about 0.21dB difference.
Just a tad too much when comparing directly but you won't be switching between outputs directly.
Next attempt... try to figure out which DAC is which in a blind comparison (have someone randomly plug in 1 of the 2 DACs)
The helper can also try to bait you by unplugging DAC A and plugging it in back again, even a few times.
Make 20 'attempts' with random DAC and figure out by ear only which is which. The 0.2dB is no issue when time has passed between attempts.
You can spread this test over several days. It is fatiguing after some time.
Good luck telling them apart.
First off I enjoyed reading this topic. It's very cool to see the effort the topic starter is making to validate his hypothesis.
I would recommend (in the name of science) to really randomize which DAC is plugged in. Your helper making decisions will influence the result. Unconsciously you could start guessing what the helper is plugging in ("he wouldn't pick A 7 times in a row..."). The helper also will (unconsciously) try to make it harder to guess for you (let's do A and then A again) which in result can make it easier to guess (humans are really bad at randomness).
You want to have a random allocation with no relationship to the prior assignments. A
real double-blind test. This is not difficult if you add a few steps and use a coin and dice.
First of all, these steps need to be done without you seeing any of it.
With that out of the way, your helper will start marking one DAC A and one B.
After the first step, we need to select a sequence to split the 1-6 range of numbers later on in the process. Your helper can split range (1-3 / 3-6) or even/uneven. Just let a coin flip settle on which sequence will be used. Write down the split (range vs even/uneven).
After this step your helper will throw the dice 20 times (without you seeing it), writing down each result.
Now he ends up with a list of 20 numbers ranging 1-6.
Now we need to do one last coin flip. Your helper will use this coin flip to find how the 1-6 numbers correspond with A and B. If the initial coin flip decided on the range split let the coin decide which range (1-3 or 3-6) is A and which is B, the same goes for even/uneven. Use the split-sequence to transform the 20 numbers on the list to A or B.
Now you end up with a truly random sequence and you can start testing!
p.s. keep in mind that you have to score more than 14 out of 20 right to come to a statistically valid conclusion. p-value would be .010547 so the result is significant at p < .05.
* as mentioned below, this is not double-blind but still a well-controlled blind test.