I've as well found Harman to sound best. Rtings.com seem to also use Harman for their compensated graphs. And Spectral Flatness test at audiocheck.net has a description of human ears having 1-3kHz sensitivity increase (the author has a PhD in field and is obviously obsessed with the topic, therefore I guess he's right), so that test also fits somewhat well with Harman graphs. Plus there's the fact that as the volume decreases, we perceive bass and treble decrease as more severe than the decreases of other frequencies (but also vice versa, increased volume yields more severe bass and treble increases), so having these frequencies bumped theoretically allows one to listen at lower volumes and still hear one's preferred signature. You can see how that looks for lots of headphones and IEMs at
Reference Audio Analyzer - it's the second graph of every test report.
Since most graphs are raw (Audiobudget, ThePhonograph, crinacle's, Reference Audio Analyzer), I usually look for either graphs resembling Harman graphs (
like that of Fiio F9 Pro), or ones with the least amount of sharp peaks and drops, so as to ensure that it's possible to fix the signature with as simple an equalizer as possible (Stuff like a 20dB drop at 4kHz,
such as that of KZ ZSR, is a no-no, as far as I'm concerned. How is one to fix that on a smartphone player?).
Hence my opinion is that it's a good idea to make such graphs, thanks, antdroid.