Many inter-agent interactions --- on whatever services-substrate we
end up with --- will take part between parties previously unknown to
each other. How do these services find each other, and how do we
best
connect them? This is the matchmaking problem, and has attracted
attention for as long as multi-agent systems have existed. Most
matchmaking research is focused on semantically matching requests with
offers, whether by measuring string similarity or doing description
logic subsumption. We take a different tack --- examining
historical
performance data --- to find groups that collaborate successfully.