I've submitted the ATI to the AP. Performance is up to expectations:
Frequency response <5Hz-200+kHz BW (+0,-0.5dB)
EIN of the measurement section better than -103dBu@unity gain, -88@20dB gain (22Hz-22kHz BW)
THD+N better than 0.0035% from 20Hz to 20kHz (measurement BW 30kHz)
Residual noise of generator section -108dBu @ 0dB attenuation (22Hz-22kHz BW)
The only performance point I had not been capable of predicting was the real capability of the circuit to drive unbalanced sources with the attenuator in-circuit. Unlike a xfmr-output, the 1646 has a finite CM impedance, which alters the level when on eof the legs is shorted to ground.
I have found this difference to be less than 0.2dB for 50r and 150r output Z, and less than 0.5dB for 600r output Z.
It's not a big problem, but still one has to keep in mind.
Overall I'm very pleased with the results, now I should receive my Q400 next week, as well as the metalwork.