Ralph brings up a very good point about feedback: the underlying theory assumes a distortionless summing point. (The summing point is the comparator input between signal input and the sampled output.) Any distortion introduced at this point of the circuit will be amplified without correction, and there is a real possibility of introducing new, higher-order terms that are not present in the forward path of the physical amplifier. Norman Crowhurst mentions this in passing in his Audio magazine articles in the late Fifties.
Don and I go to some trouble to avoid even local feedback in the Raven and Blackbird ... all cathodes are bypassed in every stage, we do not use the Ultralinear connection in the triode-connected KT88, nor global feedback. Any distortion that is present is the result of first-order effects in the tubes themselves, not anything else. It also mandates selecting tubes that have a minimum of high-order distortion terms in the intended operating range.
This overlaps with the philosophy of some SET amplifier designers, but we prefer to use complementary pairs to cancel most of the distortion without recourse to feedback (local or global). The transformers, if correctly balanced, cancel out most of the distortion before passing the signal to the following stage, thus, each grid-pair is fed a cleaner signal.
This comes in particularly useful for the 300B grids, which can be a difficult nonlinear load for the preceding stage. Unlike RC coupling, when one grid demands current, both phases of the driver pair are available to deliver current to the 300B grid that needs it. This enables seamless and jump-free transition to A2 (positive grid drive) when the signal demands it.