IIRC Bohrok2610 does not use X5R bypass caps for the shift registers, among the other various differences mentioned?
Anyway, seems to me there are maybe three of four various areas of focus to consider which have been mentioned so far:
1. Output stage distortion and or mushy sound
2. Possible Vref effects
3. RTZ generation
4. Shift register bypass layout
Area there any others I missed or that should be added to the list?
If not, my personal responses to the list would be as follows:
1. Go ahead and try the OPA1632 output stage, with or without transformer isolation
2. IME its hard to beat a simple, maybe 3-4 transistor shunt physically close to the dac D/A conversion
3. IME Clocked is going to be more precise sounding
4. IME Ground fill under the shift registers will turn out to work better than ducking bypass through vias
IOW, some of this stiff has been seen by other people already and some results have been obtained. Unfortunately however, no measurements to share.
Also its not like all the circuitry is in an IC, so indirect methods don't have to be used to investigate; this circuitry can be modified and probed fairly easily, all without the cost of a new mask.
Moreover, we should try to avoid any unnecessary "analysis paralysis."
As an aside, the above comments are offered in he spirit of getting the best out of the dac even if it costs substantially more to make. Maybe one goal of the project as Marcel has envisioned it is cost constraint. Keeps it simple and low cost enough so that more people can afford to build one?
Also, one other thing I would suggest would be to get rid of the Amanero USB boards. The latest I2SoverUSB version is IMHO better than the last one, and it doesn't cost more than a legit Amanero. Otherwise for most people USB noise will continue to be an insidious problem.