I'm trying to verify correct operation of a programmable attenuator with LSB attenuation of 0.5 dB. Previous analysis of the SA power uncertainty included SA detector accuracy, power meter uncertainty, and S21 transmission tracking error. In my uncertainty calculation I RSS uncertainty from 2 measurements (no attenuation and tested attenuation). This results in an uncertainty ratio far from meeting a required 4:1 ratio for many attenuator bits. Does any of the uncertainty cancel out in a delta power measurement? Should I be able to reliably resolve a 0.5 dB difference in power?