In relation to the Harman target scoring ...
It gets smoothed to 12PPO.
Now that I think about it, I fear that a simple smoothing per octave model is not neccessarily accurate for a headphone FR scoring. The practical issue is that the Harman target does not make assumptions on the particular ear geometry. It was meant as a
general guideline that deliberately leaves out high Q resonances caused by phase cancellations in the short-wavelength treble region. Those resonances are natural components of the individual human HRTF though and also present - at least exemplarily - when measuring with a 45CA that simulates a human pinna and ear canal.
Take the notch between 8 kHz and 10 kHz in the concha area as an example. This phenomenon can be seen on almost any headphone measurement
@amirm has published so far. I am pretty sure that we will be see much more variance once we have gathered a larger group of different models by different manufacturers, though. The fact of the matter is that not every design excites ear resonances the same way. Also, placement can play a huge role for activating, deactivating and shifting such interactions.
Now, if we want to come up with a reliable scoring (as reliable as headphone measurements can be) then we would either need much more data with as many seating positions as possible and then take the average of those OR we would have to apply some filtering that ignores all the high Q peaks and notches and just leaves the general trend of the curve.