That isn't how SBMM works. If you play a team with higher ELO, the amount of ELO you win/lose changes based on the skill differential. The eventual averaging is

That isn't how SBMM works. If you play a team with higher ELO, the amount of ELO you win/lose changes based on the skill differential. The eventual averaging is not how the algo finds your skill level. If you play a worse team than you are expected to win with X probability, that is what you are judged against. Also, just to be clear, no implementation is this simplistic. Most models are Bayesian and so incorporate uncertainty around the estimate of skill. And most also include your individual stats (this is game-dependent but I believe Overwatch has a class/map-based ELO that you are rated against...as an example). I agree with your general point though. Most SBMM systems do not feel close games every time. Usually, someone is smurfing, someone is throwing, someone won't come on comms, etc. (yes, I play Overwatch and I am still angry about it).