So they used twisted pair DC cabling to minimize torque errors due to DC magnetic field (and the torque follows direction of cavity orientation), but.... I'm wondering if there is DC current circulating in the cavity due to rectification from oxides or impurities on the copper surface; i.e. Passive Intermodulation Distortion (disclaimer I'm an RF EE). They said the torque was strongly non-linear with applied power; which is a hint.
DC cavity currents were not listed as a potential error source, but these phenomenon could show up with high power and high Q.
Anyway, I just emailed a couple of the NASA guys; see what they say.