We will need a way to level match the output versus input.
I can imagine listening real-time and listening for the one with the least artifacts. For those artifacts which do exist, I wonder if those which sound least bad correlate with the better sounding amplifier....
Yes, that's the idea.
You can record the dif + source file using a soundcard and analyse afterwards as well.
If it needs to be done all tube you will need to find tubes + low distortion caps that buffer the inputs and an attenuator that can be set extremely accurately.
It all hangs on how balanced you can get it (probably around 1kHz using a sinewave).
I used a 20 rotation potmeter but I suggest to use 2 potmeters instead.
One for 'course' tuning and a second, small value one, in series with the input for fine adjust.
Then comes the difficult part.
The output signal represents the difference between the 2 inputs so when you have a difference signal it could be that what you hear is something that has been added but also something that may have been 'removed' from the original signal.
That will be an anlysing thing.
A difference in FR will show in an amplitude difference, so will a phase difference.
My questionmarks with traffos vs opamps lie in the fact that opamps can go from DC to MHz, transformers well...
With transformers a question will be how well both inputs balance.
You will have to get a baseline by connecting the same signal and see how that nulls.
Still a nice idea and as Donald mentioned you could aim for the least amount of diff (if you are looking for hifi) or you can listen out for the most pleasant sounding diff file.
Another test that may be interesting to see is the dynamics test I proposed.
Chances are the -70dB signal may not look like the original.
You may have to clip the 0dB signal (diodes in antiparallel) to avoid problems with some scope inputs.