The average player is average at the game.
The matching system tries to average mmr to create an average probability of ‘fair’ play between the teams. So the average average is averaged.
Player populace (mmr distribution) falls into certain peaks were some values are more common than others, which influences where an ‘average’ of the mmr is going to fall. If a team has an average of 2000 mmr, than that might be a composition of: 1800,1900,2000,2100,2200 rather than 1980,1990,2000,2010,2020.
Before players ramp up their ‘win streak’ they may be in the same averaged block, but they’re matched at the lower end of the mmr (1800 in this case) so there are ‘better’ allies in their group that can be more cooperative to play alongside. Player wins some games, mmr increases, but they’re still in the same relative average mmr block (2000 in this case)
After wins pile up, their personal deviance from the average has increased, but the matching average is still relatively the same. Their game is filled with “worse allies” relative to the previous few games they were playing.
Now they’re the 2200 in the spread of 1800,1900,2000,2100, but they are the one above the others in that match For the player experience this can feel like the “forced loss” scenario while just being the 'working as intended" model for the mmr averages.
Part of why some suggest taking a break, or queuing as a different hero is to then shake up the systems expected average it has for matching. If you’ve already played several game, the matching has an ‘average’ it’s looking to make, so its not uncommon for players to end up in the same queue as players they may have already played with/against because they’ll all still in the same relative average mmr block. So taking a break, or queueing as heroes that adjust the mmr averaging then shift the block a player is matched against.
So… basic math is pretty much why the system can seem ‘forced/broken’ and “working as intended” at the same time. However, instead of pointing out how averages can work (and realizing the extent of player populace in actual ‘skill’) people usually draw up boogie man stories and have to profess the universe/blizzard/activision/whatever is personally out to get them. It’s much easier to blame something as some anthropomorphic evil impossibility (that would lead to better matching if it were true) than to do some arithmetic.
Cuz afterall, once school is out, math is the greatest evil of them all!
edit note: I am referring to mean, median, and mode in this post. “Average” doesn’t just refer to one thing.