The final solution to CW matchmaking
Google uses it. Netflix and amazon use it to recommend shows and products. Every research area from drug design to linguistics is using it. Why hasn't the gaming community embrace it? I'm talking about deep learning. Over the years, SC has acquired millions if not billions of data points from each war: what troop levels and building defense levels each player had in each war. These could be used to train a deep neural network, which would predict exactly one thing: given a player's set of offensive troop capabilities, how many stars would they obtain on a given base.
Matchmaking is really easy then. Just predict for each player's profile (troop/hero/spell levels/camp size) how many stars they'd get on any other player (this will not be 0,1,2,3, it'll be something like 0.47 or 1.68 to reflect uncertainty). Now use the optimal hungarian algorithm for scheduling tasks to decide how to do the optimal matchups for the first and second attacks (bonus! this would allow them to finally make the recommended base actually work!). This is quite easy and fast computationally (much faster than animation algorithms that are used all over in CoC), wouldn't add a bit to war search time. The MM algorithm would then only match up wars where each side would have an equal number of expected stars, within some tolerance.
This neatly and perfectly handles the case of those engineered bases; the deep learning algorithm would quickly see that the offense could easily trump a maxed TH of similar level. From a software engineering point of view, this is the only right way to address the current problems that exist in match-making.
The end result is no further reward for engineered bases. They'd just end up matched against other engineered bases. No more fluffy-bunny throws the war. Only close wars that are determined by skill and player participation.
For the record, I'm a deep learning practitioner and would be happy to serve as a consultant if SC wants to finally fix the problem by going this route.