I wouldnt dare investing big money on obsolete model unless there is strong assurance it will run ok for a long time. But i dont mind buying it at small money with slight damage. If i can fix it i saved alot of money, if not, i lose small amount of money. But that is me, ymmv. I think mathematical model of current technology is good enough to 'de embed' system's respond or characteristic. So when attenuators are included in calibration, they are considered in-system elements and de -embedded from final dut result, hence not necessitating of low signal vna output and high input signal level for the purpose mentioned. I read briefly on this 'de embedding' in app notes, agilent/keysight iirc, but not really 100% sure understand since i have no intention at designing a vna, and the math is a bit daunting. Ymmv.