One of the most interesting and controversial aspects of Phil Zimmerman’s PGP was that it avoided any central repositories of information, relying instead on what Phil labeled the “web of trust.” The idea was that Alice “trusts” Bob, and Bob “trusts” Charlie, there’s some transitive trust that you can establish. (I’m going to stop putting trust in quotes, but keep using it in the sense of this web of trust.) These trust relationships are one way and publicly expressed in the form of signatures. That is, Alice indicates her with Bob by means of signing Bob’s key. Bob may choose to sign Alice’s key, but doesn’t have to.
Setting aside all of the security properties of the idea, it creates a fascinating set of published data around social networks. In “Wotsap: Dissecting the Leaf of Trust,” Jörgen Cederlöf writes:
After implementing the group matrices I figured it would be nice to see the group matrix of a much larger group. (If you are a mathematician, the Web of Trust is a large directed graph where the vertices are called “keys” and the edges “signatures”. There are four different types of signatures, just think of them as four colors of the edges. The group matrix is the adjacency matrix of this graph. You probably want to take a look at the FAQ, especially the part about MSD.) I generated a PNG image with the keys sorted in MSD order and expected little more than random noise. When I first saw the result I thought I had done something wrong, but a little bit of thinking revealed that the resulting leaf-like shape was perfectly natural, almost unavoidable.
Thanks to Nicko for pointing out the emergent properties of the web of trust.
 I attempted to quantify that in a message to the cypherpunks list, “reputation credts.” Rafe pointed out that the system could be made to oscillate, and I abandoned it. In retrospect, I’m pretty pleased with what I wrote, even if the system wouldn’t have worked–there’s a lot that’s still applicable to reputation and social networking and identity projects.