This talk will address the distributed information of a noisy 
binary source through a parallel relay network. The relays are 
only allowed to forward partial information about their observation 
to a fusion point, which makes an optimal decision. We show that
in a two relay system, if one relay is constrained to an output 
alphabet size of $M$, then no more than $M+1$ alphabet symbols 
are required from the other relay for optimal operation.  
Several counter-intuitive examples, a fundamental framework for
analysis using the error curve, and the structure of the optimal 
solution for the general case will be presented.