I was under the assumption that only duplicate records go into UV table while other error records go into ET table; but I see records having Numeric Overflow go into UV table. Could someone explain the criteria for records to go into ET or UV table?
Perhaps you mixed up the ordering of the error tables in the load job ? the ET tables comes first and then the UT... remember that names as such has little importance to the load utility, it's the order that you mention in the load job that matters.
The first (ET) error table is primarily for errors detected during the Acquisition phase.
The second (UV) error table is for most errors detected during the Application phase. In general, this would include Uniqueness or other Constraint violations, or overflow on anything other than PI fields.
The exception is that errors not detected until Application phase but which make it impossible to build the PI (necessary to log the error in the second table) will go in the first table after all. This would include numeric overflow in PI columns.