Tbh, they keep smaller seasons and placement matches to try minimize the deviation that always would happen between the 2.
The most noise happens to be from the how the player plays the game. The MMR system were built to handle the player performance, they “adapted” to role performance. That change already makes a ton of noise. Because prior to it, the player could actually do something useful to improve the odds or just straight up throw it, with role locks you actually have less options at your disposal and less potential of affect the outcome of the match.
QP matches often are more balanced than Comp counterparts, aside from Smurfing. Because while they have some MMR in works they have more flexibility on it, both on “value” but also because aren’t also limited from SR, like comp does. Surprisingly QP Classic aside of “oddball compositions” often have more balanced matches in terms of individual player skills.
Considering they done the game as casual and later added competitive aspect of it, makes sense why the system can be a bit noisy. The SR tries to follow MMR but takes time and the entrophy to kick in. Having RQ just mess everything up. In a RQ system would be more adequate to consider SR from individual hero not role, but that would need to be selected prior to queue. MMR system wouldn’t work well on RQ system if the player only pick a single hero either.
They created a problem with poorly planned decisions post launch. Having ability to swap on the go were fine with their MMR system. While prior RQ they implemented Competitive matches those matches were inherently less balanced than actually QP matches. While would funnel SR, not always SR would be near to the MMR value, which wasn’t a bigger problem in the past because folks simply could pick any role they wanted if needed, counterbalancing the issue.
With RQ and locking 2-2-2, they improved the learning curve but also made more folks stick to 1 char or one way to play. You can’t balance the player in a role if he only plays with a single hero. That’s why RQ would never have a good MMR matchmaking if you don’t actually select hero prior to it and even if you do you would need to lock the hero swap to rely on that. RQ would work on SR range tho, because could enable the player to select “nearby” SR on their other heroes while locking the outside ones, while they could also do it on MMR, but because the instability of performance of each hero the MMR would be all over the place while SR would be a easy metric based on win/loss.
I agree that mixing both can be problematic, but I would argue that only became really problematic because they restricted the options from the player’s perspective without actually filtering it. They know the performance of the player on each hero and know what players would likely select but don’t use that data to actually make their MMR and SR system be more consistent, because if they used, they would force the player to select what kind of hero he would play and based on the selection would have others within the same range at his disposal during the match.
OW1 was made without both competitive and RQ in mind, they forced it without major revision on it’s behavior and the consequences about locking stuff up without effectively using the “player prefference” instead of the RNG lock. A player who one trick cassidy could be a really bad soldier and neither current SR or MMR would know that until the player actually pick it during the match.
That’s my stance about the actually problem on their system, MMR or SR could work as different stuff or even work together if the game didn’t had much restraints without forcing the user actually direct what kind of behavior he would do, but the moment you put limits on each you need to actually force the user to use what they want and account for that not from some RNG inside of the match which messes the Rank from most One tricks and mains generating a lot of noise in the system.
Comendable initiative about the Open Source project, if you can share more details afterwards I would like to see.