Maybe slightly better with E1DA scaler + ADC:
View attachment 356709
THD rising <70Hz for the internal reason of ES9822, I didn't find a way to reduce that anyhow. About THD rising at >7kHz, this is ES9039Pro profile. You can measure 10k THD with APU and see that H3 will be about -110db, IMHO, it is a shame. BTW, ES9039Q2M has -135db of H3 at 10k test.
As I told you, Cosmos APU has residual THD+N about -135db@1khz and 10Vrms(or 5V with +6db), roughly it is 3x times less than APx555 on a lucky day. All DACs in the range of THD+N -126-130db@1k APx555 will show as -124-125db due to it being the inherent limit of that noisy analyzer. When L7 tested Cosmos APU he tried to measure THD+N of some DAC(not sure which model but 9038Pro) which got -125db at APx555b and next with Cosmos APU->Cosmos ADC->APx500_SW, the result was -127.5db as I remember. Please note, L7 APx555b is a few years newer vs Amir's and L7 always gets a bit lower THD+N on it.
PS: Paul, I guess you (and John-REW) haven't seen my Excel sheet-tool for an ADC residual noise compensation? I intentionally set here A-weighted units to minimize the error of subtracting scalar noise values by narrowing of BW(A-curve is about 100-12500Hz where a noise floor is always flat). And it works not too bad, for instance, Meizu HiFi DAC with bare Cosmos ADC shows SNR -122db(A), after Excel sheet compensation -128db(A), and with APU -130.5db(A). You can subtract noise vectors (bin by bin) and don't care too much if the noise floor isn't flat. Due to the THD+N combo being limited by the N component in most cases, you may compensate for the APU+ADC residual noise and move -135db limit even deeper. The same thing is possible for APx500 but seems AP also haven't seen my Excel sheet