Even before the update, we had no idea how TrueSkill2 might have been tweaked since H5. Now, the update notes that the matchmaking/MMR will “take into account learnings from live data” and all sorts of other vague statements that assert changes without providing any detail.
I suppose it is a mixed bag telling players exactly how it works although it sure was nice in H5. But, can we now acknowledge that the answers to everything aren’t in the TS2 paper? I’m pretty sure we officially have no idea how MMR works at this point.
The TS2 paper lays out that they are fitting a statistical model based on a large number of metrics from past games to determine the main factors that determine win/loss probabilities. That then gets condensed down to a number which is used to determine rank.
When the underlying data used to build the model changes of course the results will change. During a season only incremental updates to the rank are performed instead of recalculating everything across the whole player base. In all likelihood the original model was fitted based on the test flights and internal play test data.
It’s of course possible that they went in there and manually encouraged the system to give more weight to certain metrics. But aside from that I believe that they’re mainly running the seasonal model update early (which also makes sense given that we’re running a longer season than typically planned).
All in all TS2 itself can’t tell you which factors determine CSR gain/loss. Only looking at the model resulting from TS2 can and this can change over time based on player behavior (so far it seems to have heavily weighted KDA in it’s results - this might change, given that people noticed it and started to favor slaying over playing the objective and fitting a new model would in theory take this into account (number of slays is no longer as indicative of win/loss in objective modes).