Smart grid support for demand response provides strategies for an electricity service provider to shed loads during peak usage periods with minimal consumer inconvenience. Direct load control is a strategy for doing this in which consumers enroll appliances such as electric water heaters, air conditioners, and battery vehicles in a program to respond to load shed instructions in exchange for a discount on electricity prices or other incentive. The effectiveness of direct control depends on the ability of the provider to verify that appliances respond to load shed instructions. A technique called non-intrusive load monitoring, in which electric power meter readings are used to identify loads generated by specific appliances, provides a practical strategy for load shed verification in residences. Non-intrusive load shed verification (NLSV) promises to greatly simplify trust assumptions required for the deployment of direct control. Implementing NLSV on resource constrained smart meters is problematic. This paper describes and evaluates a distributed non-intrusive load monitoring algorithm that is split between a capable backend system and a typical smart meter in the field.