Kaawumba has worked hard for many years to curate the guide and make it very clear for people. It’s really an excellent compilation of all the things devs have said over the years and an excellent consolidation and translation. I don’t think the devs need to say much more than they already have.
The problem is that MMR works across the whole player base very well. Overall, most players, and especially players that play a large number of games are in the right rank.
Individuals sometimes have frustrating experiences, but that doesn’t mean that MMR is broken. We do not have access to the details behind the scenes, but Jeff has said that it works well.
For MMR to work well, then win streaks or lose streaks do not need to be eliminated. MMR works well across a very large number of games, for both the player base as a whole and for individuals. Drops of 200 SR over a day or two is not evidence of the system being broken. It just represents the complications of a competitive system and the fact that MMR isn’t and can’t be a perfect predictor of the fairness of a match. Over a large number of games, it averages out to be accurate, but individual games cannot be perfectly predicted.
Further, even if the matchmaker were TRYING to guarantee a win or a loss to an individual, it couldn’t do that with specific games, that’s not how statistical predictions work. The matchmaker CANNOT enforce the outcome of individual games. It can only set up the conditions that across a statistically significant number of games would average out to be fair. So people at the same time have unrealistic expectations of the consistency of game outcomes, AND of the ability of the matchmaker to know the outcome of an individual game.
This is not correct. MMR is a very precise decimal that moves based on the information from every match.
We do not know the rules about MMR changes, nor do we know exactly when or how much it moves, but they have said that it moves according to both information about the past performance of the player and the new information of the performance in a given match. Presumably there is AI and data modeling involved which learns based on the whole body of statistics across all the games. So the movements of MMR after each of your matches use complicated rules that are not published, but which have been stated to be STATISTICALLY accurate across the whole player base. The measurement of accurate is that the overall win rate for players approaches 50% over time, which means that eventually, the MMR becomes accurate, and people win around half of their games, which implies that the difficulty is correct.