Here’s another update.
We tried something completely different since my last post. This time we copied the alliance specific war history to every member of the alliance. In other words, we ran the matchmaking based on the old data (the war history of the alliance) but using the new implementation that reads it from the player history. After this initial seeding, we kept making matches using the new player-specific history for the past two wars.
So far the results are encouraging. Our metrics show that the matchmaking quality has been comparable to the old implementation. As a reminder: The benefit of the new implementation is that it solves the issue where multiple experienced players start a new alliance and get too weak opponents.
We’ll monitor the situation to make sure that the results don’t degrade over time.