Overall its really a pleasant listen, there's nothing its doing which initially draws attention away from the music. It has a few weaknesses though, the main one being soggy (my wife says 'one note') bass. Ambience cues are confused, on a YT recording made in a cathedral I hear plenty of ambience, too much in fact because its not anchored in the architectural features of the building rather it 'sloshes around'. Dynamics are reduced in comparison with my Celibidache NOS design, this IME is a result of the lack of filtering of images.
I'm going to have a play with the PSU part - running a complementary pair of transistors without bias in order to generate a 0V is a bit dodgy in my estimation. I reckon a lower noise supply will help it.