Designing the efficient information-processing organization

We model a hierarchy consisting of a fixed number of heterogeneous agents, each of which reports (possibly distorted) information to its direct superior who can detect and correct the misreport with some probability. Two problems are simultaneously considered: designing the efficient hierarchy structure and configuring the agents who are heterogeneous in their trustworthiness. We develop the Unimprovability Principle, which is an analogy of Bellman Principle. By applying the principle, we show that efficiency requires only the least reliable agent in any rank to have subordinates.