Just another quick sort of on topic off topic. Gustard was able to eliminate the hump in X16 but their way of fixing it increases THD+N. So they chose to keep the hump and have the best THD+N/SINAD they could have. I don't see the issue considering the hump is already quite small. It's sort of prominent but that's because the noise floor is so low on that device.
It could be done with no hump and the same THD+N though. I can't tell exactly what causes or fixes the hump if you ask ESS they'd tell you the same. But certainly Vref, clock, layout related. Also for worst cases, resistor matching would help, changing opamp would help. It's one thing that everything just adds up to the final result. In a good design you don't even need differential amplifier in the end to have low distortion (unlike John from Benchmark had said before). I thought just using one arm of the output would be horrible until I actually measured it, difference is just a -130dB 2nd harmonic.