It sounds like a problem due to an attenuator in the output chain. Since you said that in a given power range, exactly 10 dB are lost, in another signal is lost, then it appears again but at a fixed level, seems to me that the attenuator control fails.
I think you should investigate first the control logic that drives the attenuator (or attenuators, if they are two or more).
Just my feeling: I don't know the block diagram of the instrument, so I cannot be precise.