I think it's proven enough. I spend all day gathering and staring at data that tells me stuff that I can't prove, I just have to take my best guess or ask for gobs of money to be thrown at the problem to get more. It's called engineering.
Every war has been a data point. I can remember the outliers and the a-ha moments. I can remember things that shaped the way I viewed the formulas.
The whole base rank doesn't hold up on matchmake rank is so absurd that it isn't really worth considering. Why would SC design two systems? What possible reason would they have to do the extra coding and why would they set it up in such a way to rank bases 1 to n on one criteria and then rank clans 1 to n on a completely different set of criteria. This is just madness to think that someone would take the time to do this. Nobody, nobody, has ever suggested such a thing anywhere but here. I am not really seriously considering this dual matchmaking pattern at all.
Here's what I think is a lock down etched in stone fact
- Offensive firepower has zero bearing at all on matchmaking. None.
- Proof: Viewing shuke/yoga war rating versus TH7 with TH7 troops.
- It seems that certain defensive structures also have no bearing on matchmaking. These seem to be but may not be limited to: (Anti Air, Teslas, Archer, Cannon, Mortars)
- Proof: Watching how low TH stack up versus higher ones as well as our own one cannons
- Three structures independently seem to drive 80% or more of the score (Infernos, Xbow, Wizard).
- Proof: Yoga vs. ZZang. 4 tesla levels and 14+ cannon levels and full traps could not out weigh one Wizard tower at Level 1.
- Proof: Basically any war you can look at those three towers and get within +/- 2 of the base rank looking exclusively at those three tower types.
- Heroes count more than you think, but not enough. Walls don't count at all. Bombs and traps count a little.
- Roughly 10 hero levels is enough to overcome inferno/xbow/wiz upgrades
From that point forward I do believe, and have since Reddit Phoenix proposed it, that they apply a geometric mean across the clan. This, is likely what Clayton and you are misinterpreting as a "second algorithm." I think there is enough to go on to say that they take a snapshot in groups of 5 and apply a GM. GM criteria can easily look like a completely separate infrastructure applied on a one shot. I sort of came to this realization when I would see what the bottom 15 and top 15 would shape up like in different wars, especially when we got some one cannon and buttworms in. When Phoenix proposed this I couldn't believe I had missed it for so long.
I would be stunned if it wasn't a 5 base GM applied to the base rankings they show. There is minimal overhead for this and their back end overhead would be minimized to produce a speedy-ish match. They restrict the war sizes to groups of 5 so it's natural to a 5xGM for something like this.
tldr - stop building or upgrading infernos and xbows. Trust me.
u
