Any details of the magic-gold-dust-soundengine-technology are very sparse, all I could find what this presentation by 'Bent' who apparently is the founder / main tech dude over at Hegel. Funny that he is completely missing from the Hegel site (
https://www.hegel.com/en/about/people-of-hegel -- all people on there are sales dudes, but won't go too deep into the conspiracy rabbit hole over that fact).
Anyway, he speaks about the 'analog computer' and a 'threshold detector'. AFAIK he basically says the error correction feed-forward signal is not present at all times. This threshold detector decides when the distortion is audible and corrects only when it deems it so. This is certainly different (or more 'dynamic' as he puts it) compared to conventional global feedback designs (where the feedback is always present).
Can it be that simple sine-sweeps used in the standard measurements (for whatever reason) doesn't trigger this threshold detector? Thus what we are seeing on measurements are the 'untouched' signal path trough Hegel's class A/B topology without any error correction applied at all? He even says that 'most of the time' no correction is applied.
Could this explain the mediocre-at-best results? I.e. what we are seeing is a standard A/B design but without Hegel's claim-to-fame-magic-dust applied?