Try to make certain that phase inversion tests are absolutely in mono when trying to cancel compare. Also the two versions must be played absolutely identically in time...
It's a reamp so they are played in time, it's the KPA that changes latency dinamically or something like that... null tests are not valid because of that.