I am not a specialist with Diffmaker nor have I ever tried it personally (when I wanted to do that years ago it just kept crashing on start-up already), so I have to rely on the available information found in the GS thread.
Diffmaker is said to do a null test between the original file and the newly recorded file. This nulling is done automatically and with sub-sample accuracy, otherwise it woudn't work at all (at 44.1 the chances are like zero to have both files sample-accurate).
Diffmaker's results have people wondering since years what is wrong with it, as its over-exagerated results seem to have an internal weighting that often contradicts other measurement methods.
Maybe the filters of the ADI-2 Pro now easily explain it all. When Diffmaker is doing the nulling, parts of the file are still not nulled - because any analog and digital signal path includes phase shift, at the lower end and the higher end. This phase shift is inaudible, a proven fact.
With the ADI-2 Pro you have phase-linear and minimum phase filters at your disposal, on both AD and DA side freely selectable. Now we see that the most transparent and less damaging filters SD Sharp (which have said phase shift) deliver bad results, while the intransparent, treble reducing and partly aliasing causing linear-phase filters are 'measured' as superior. These results indicate that Diffmaker ist completely centered on the full nulling with a weighting that is unusable in real-world, and will even indicate the opposite of what is true.
New test results for the ADI 2 pro is coming out.
ADM: -1sec, 0,351dB (L), 0,344dB (R)..Corr Depth: 33,6 dB (L), 35,2 dB (R) Difference: -52.4 dBFS (L), -53.5 dBFS (R)
Matlab: -1.000113818 s 0.3521 dB (L), 0.3453 dB (R) Difference: -52.8 dBFS (L), -54.3 dBFS (R)
These are acceptable numbers for the 1600 EURO price.
Old numbers not so good:
ADM: -1sec, 0,366dB (L), 0,357dB (R)..Corr Depth: 33,6 dB (L), 35,3 dB (R) Difference: -43.2 dBFS (L), -44.6 dBFS (R)
Matlab: -1.000125149 s, 0.3521 dB (L), 0.3504 dB (R) Difference: -44.5 dBFS (L), -45.5 dBFS (R)
Everything now indicate that RME has to just tweek their slow filters to get still better results.
The penny hasn't dropped it seems, so I give you a simple example:
The test measures how well one pass of loopBack preserves the "mix". In a mastering situation you may have several of these passes.
Do exactly that. Connect input to output at same ref levels with everything EQ etc turned off, then re-record one file 10 times in a loop. A 10 second sample is enough. Use SD Sharp on AD and DA. I guarantee that you will have a very hard time to hear a difference between the original and the 10 pass file, if you manage that at all.
Please note that intentionally the DA has a bit lower level than the AD, to prevent overloads in loopback applications. The difference is small (around 0.1 dB), but after 10 times it will be more than one dB, so you need to normalize or manually correct the level of the 10 pass file to be able to reliably listen for sound differences.
Now do the same test with the Slow or NOS filters (phase linear). Treble is lost already on the first pass! After 10 passes you get a dull sound, all glitter removed.
Thinking about it that would also be a perfect reason for the developers of this project to abandon it. Nulling can't work with analog/digital conversion and all the filters that are needed and unavoidable.
It also opens the way for another theory: units that show extreme high Diffmaker scores are lucky to have the DA phase shift be compensated by an inverse AD phase shift. They will get super scores for something inaudible, while the real sound quality based on SNR, THD and frequency response is not, or only a little involved in that score.
Hmmm, I spent too much time with this stuff again...