

The last part of the grind part was a bit shocking to me.  I thought it was a fluke when I did the first one, but the next two had similar results.

 

You'd have to track the full grind to be sure that your numbers aren't also true for before the last 15%.

I'll have to buy another tier 8 and track it across the entire grind.  I almost have the 45.02A researched.  I also will have to try it with some lower tiered tanks.

Share this post


Link to post
Share on other sites
 

could you provide the screenshots you found form last year?

I'll be able to get the rest of the screenshots and post them when I have access to my desktop again in a few weeks.  I'm in the process of moving.  Here is what I have on my laptop.

Prime time battles:

Almost 41,000 players total. Notice the 15,000 on West?

Sept%201-2014%20945%20PM.jpg

Almost 40,000 players total. Notice the 15,000 on West?

Sept%201%202015%201005%20PM.jpg

Almost 36,000 players total. Notice the 11,000 on West?

Sept%206-2014%201006%20PM.jpg

 

Afternoon battles:

25,505 players total, 19,000 on East, which is a typical prime time server population now.

Sept%2012-2014%20446%20PM.jpg

18,296 players total in the afternoon.

Sept%2015-2014%20339%20PM.jpg

21,066 players total, which is a typical prime time server population now.

Sept%2015-2014%20357%20PM.jpg

26,443 players total.  Almost 21,000 on NA East, which is a typical prime time server population now.

Sept%2021-2014%20127%20PM.jpg

17,714 players total at 3:15 PM.

Sept%2022-2014%20318%20PM.jpg

 

 

Yes, as long as you enable replay collection and it applied to a significant proportion of battles. However, if they were experimenting with skill-based MM, it's pretty implausible that they'd do so on NA.

Otherwise #1 is cherry-picked and statistically insignificant. #2 sounds like an XVM bug. #3 is reasonable for low-pop servers.

I'm not sure it should be considered cherry picked.  I had the same results for three different tier 8 tanks and chose those three because they were the three tier 8's I wanted to finish the grind on at the time.  If it was random, wouldn't I have different results for each tier 8?  The results seemed pretty consistent.  It should be looked into further with record keeping of an entire tank grind for multiple tanks of various tiers.

I could be an XVM bug.  I'm not sure what would cause a bug like that with XVM.  XVM would have to be ignoring one or more players on my team for every battle for some reason.

Share this post


Link to post
Share on other sites

I just finished grinding the KV-4 over the past few days, mostly on West.  I had mostly T8 battles, and only a few T9 battles, and even those were only a few T9 tanks each side, and I can only remember maybe three T10 battles.  Perhaps there were one or two more.  But it was definitely by far T8 battles.  I was thinking, hey, this is great, I'm finishing up the grind and I get to be top tank most of the time. 

You can't come to general conclusions from a few anecdotes.  

Share this post


Link to post
Share on other sites
 

You can't come to general conclusions from a few anecdotes.  

ancedote - I do not think it means what you think it means.

 

Data about three separate tank grinds where the tank was bottom tier 44-48% of the time in the last 15% of the XP needed for the next tank is not an anecdote.  Your KV-4 story is an ancedote. 

Share this post


Link to post
Share on other sites
 

ancedote - I do not think it means what you think it means.

 

Data about three separate tank grinds where the tank was bottom tier 44-48% of the time in the last 15% of the XP needed for the next tank is not an anecdote.  Your KV-4 story is an ancedote. 

 it is closer to an anecdote than a statistically significant sample size.

 

Share this post


Link to post
Share on other sites
 

For the past ~6 months, the NA servers have had an average of under 17,000 players according to this:

https://stats.wotapi.ru/stats/wot/na/total

NA east has been averaging 12,500 players and NA West has been averaging 4,700 players.

....

This has me wondering:

  • Is the low server population of the NA servers causing the unusual things to happen with matchmaking?  
  • If Wargaming was experimenting with skill based matchmaking, would it actually be easy to detect?
  • Is data collected on the NA server as valid as data collected on other servers due to lower player population causing unusual things to happen?
  • Does the low player population on the NA server along with the death of tier 10 clan wars cause tier 8 tanks to get more tier 10 battles now?

 

In the past, I noticed that the XVM chance to win formula was very inaccurate outside of the 40-60% range and wanted to develop a new formula that was simpler and more accurate.  I had noticed that just using the average skill difference between the two teams as the basis for a formula was almost as accurate as the XVM chance to win formula(using a small sample size) and was more accurate outside of the XVM 40-60% chance to win.  However I don't think I can collect enough accurate data on the NA server to develop the formula due to the low player population.

 

Others factors to consider, regarding matchmaking on the NA server:

1.  Being mutual friends or having an one way 'friend-ing' relationship, is taken into effect by the matchmaking system.  If you friend a person, you will see them more frequently in random matches compared to not being a friend.

2. Blacklisting individuals, adds a layer of rules to remove individuals from your teams more frequently than not.  However if and when the population is 'low', given the constraints of your server, this Blacklist layer of rules is relaxed more frequently.

3. The matchmaking battle tier range of your vehicle selection does matter, whether you've see less than 15 individuals on a given team.  For example, playing in American T34 at 11pm PST on the NA Server can sometimes return teams with only 7 people per team.  I had that occur yesterday in fact, on NA West server on Mountain Pass map.  Granted it did feel more like a tournament match-up, than a regular random pub match because of the limited numbers of players.  Fewer ones teammates, the more teamwork matters.

4.  For random matches, the modes you have selected limit or enlarge the population pool whom you queried with, when forming a match.  If you have Encounter and Assault modes turned on, you should have a great pool of individuals to pull from and see in a given random match.  Whereas if you only have Standard mode enabled, your limited your pool of possible opponents & teammates from a given total population.

 

All this factors are subject to manipulation, by individuals, clans, Wargaming Inc, bots, and various domestic-foreign government agencies.  Point being, what you see is not necessarily a random outcome in terms of matchmaking.   This is not a good outcome in my point of view.  Its too easy to rig things, whatever their various individual and/or group motives maybe.  When a competition is seen as rigged by the greater public, little good will come for any gaming companies' product & services.  Customers can and do walk away.  

If you doubt this claim, take a look at various professional sports' associations records, before and after, when 'rigging' becomes known to their audiences.   Majority of customers, fans seek and hope for an honest sporting event -- an so called level playing field. Why play or watch if it's a predetermined outcome?  Why support that? 

 

 

Yeah, the data of 70 battles split between 3 tanks means jack squat.

    Data isn't that difficult to find, when one has a personal archive of over 31000+ battle replays on file. 

There's a good chance, if a person has played random matches on the NA server at some point in time, I've come across them in-game at least once.  Granted my sample has a bias of battle tiers 1 to 10, which means it has a blind spot for small amount of people who only play battle tier 10 & 11 vehicles always (e.g. AbsoluteBum et la.)

 

  Of course, Wargaming has altered their matchmaking system over time: vehicle battle tier system (e.g. artillery, scouts et la); supposedly sought to stop creating limited premium vehicles.  World of Tanks has changed, as we all have changed over time too.  

 

 A long time ago, when I was a total newbie at the game there was an infamous 'clubber' who drove a T-26 around far too much.  What did he suggest doing to avoid some of the faults of the Matchmaking System?  Play limited matchmaking vehicle MORE when the low population.  Why?  To seek to avoid the far too common problem of bottom tier matchmaking status many times in a row.  

Point being if and when a Server population becomes too low, individuals tend to chose what vehicle to play differently.  The traits of a vehicle sought become different, than if a Server population is much higher.

Share this post


Link to post
Share on other sites
 

Second unusual thing:

In my previous analysis of matchmaker and XVM chance to win, I recorded the XVM chance to win and if the battle was won or lost.  I rarely saw a streak of low or high XVM chance to win battles.  I tried starting another analysis several times in the past 9 months and ran into something else unusual.  I'd get streaks of 125 or more battles over the course of a week where my average XVM chance to win was 40% or lower and I'd get streaks of sub 35% chance to win.  It would happen when playing a variety of tanks from tier 6 through 10.  This made me think there was something wrong/broken with XVM and I'd delete the data and start over. 

 

I had the same on EU, all the time. The streaks are insane and not explainable with random distribution. Longest streak with constant shit teams was 1000 games. Which made me quit WoT for good 3 months ago. Have not looked back.

WoWS is a bit strange too, but it does not bother me that much there, because there are less steamrolls and you can still have a fun game and get your damage in. Also it "carried over". Had 650 total shit games in WoT (my last 650) and then stopped and started WoWS. Had 34% win rate in my first 350 games. Yes 34%. Then it stopped and went to over 60% now normal for 1000 games.

They really should stop fucking around with that and let it be random. Or stop pretending and actually implement skill based matchmaking officially.

 

Share this post


Link to post
Share on other sites

No mention if you did take account time of the day you play matches? Unless you always play same time, it skews #1 totally. At least in EU there is LOT more top tier tanks in queue later night than daytime,

Share this post


Link to post
Share on other sites
 

I'm not sure it should be considered cherry picked.  I had the same results for three different tier 8 tanks and chose those three because they were the three tier 8's I wanted to finish the grind on at the time.  If it was random, wouldn't I have different results for each tier 8?

If it was random, then this result would still happen a significant proportion of the time, and you'd be far more likely to publish it if there was an apparent pattern.

 

However, it is possible to bias your matchmaking on low-pop servers. Let's say you survive a tier 10 battle and hit the "battle" button shortly afterwards. Because the other players who survived that battle are likely to do the same, an ideal mix of tanks for a tier 10 battle will be added to the queue, and so a tier 10 match is more likely to fill up. This can theoretically lead to runs of good or bad MM.

You can exploit this by waiting a few minutes before hitting "battle", or by switching tiers to match the last battle's tier. Note that the effect inverts if you don't survive the battle.

Share this post


Link to post
Share on other sites
 

Others factors to consider, regarding matchmaking on the NA server:

1.  Being mutual friends or having an one way 'friend-ing' relationship, is taken into effect by the matchmaking system.  If you friend a person, you will see them more frequently in random matches compared to not being a friend.

2. Blacklisting individuals, adds a layer of rules to remove individuals from your teams more frequently than not.  However if and when the population is 'low', given the constraints of your server, this Blacklist layer of rules is relaxed more frequently.

Do you have source for this?  I've never heard it mentioned before.  If true, this means you could blacklist unicums and friend tomatoes have a better MM?

Share this post


Link to post
Share on other sites
 

Do you have source for this?  I've never heard it mentioned before.  If true, this means you could blacklist unicums and friend tomatoes have a better MM?

I have never heard of this either.

Share this post


Link to post
Share on other sites

Yeah, I'm not sure if Valan was saying that is a possible WG tweak of MM or something he thinks is in the game, but I haven't heard anything about it, nor seen anything to indicate it's true.   A bit too much tinfoil hat in some of these comments; even if I disagree with the OP conclusions he did actually track and post data points.

 

Unfortunately, you can't retroactively apply XVM WR chances on replays; you will see the chance to win based on player current skill rather than when the battle was played.

Share this post


Link to post
Share on other sites
 

Do you have source for this?  I've never heard it mentioned before.  If true, this means you could blacklist unicums and friend tomatoes have a better MM?

 

 When one keeps on sees the same mutual friends over and over in random pub matches each day, year after year of play -- the thought does arise.  Especially after going through a process of friending, unfriending, then re-friending multiple individuals over a long period of time.  It's not hard proof per se.   However when one considers a series questions related to why did Wargaming Developers previously 'limit' the sum of every person's Blacklist + Friend list?

 

 Suggestion being, the combination of a Blacklist + Friend list direct impacts how the matchmaking system operates.  If you want to learn how to improve your game, you want to play with better individuals' than you current are.  Additionally you seek to avoid playing with individuals who deal team damage and/or actively harm your team systematically.  Thus it's a pushing away of those whom actively harm, and pulling in those who active aid or help a team winning.   If you care about learning, doing better in terms of winning, there's no foul in acknowledging some individuals are better at things that you yourself are -- yes?  

 

  In conclusion it's the very opposite, seek out those who have better overall skills & judgment at the battle tier range in question.  You'll still have bottom tier matchmaking status as much, however they will be more frequently in your games because Wargaming Inc seeks to promote the social element of this game.  Why? Friends tend to keep individuals playing this game longer.

 It's no accident Wargaming Inc promotes 'Clan' formation, it costs gold recall.  What new feature arrived with Patch 9.13?  

  Answer:  View Player's Clan profile (e.g. right click a friend list name to see it pop up).   It shows all clan members names, a few statistics, plus Stronghold map & times of play. Overall it's an user interface improvement, long overdue.

     

Why does seeing the same people over and over matter in the end?

It ends up effecting the a person's feedback cycle, which impacts ones learning curve rate.  It's somewhat like a difference between a low alpha shell rapid D.P.M gun hitting a tank, versus a high alpha shell with a long reload cycle hitting the same tank.  Point being a many small hits tell a clueless newbie where the enemy is at, more easily than an one time shot does.  Its more predictable, for both.   

 

What is one measure of how to determine whom is actively harming a team consistently? Or is a very poor learner?  

(#1)  [Individual Personal Rating] minus [1000 * Average Tier played]. That's one measure of how. Definder does rather poorly on this measure.  Bot like individuals, and beginners who try to learn this game with tier eight premium vehicles to start with, will do badly on this measurement commonly.

(#2) Individuals who deal team damage actively, to anyone on either team.  If team mate begins a match by killing a team mate. That's absolutely horrible.

(#3) Individuals who causes chaos in the team chat.  There's only so much bad mouthing one can take.  Point being, is it constructive criticism or is seeking to destroy one's emotional state?  

 

If team mates are being supportive. Giving constructive criticism. Great.  

If team mates are instead actively seeking to destroy their own team's enjoyment of this game.  That's where I draw a line in the sand, and Blacklist them for life.  

Things become more 'gray' when its one of the best of best, having an emotional breakdown and trash talking to harm of their own team.   Question is, is it worth it?  Fingers giveth, Tongue taketh. Everyone has their own lines in the sand, so to speak.

 

Yeah, I'm not sure if Valan was saying that is a possible WG tweak of MM or something he thinks is in the game, but I haven't heard anything about it, nor seen anything to indicate it's true. 

 To to clear, yes I strongly suspect Blacklist and Friends List are part of the matchmaking system.  It'll effect the frequently at which one will or will not, see certain individuals in so-called 'random battles'.

 

Meaning if you do not want to fight your unicum friends' platoon in a random match, check to confirm they are on a different server and/or in a match already BEFORE clicking the 'battle' button.   It's no accident many of the top individuals in terms of win rate in the Hall of Fame have done this for years.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

  • Recently Browsing   0 members

    No registered users viewing this page.