Originally Posted by FractalAudio
The key word is "claims". We know that the core modeling is based on the Zenterra, which was (reportedly) EQ->waveshaper->EQ, so I find the claims specious.
The only other possibility, given the black-box approach, is EQ->dynamic waveshaper->EQ. The dynamic waveshaper could be of the Moog sort or a waveshaper with memory using RK or Euler methods (state-space). Those two methods are essentially equal but differ in their implementation. FWIW, these methods are excellent and can replicate a tube with very good realism.
This method is completely suitable for automation if you treat the EQ as IRs rather than trying to curve fit recursive filters. So the actual model becomes IR->waveshaper->IR. Obtaining the two IRs is almost trivial and you can do it yourself with readily available software tools.
As a point of fact, I consider this to be a very good method for automatic "modeling" and would probably use a similar technique.