I entered the discussion about computer audio expecting to learn something from your long experience. Unhappily the only advice could be summarized in a few words: get an USB asynchronous interface, a DAC with good measurements, forget about your preferences and the small differences, enjoy the music and perhaps get a boat. As I am living far from navigable waters, I do not feel tempted to follow any of your friendly advices.
There is a bit more to it than that but not a whole lot Micro
. With respect to a digital transport, there are three things that engineering tells us:
1. Digital PCM samples. Unless something is broken, these get through reliably whether you the world's worst transport or best. Should anything happen to them, they would easily and readily show up in measurements. So there is no reason to worry about improvements especially in the class of products we are talking about.
2. Digital sample timing. A lot can vary these from the source, to cable to receiver. There are also mitigation techniques in the DAC for upstream issues. What makes this situation the most difficult, is the DAC attempting to guess at the sample timing on every input pulse. And using a transport like S/PDIF that embeds such clock in the data stream itself. High-performance implementations (in DACs) can deal with this very effectively. In my WSR magazine article I showed for example how even over the "dirty HDMI" connection where every AVR did poorly, the Mark Levinson 503 processor passed this test nearly with flying colors. I say nearly because one could do even better with good async USB but we do know how to do extract the clock well. It is just that it takes a lot of engineering skill and material cost. It involves what is called "mix-signal design" meaning you need to know both analog and digital design which many engineers are not trained or experienced in.
The easy way out of the above, with a computer server or proprietary connection to a DAC, is to reverse the roles. Put the DAC in charge of the clock, have it generate a super stable one and then tell the upstream source to track it. An async USB interface accomplishes this by generating a local clock and having the PC (the USB source) track it. What is left then is the accuracy of local clock in the async interface. In the case of external async USB, you still have a bit of consideration left with respect to the S/PDIF output. With internal ones this can be avoided.
3. Noise. PCs are very noisy devices. That noise can easily bleed into the DAC. On my "work" PC, I once hooked an HP Fax/Printer over USB. The audio was provided by a "high-end" Creative Labs "gold" card or some such nonsense with claimed 100 db or so signal to noise ratio. Despite such lofty numbers, if I attempted to send a fax or print something, massive amount of noise would bleed into my internal sound card's output. By the same token, we want isolation between the USB and the clock side of the async USB. Isolation is a generic term however. You can have a little or a lot of it. Here is a version of it on the Berkeley Alpha:
On the right is the USB interface. Notice that large vertical component. Look to the right and that is the "digital" (computer) interface. You see a very clean physical isolation beyond any electrical ones. Compare that to some async interfaces that are just a bulge in a USB cable. Clearly there is not as much room there to isolate things. Here is an example of one that is board level but clearly no attempt at physical isolation (Musical Fidelity V-Link II):
Good news is that we can measure #2 and #3. Bad news is that we can't necessarily show what is audible or not. In the case of jitter, we have models for simple sinusoidal tones. I have used that to analyze the jitter profiles in the products that I tested. Noise from PC is unpredictable. It can be any and all things. This is why it is important to have good isolation. How good, I can't tell of course because that depends on the nature of the noise at the source and how good the DAC is at rejecting it.
You say I have a lot of experience. I don't in how many devices I have tested. My experience however is in analyzing such devices from design, theory and psychoacoustics of noise and distortion. That analysis is stipulated above. If you look for those factors, and have them minimized, then you are golden. How many decimal places you go past inaudibility, is a choice you have to make.
Now if we put aside everything we know about engineering, and wish to have factors and measurements that are unknown, then I am out of commission. I can't engage in discussion of topics such as preference with respect to transports for example. You don't control the jitter and noise so I don't know how one would optimize for them based on preference. You can control which speaker you buy so you can optimize there. But not in digital transports where we have not even converted the digital samples to analog i don't know how preference is established.