Anti-alias filtering is part of the measurement

Sampling folds energy above half the sample rate back down into the measured band, and once it has folded there is nothing downstream that can separate it out. A tone at 90 kHz reaching an 80 kSPS converter appears at 10 kHz and looks exactly like a signal the instrument exists to find. Everything after the converter would treat it as real.

I wrote the filter as a specification rather than a component choice: a 3 to 30 kHz passband with variation within 3 dB, at least 20 dB of attenuation above approximately 40 to 45 kHz, and suppression of very-low-frequency drift and DC offsets at the bottom end. The acceptance criterion is stated from the failure side — the converter must show no significant alias frequencies.

The test follows from that. A constant-amplitude sweep from below 1 kHz to above 100 kHz, with the complete response plotted, and then an out-of-band tone injected while watching for a false in-band signal appearing in the digital record.

A filter that has not been measured is an assumption. Rev A places the network; it does not establish what it does, and the response has to be deliberately designed rather than inherited from whatever the component values happen to give.