This is a brief discussion of the reason why people who have a low level of error-covariance over a broad range of questions will also be compatible.
There are two types of error-covariance which I have discussed. One is that among an odd number of people. I have written most often about a set of three people, and described that as if it was a tribunal, but used secret ballots. The basic argument was that such a group could be assembled so that their error tendencies cancelled out. Such a group could get almost all answers right, whereas a badly chosen group might get almost all answers right.
The key question to be asked is just this. What kind of a social relationship would we expect these people to have?
Before answering that, let us look at a related issue. I have written of estimation by a pair of individuals, where the correct answer is almost always between the answers they give. With the right choice of people, this can be almost guaranteed. What kind of social relationship would we expect these two people to have.
I say that the group of three people would be most successful making decisions, even if they did so, not with secret ballots, but in person. If that is true, then they would be very good at making decisions about social activities, do this, do that, etc. A set of three people making such decisions sounds to me like a set of three friends, and the fact that when then do make social decisions the results are uniformly good suggests that as friends, they would be mutually compatible.
The other case reminds me of a married couple. So often what a couple needs to do is estimate, how much to spend on this or that, how long to wait before having children, how far from their jobs to live, and so on. True, there are yes-no questions, but they may be dealt with by getting advice from the best of many possible third person moderators. Though that may be needed, the basic ability to do estimations suggests to me that they would be compatible.
This is clearly a matter for empirical study, but I am quite sure that the social relationships between pairs or small groups of people bound together by minimized error-covariance will be strong ones.