I’m not sure how the score for Harman target compliance is calculated on that site, but just looking at the charts it’s pretty obvious it’s flawed. The chart for the Stealth has the only error of note being a couple of dB more low and sub bass than the target, everything else is +-1dB. The one exception is the deep notch around 10kHz, which is unlikely to be audible given the very narrow range of the dip, and may well simply be a measurement artefact.
The Sundara has larger errors over a broader frequency range through the midrange and low treble yet somehow manages to score substantially better.
Subjectively speaking, it is my feeling that the majority of people will find errors in the midrange and treble way more impactful than a couple of dB of extra sub bass or a sharp drop over a very narrow frequency range, especially at 10kHz.