This document presents PDDP, a practical system for performing statistical queries over distributed private user data while preserving user privacy. PDDP uses differential privacy to add noise to query results. It limits the ability of malicious users to distort results by splitting answers into binary buckets. The proxy adds blind noise in the form of encrypted coins contributed by clients to achieve differential privacy. PDDP scales to millions of users and tolerates client churn while providing strong privacy guarantees and bounding result distortion.