If there is no differential signal, then, by definition, there is no common-mode signal.
If there is no common mode signal to reject, then there is no common mode rejection ratio.
To feed a subsystem with 100dB CMRR into a subsystem with 50dB CMRR, then the 100dB CMRR system must be differential-in differential-out. Does the 100dB CMRR system output a signal with any common-mode component?

 

If there is no differential signal, then, by definition, there is no common-mode signal.
If there is no common mode signal to reject, then there is no common mode rejection ratio.
To feed a subsystem with 100dB CMRR into a subsystem with 50dB CMRR, then the 100dB CMRR system must be differential-in differential-out. Does the 100dB CMRR system output a signal with any common-mode component?

About the above. So I guess it is correct that a 100dB CMRR subsystem when fed to a 50dB CMRR main system can turn final signal into a 100dB CMRR as long as the cable is short and no additional noise, right?

Also I assume there is differential-in and differential-out in the subsystem described due to the fact there is Reference which automatically implies differential capabilities, correct?

Lastly. It seems the combination of PGA (Programmable Gain Amplifiers with gain just enough to get into the voltage level of the ADC) + 24 bit ADC (in the system described it is maximum of 250mV) is already the perfect system that beats even a pure sole instrumentation amp (like level of 3.3V) due to the fact you still have analog signal in the latter and still need ADC? Right? But then the question of this thread is what if the amplification is sole done by the instrumentation amp and you don't rely on the ADC for the gain, what is the result vs the gain relying on the ADC via PGA?