• Do you have an SQL queries to substantiate that your keys are unique(rows are not duplicated)?

    Could you try the SQL that is used during cube processing?