My leading Bias that finds me butting heads with folks.
MMR systems work just fine for team games like League. Its math and provable. Far to often people are expressing views with no backing knowledge. In the worst cases for me its like watching some one who can't do basic math try and tell other people how they think multiplication works. The moment you see some one telling others 4x4=44 you know something about their understanding is wrong. Im going to start be addressing incorrect MMR claims I keep seeing like this. "forced 50%" "it cant work for 5 man games" "It was designed for CHESS you dumdum."
Forced 50% destroys the MMR value used to calculate games . If the act of rigging teams harms the accuracy of the number used to rig teams then you got a system that becomes increasingly less accurate that would need accuracy. A forced 50% system would struggle if MMR is the core, you would need to swap to a different system. Proactively setting results during matchmaking is a lot harder then reactive adjustments. There are a massive # of RNG variables completely out of system control. There are many reasons players say Riot will do this, most common is # of games invested. But the thing is that MMR systems expectations for an average skill player to reach the top % of the ladder is ∞. By definition an average player will never make it there so the claims its about # of matches fall flat to me.
5v5 rng teammates. The concept works just fine. Game of skill = Skill affects ability to win = Skill impacts win%. Directly porting over Arpads Formula for a game like league? Copy and pasting the OG chess stuff? That does not work, and we know it because you can fill in the variables and test it. The mistake people make is treating every player on a team as equals 100%. Fact is not every one should have the same stake in a match. Many times when some one is learning about MMR that's the first issue they hit when they apply it to league. The make every one have the same stakes in a match and you cant do that with RNG teammates. If its a team rooster that's the same? yes because you can treat the team as a singular entity. when the teammates are constantly shifting in skill and experience you have to account for this. The way you do that is by keeping track of each player individually, and having a Matchmaker that wont build games outside of what you consider a fair range. A player with 1 game tracked game vs a player with 2000 tracked games need have different interactions. 2k games will give you a lot of data on the player. 1 match wont. Again, you can test this by filling in the variables and looking at how the system functions, for team games like lol that's about your Matchmaker Range it will pull from as fair and tracking/adjusting players individually. Microsoft Trueskill (an Elo derivative like MMR) claims to be able to work with 8v8 player games and become accurate @ 91 matches.
Its just for chess dude! Arpad Elo's system only looked to the game of chess for 3 Data points. They are the end result of a match, was it a Win, Loss, or Draw. Its not about how good a Play on turn 4 is. Its not how many Queens you take, how well you move your knights. The basic assumption is More skill = More wins. Its pretty dam obvious but let me say it and bold this just because I really want this argument to stop being used. Wins, Losses, And Draws are not exclusive to chess. Skill impact in skill games is also not exclusive to chess. This is a weak argument and what makes me say some folks are trying to tell others 4x4=44 but they really don't know WTF they are discussing.
Things I believe Riot should improve.
Setting accurate player expectations. MMR systems can be tested, as I keep saying over and over. You fill in the variables and run a RNG simulation. In doing so you find out the average for how many games it takes to become accurate at making the workable range for Matchmaker to build games so Mismatches are a Minority. I wish riot constantly provided # this to the players. I believe the reason the don't because its going to be a turn off for a lot of folks. I think the last time I saw a Red post a value it was 150 to 300 matches, that's a lot for far to many players in a single year. That being said I have seen a MMR system built and tested that ran about 400-600. In theory Soft resets do a good job of lowering this number in subsequent years but again that's something that could be tested and that value would also be nice to know. The reward for this is players tend to know when a Tier is an accurate reflection of an account.
Less efforts for appeasement. Look at the fiasco with Positional Ques this year and changing the starting MMR values. Changing the Starting MMR value impacts the entire ladder. When I say appeasement I'm talking about decisions made for player perception over competitive quality. The Entire LP system is made for hiding MMR loss from players. The expectations are that about half the players will lose MMR from their starting values. "Using MMR as the sole mark of achievement in League punishes half of the player base as their MMR will decline over the course of the season" -Riot Gortark Que times being set as a very high priority because people get impatient at the cost of quality MMR ranges? That's appeasement and hurts match quality. I don't know if riot does this. I know Que times from when I started playing to this season are clearly shorter. Most people feel the game has a lower Playerbase then a few years back and thus my expectation is the que time should be a bit longer now then it was then. Im going off the ever unreliable Anecdote evidence on this claim so take it with salt. I make it because I honestly don't think this experience is going to be different for any other player. So If you have been around and disagree that que times have gotten shorter over the years please speak up and call me out on this. Sacrificing Quality for Que times is a negative in my book and to bad for the folks who are unwilling to was 120 or so extra seconds. Moving Starting Tier to Iron was to make Silver players feel better about new accounts. Adding in extra steps to compensate it to accomplish the same end result is just a lot of work and chaos. Now you can probably achieve it at a "workable" level but its gonna take some Trial and Error.... and Live vs on paper are gonna vary... So each attempts gonna impact the entire ladder when you make that effort, and I just don't see how you don't end up dropping the entire ladder down in values.
Rework Duoing. When you know that MMR systems use MMR to make matches and then look At riots Duoing use Tiers not MMR values for what it allows players to do. It is instantly apparently that there is a very large Gap of MMR values. Tiers move slower then MMR, this is intentional. Silver and Gold can Que, Silver 4 can have bronze MMR. Gold 1 Can hit into Plat. Bronze to Plat is a very wide gap in MMR for a matchmaker system to account for. I just don't see any way this can be compensated for with out just adjusting MMR values at the end of a match for impacted accounts. You can offset its impact on others attempts at climbing but it wont change the fact your gonna see some gross mismatches and unfun games. If you don't know what I mean here then try to que with a friend at the highest end and lowest end of what duoing will allow your account to do for a week.
Duoing goes into my comments about Appeasement, this is something folks like to do and makes it more fun for them to play. But its not about making a better match or better matchmaking. I don't think its over the top or unreasonable to ask for duoing to be tied to MMR on accounts and that the MMR gap can not be higher then the value your MM thinks of as a "fair" game. Honestly id also want it to be off experience so that if there is to much of a gap in uncertainty then you can duo as well but its unlikely riot would make that concession.
If any one is curious about MMR systems etc or has some questions about functionality hit me up. If your replying to the post id recommend opening with @ [bolded topic]. If your reply is not directly related to anything I said but on the topic of matchmaker I strongly encourage you to not reply to this but make your own separate post in the thread.
