Hello, jb!
Actually, the Harman target is supposed to "mimic" the frequency response of flat speakers in a well treated room. It doesn't include the crossfeed aspect of speakers nor any of this, just the frequency response aspect. The preference part is kind of the icing on the cake, where listeners further "sculpt" the sound of the target curve. The treble preference didn't vary much from the original target, the main differences were in the bass region (105Hz and below, a low shelf).
Now, don't get me wrong, the Harman target isn't the definitive target curve, mainly because the GRAS rig doesn't include the ear cannal's impact on frequency response.
For contextualization on the above mentioned, every measuring rig have their own impact on the final measured frequency response. The GRAS rig was the one used to make the original Harman target, therefore you want to use measurements done with that rig to get results that properly translate with the target. The Oratory1990 measurements are the most well known and are mainly the ones available in Autoeq.
Currently, there is a new rig available made by Bruel & Kjaer called "5128" that includes the ear cannal's influence in sound. There isn't a equivalent to the Harman target for that rig yet, but there is an attempt that tries to mimic the sound of it. If you have headphones that were measured with that rig, I'd suggest you to try the EQ profiles. It's probably the best approximation to a flat sound nowadays, at least that I know of.
Here's a video that explains that attempt: