Good point, that nonlinear processes generate frequencies not present in the input. One of the biggest challenges in making analog circuits with good-sounding

Good point, that nonlinear processes generate frequencies not present in the input. One of the biggest challenges in making analog circuits with good-sounding clipping distortion is designing proper lowpass and bandpass characteristics into the coupling between stages to roll off undesired artifacts. The advantage in being able to actually listen to a real-time simulation modeled at the level of resistors, capacitors, inductors, and tubes or transistors is that one would be able to immediately hear the effect of component value changes, rather than just see them on the transient or frequency response plots. Modeling the system at the level of DSP blocks would be more computationally efficient, but then one is left to derive the equivalent analog circuit. It's not rocket science, but it does add a level of indirection. In any case, being able to simulate 5 triode sections and the linear filters between them in real time at a 48 kHz sampling on a modern CPU very well might be good enough for rock and roll. My use case is a design tool, not to obtain a substitute for a physical device.