Resource management in data processing is an open problem in a distributed system. The brain faces this problem when allocating resources to different processing stages. Filter graphs are an abstraction that allows us to consider solutions to this problem. We propose a solution using a transactional framework analogous to prediction markets, which would yield a system that is able to efficiently combine the activity of a distributed set of filter graphs to solve arbitrary inference problems. To demonstrate that we have found this solution we will focus on problems that form naturally distributed online inference problems. These are automated code debugging, air quality monitoring and the RoboCup football challenge.