Agree, you don't really want to pre-aggregate your temporal data, or it will effectively only aggregate at each row-time boundary and the value is lower than just keeping the individual calculations.