That is true but I don't see how it invalidates my hypothesis. For example: Audyssey measures the SPL at the MLP from mic position 1 and determines that it requires a trim of -5dB to set the (pre-Audyssey) level to reference (assume this is for RF speaker for the purpose of the example). Then Audyssey proceeds with the other measurements and creates filters to correct room issues. In creating the filter set for the RF speaker, it observes a significant peak over a widish frequency range. It cuts this peak by an average of 5dB. This cut reduces the overall energy from the RF speaker, which Audyssey had previously determined to require a trim of -5dB to hit reference. So the additional cut in energy level made by the EQ would now render that speaker below reference. Audyssey knows a) the initial setting (-5dB) and b) the level of the cut required by EQ (5dB), so Audyssey now compensates by setting a trim level of, say, -1dB. Thus the post Audyssey trim is -1dB but if you use the built-in AVR test tones, these bypass Audyssey, so the levels they deliver will be incorrect when compared with the EQ's levels. I think this is what Roger is saying in his reply above too, BICBW. I am only hypothesising - IDK if Audyssey works the way I suggested.
I don't think it's necessary to sweat on this issue too much more Keith. Just let everyone do as batpig suggests: use the internal test tones and call it a day.
BTW, what is a wideband(ish) peak?