It seems to be accepted that there can't be done much about PGA'ing players. Actually it can and it would be quite easy.

If you plot a network graph of all games played, attending players, and their individual results, the graph will include detailed information about who PGA'd and who did not. PGA'ing players will show a very high correlation between a certain set of peer players attending and average scores reached.

Kdice is all about randomness (random maps, random setup, random throws). The current ranking system is based on the idea that excellent players can sustain better scores over time than just lucky ones. Randomness is fair as long as its the same randomness for everybody.

It's not random anymore when certain clans are allowed to play at an lower risk of being attacked because of secret and unfair PGAs. The average variance (entropy, randomness, whatever) of people you are able to maintain a strong game against should be included in the rank! Somebody who is able to maintain 30% 1st & 2nd place against a truly random set of opponents is a much stronger player than one who is only able to reach it by unfair methods.

If implemented in a wrong way the proposed method would introduce an asocial factor into Kdice. Instead of being happy to meet a friend online you would have to fear about your rank if you play against him too often. This side effect can be completly avoided if we distinguishe between dices won and dices lost against peers. A freuquent peer you win AND lose against, as you would do against anybody else, would automatically be counted neutral without the need for further modifications.

Spontaneous, tactical truces would also be unaffected as they don't alter correlation. It only makes a difference if you are cheatingly trucing the same people (or your own secondary fake accounts) over and over.