OK so I did a few more and the dip is always there. Here’s the same as above but D.Mode OFF (upped the volume to 54/60 to stay away from the noise floor but still be at least -3 dB of max):
Then I went low-gain but also switched to BAL-out to cover more ground. Unfortunately there’s something about my cabling for this scenario where using a headphone cable + 2 types of adapters at both ends and using only 1 channel into the line-in looks like it creates 30 whole-ass dB of extra noise on BAL, to the point that it’s even visible on all the FR plots as I go back up in power levels. Same happens to the AkLiam PD4 BAL-out, so I would ignore that and just focus on everything else, i.e. the dip is still there, just that the overall shape seems to change toward the brighter somehow. That could still be from the extra noise being shaped that way, not sure. (In A/B listening I still hear it as veiled + some weird sparkle above the veil area, so it’s not like it makes it seem any flatter.)
And lastly I reflashed the newer firmware v2.00 that fixes some overheating and auto-shutdown scenarios(?) and adds multiple button modes so you can have track switching on the volume buttons. The dip seems to be a tad lower this time and also the 20k rolloff now looks like another dip at 17k-ish, followed by the start of another rise. In having more hills and valleys I guess this could be called worse than v0.99, like those people on YouTube were saying.
