My apologies for not commenting sooner, and for the reported mismatches.
As I posted here War Matchmaking Issue -- Proposed Solutions (Developer response: post 107) - #113 by mhalttu, we have been working on an improvement on the war matchmaking.
In the past two weekends, we have been testing the new model. We do this using a so called AB test where 50% of the alliances get matched using the old model and 50% get matched using the new model. The benefit of this approach is that we can actually see if the new version is producing better results or not.
Some of the mismatches are probably caused by the fact that the pool of alliances is temporarily 50% smaller because it has been split into two groups. The bigger mismatches (when it comes to the difference in the war score) are probably related to the fact that we still show the old war score in the UI, even if the match has been made using the new logic. So in those cases the new logic has made a clearly different decision than the old one would have.
So far, the results have not been very promising. According to our metrics, the new model is indeed performing worse than the old one. Today’s matchmaking was again performed using the old model for 100% of the alliances.
There is still one more thing we want to try with the new model, and we are planning to run one more AB test over the weekend. If the results don’t clearly improve, we are most likely going to stick with the old model. If the results do improve, we will probably want to conduct further testing.
At the same time, we are working on a separate change that should make it more likely that you’ll get a good match even when an AB test is running and the pool of potential alliances is halved.
I want to remind you that the goal of this project is to improve the matchmaking in the long term. However, I do want to apologize for the problems and for not keeping you guys in the loop.