I think Ben wins, but not for the reasons he gives. On the topic of evaluating a network in terms of the value of personal connections: I could use some math/analysis help. I notice that all of my phone services (at the moment) charge two distinct fees: toll charges for connections made and a basic service fee for membership in the network. (Things like VOIP may throw things over towards nothing but the basic service fee --- see below). I think the toll fee value of this network clearly grows N log N *up until* we reach the human capacity limit for connection after which the toll fee value grows just linearly with N. (And, in history, aren't we getting at least pretty close to that point?) Meanwhile, the cost of toll service falls in Moore-ish ways thanks to technology improvements. What about that basic service fee? Is it Metcalfian? N log N? Linear? What? It's a little bit hard to tell from prices because the cost of networks keeps crashing, what with Moore and Moore-like phenomenon and even lesser-order but significant effects. I'm wondering, if we adjust for those technology changes, what the basic service fee prices look like in history. I notice that telephony providers keep moving closer and closer to nothing but the basic service fee ("hello VOIP"). That is consistent with toll fees (personal connection fees) maxing out to linear growth in value while marginal cost of prodution is super-linearly crashing thanks to technology/infrastructure build-up in a competitive provider environment. What with competition, and efficient markets, seemingly robust and universal demand, and yadda yadda, the price of basic service is going to converge on the marginal cost of production of a new endpoint. What we need, therefore, is a robust lower bound on the cost of a network that connects N users consuming their human-capacity upper-bound of connectivity. There's a search problem involved in such a network. I've dialed a number -- you find the route to complete my call. To be able to solve that search problem in log N, there's a sort problem involved: N log N. And, finally, there's a wiring problem involved. "If we all went to the same New York deli on the same night and ordered blintzes, there'd be chaos" [W. Allen] but there are slashdot-like effects (e.g,. New York phones immediately after the 9/11 incidents). I'm a softy on the graph theory to be able to answer the wiring question off the top of my head but I'll confidently conjecture N log N. I think Ben might be right about N log N for phones and the internet and such -- but I think a better way to see this both a priori and empirically is to look at the cost of the network and how it relates to price (all in the context of seemingly maximal demand -- to the limit of human capacity). I originally started off arguing that persistence -- an accumulation of edits -- could make a hypertext-qua-network Metcalfe-ian. I have to think about that some more. Perhaps it's N log^2 N or some such. Makes no nevermind to Stephen, of course. -t