Providing for record filtering in distributed dynamic clustering algorithms for coordinated multipoint (CoMP) wireless communication is described herein. By way of example, strategy selection records distributed as part of a belief propagation network are pruned at recipient nodes, thereby reducing processing overhead for dynamic clustering. As a result, cooperative policies can be determined with greater efficiency, and with greater relevance to local clusters of cooperating base stations. In some aspects, record pruning can comprise identifying and discarding redundant or incompatible sets of policy decisions. In at least one aspect, a number of evaluated records can be capped based on relevance, while preserving deployment-wide applicability of the belief propagation network. Accordingly, dynamic distributed CoMP decisions are optimized on a deployment-wide scale that more efficiently converges to maximum utility solutions.