When I tried the 598's the look made me feel like I had just stolen them from some dude's luxury yacht. They sounded a bit more rich/lush compared to my AKG's but I felt they had a clearer and more detailed sound at the same time and weren't sibilant in nature either. They were also super comfortable.

I had a very positive experience and am wondering if the 558's would be just as good to my ears.

Also the sweep results (same 80 mVrms, using 2i2 and HD558 as load).

This seems to show some D/A or A/D related problems, since there is a high amount of aliasing/imaging, and apparently a 22 kHz lowpass filter, even though both the playback and recording sample rate was 96 kHz. Is it possible that the signal was resampled by Windows ?

Yes, it still seems to be resampled at some point to 44.1 kHz.

Alright... how about this one...

Is it a digital loopback ? It does not have any distortion and the frequency response is perfectly flat below about 16 kHz, but there is still a brick wall lowpass filter at ~20 kHz, and noise shaping. I think I will use the first 150 mVrms file for now, but I have also created a 44.1/16 version of the test file, maybe if it is already 44.1 kHz, then it will not be resampled by the operating system or drivers.

Here is the sweep recording of the 44.1/16 version through the headphones SPL (headphone characterization rig)

Unfortunately, it does not seem to have gone through any analog hardware (the distortion and frequency response is again too good for that), but it is not a bit perfect copy of my sample either. I think with the new settings you use since "sweep_HD558_150mVrms_2.flac", the recording is just a digital loopback through the audio driver.

Edit: I have a few graphs from sweep_HD558_150mVrms.flac:
Frequency response (I could have set a higher vertical resolution; the high frequency ripple and roll-off shows that there was probably some low quality sample rate conversion applied somewhere)
Crosstalk
Distortion vs. frequency (only D2 and D3 are included, anything higher mostly just adds noise)
Distortion vs. level (0 dBr corresponds to 0 dBFS in the original sample file; the 15 kHz result is not shown because of the playback/recording problems)

From the same sweeps it would also have been possible to generate these plots:
- THD vs. frequency with only one channel driven (it did not really differ from the both channels case)
- THD vs. frequency into the other channel when only one is driven (this did not happen in any significant amount)
- linearity (basically, gain vs. input level - it would be somewhat interesting with high distortion)
- distortion plots for specific harmonics