OK, now we're getting somewhere. It's nice that he responded so fast.
That's not the whole story though. If you're doing high frequency corrections, you don't want to try and fix sharp dips and peaks that are highly mic position dependent. The spatial average keeps you from doing that (has a smoothing effect). Here's a single MQX measurement compared with the aggregate (~25 mic positions or so) with all smoothing turned off:
View attachment 228967
View attachment 228968
You can see the dramatic smoothing effect a good spatial average gets you (and a big advantage MQX has by allowing so many measurements). With the lowest level of smoothing applied:
View attachment 228969
View attachment 228970
You can see they do apply smoothing before calculating the correction filter--which is a good thing, you certainly don't want to try and correct all that hash:
View attachment 228971
It looks like their filter is pretty much based on their middle smoothing setting, 1/12 looks about right:
View attachment 228972
And as we've now confirmed, it is based on the steady state response. Here's a comparison of the MQX aggregate to a MMM using a Umik1 for that speaker at the listening position:
View attachment 228974
The deviations at low frequencies are because I modified the speaker which reduced the sensitivity of the woofer at low frequencies. The deviations at high frequencies track well with the differences in calibration between the two mics (which I have measured and can compensate for in the target curve, but didn't for this particular file). Again I'll mention the number of measurements MultEQ-X allows you to take really helps you get a much better, repeatable spatial average.
OK, now that that's out of the way....
It's not that one measurement is "correct" and the other is not, or that one is "true" and the other is not. You're measuring different things. Both are correct. And the Harman curve is based upon the red line, not the green one. If you correct a windowed response to that curve, you aren't really using that curve.
If you read Toole's book, all the papers, all the research from Harman, every time they refer to an "in room response" they are referring to the steady state response--the red line (unless specified otherwise). The Predicted In Room Response from the spin measurements is predicting the steady state response in a typical room. Good anechoic data can not really be obtained from your listening position in a room. I understand there are some newer/high end room correction programs that do some windowing, but I don't know enough about them to comment on how well they do at that. Historically, pretty much all room correction systems have used steady state measurements, and that's what all the various room curves people pass around are based upon.
Generally, that's the way most like to do it. But it's an oversimplification to say you can't do that with RC that uses steady state measurements. It takes some effort, knowledge and anechoic data from the speaker but it can be used for that purpose (it's just a lot more work that buying a JBL SDP-75 which allows you to load "Anechoic EQ"--which from those I've seen amounts to EQing the listening window flat) before you even start your in-room measurements.
Sorry for getting so off-topic, Matt. But I do think it's a useful discussion.