If I understand the Harman research correctly then the research produced a probability on preference with these factors as unknown variables, in other words the variability of anatomy is priced in to the model. Unless we have the ability to quantify these variables (which end users don’t) then target compliance becomes the only objectively valid variable on likely preference that we have. As you say it’s not the beginning and end of variables that we should care about, but it is the only known one available to us.
I suppose there is still the argument that if the dummy head measurement shows a deviation from the target curve which is covering a wide enough span of frequencies with enough difference in dB that it can be an audible difference, then you could argue that applying that EQ filter to bring it onto the Target Curve would result in an audible difference within your own unit that you're listening to. Of course though, you can make the points you made that you can't guarantee the exact frequency response you're receiving due to unit to unit variation & anatomy differences, and therefore that small change that is being made in the suggested EQ filter might be pushing it in the wrong or right direction - that's applicable to all EQ. But, on average, if the research is good that created the Harman Target and given the measurement you're using for your IEM is a good statistical average, then there is some merit of trying to get as close to the Target Curve as possible when using EQ (as there is no reason to "add ways in which the unit may vary from the Target Curve") - but of course only within the relevance of whether or not you could hear the difference between that filter being activated or deactivated......although I kind of believe in the idea of cumulative gains with various filters in different parts of the frequency range interacting to create a positive audible difference even though one of the changes might not be audible on it's own.
Exactly. It needs to be understood this is all implicitly a probabilistic exercise. When I say people are not listening to the Harman target when they buy the Truthear, in full this means 'you're likely not listening to as close to the Harman target, and therefore the frequency response you're statistically most likely to prefer, as you can'. But that's a mouthful so I'm not going to say that every time. The only variable the end user can control in all this is frequency response via EQ, and having a plethora of headphone measurements and people like Oratory and
@Maiky76 producing EQs that take the frequency response very close to the Harman target, we can all achieve this. From that point of course you can then adjust the EQ filters to preference if you feel the need (which Oratory makes very easy to do with his choice of filters and explanations of their subjective effects in his pdfs). But starting at the Harman target will for the majority likely get them closest to their preference quickest (and that's why Oratory tries to match the target so closely, within reason of course, in order for it to be most likely the EQ is preferred by the majority of users).
It's quite obvious there would be an audible difference between the Truthear's response and the Harman target, just by looking at the EQ function required to bring it to the latter (top-right graph):
Not even considering the sub-bass, as the green 'Target' EQ shows, there's a 2 dB (going down to 2.5 dB at one point) broadband trough covering ~800Hz to 6 kHz, corresponding to the Truthear's peak relative to the Harman target in this region. FR peaks are more audible than troughs, broadband (low Q) is more audible than narrowband (high Q), and this is the region determined by the loudness contours to be where our ears are most sensitive. This is not an insignificant difference.
To illustrate further, the two IEM targets Harman tested, despite only differing by a couple of dB, on average over all tests (not just cherry-picking one test) were given a scaled preference rating by listeners 10 points apart, as can be seen below by the two highest rated data points at 100 (the Harman target) and 90 ('Harman target 2'):
So obviously even small differences in frequency response can produce a non-insignificant change in preference.
The reason for being so meticulous here when looking at frequency responses and variables actually under our control like EQs is precisely because of all the potential inaccuracies in variables we cannot control - you do not want to compound errors. The fact that some repeatedly fail to grasp this, on a science forum, is pretty worrying. An analogy: you're tasked with making a table, but you're given a tape measure that you've been told is only accurate to ± 2mm. Do you say, 'ah that's not very accurate, so I might as well just do it all by eye', or do you make sure you measure every piece of wood to cut as carefully and accurately as you can so you don't introduce any more variance? An even worse response would be to throw up your hands, throw in the towel and not make the table at all. That's just what the Harman Science Deniers (like climate science deniers but the result is less apocalyptic) are doing with their fatalistic FUD. Work with the best tools you have available (the Harman target and EQ), and be as accurate with them as you can so as not to compound any existing inaccuracies those tools may contain.