Maybe the parameters of the Richard Clark amp challenge can fill in some blanks. He required amps being compared to have the same power without clipping, no EQ or other response differences, 30 db channel seperation over the whole audio band, and no more than 2% THD. You had to get 12 of 12 correct twice to collect the money. Yet over more than 2000 attempts no one even got enough right to meet a p-value of 5% which would require 17 out of 24. These tests are all done with music.
If anything, the challenge shows that differences appear in 1.) the boundary conditions, and 2.) as a result of interaction with surrounding equipment.
Don has shown us that amps can respond differently into very different loads, and we knew that already. That's an example of a difference in a surrounding condition. Another example is when the gain of an amp is low enough that the driving line-stage amp is driven to nonlinearity.
And boundary conditions. I think clipping happens a lot more often with real music than people suppose, particularly with the lower-sensitivity speakers used in home systems, and particularly for those who play the music loudly.
So, the challenge sounds to be a bit like "show me you can tell when the amps are different if I constrain them to the narrow solution space where they are most likely to be the same."
Granted, the challenge was meant to undermine the notion that amps have a signature sound even within those limiting constraints those boundary conditions. Surely some do. As Bruno Putzeys put it (my paraphrase from memory): "Pass and I have different objectives. He designs amps to sound good; I design them to have no sound at all." But most don't, or when they do, most can't detect it if all they use is their ears and not their eyes. All that is theory, though. What isn't is that in the real world, amps are used differently than in those tests. They are played louder, or into different loads, and show the results of pushing against their limits. How they respond to those limits explains much of the difference in amps, it seems to me. I'm reminded of the Carver challenge, which can be explained by the high output impedance of the "reference" amp compared to the low output impedance of the challenger. The suggestion is simple (though speculation reported only third-hand that I've seen): Carver added resistance to the output to increase the output impedance so that it was altered by the speaker in the same way as the reference amp.
There was another amp challenge linked here that I cannot now recall, where they determined that the amps being compared were clipping up to 1% of the time. I suspect that happens more than people realize, and it would take either putting the output on a scope, or recording the output of the speaker and analyzing the waveform for clipping (which is probably the only thing one would get out of making that recording in support of the comparison).
When Amir tests headphone amps, he's testing for sufficient power, it seems to me. It's not like clipping is a switch--even with a steady-state test signal, a bit of clipping has a bit of an effect--maybe not even audible as such. A lot of headphone amps are driven to their limits, it seems to me. I know there are times when I really crank up my Atom, depending on the sensitivity of the headphones. But the SINAD is measured below clipping, and won't catch that.
Rick "responding before reading the whole thread--always a risk" Denney