I've always found NF to be of little practical interest in audio, except maybe when selecting transistors from datasheets.

EIN voltage is relatively easy to assess. Build the circuit, measure its gain, tehn short the input and measure output noise, you get the EIN voltage by dividing.

EIN current can only be determined by calculation, loading the input with a resistor of known value R and extracting the current noise value, using the quadratic combination rule. (EIN)²=(en)²+(R.in)²+4kTRf

The measurement sensitivity is when the R.in term is high, but It must be verified that the BW is not actually changed, due to capacitive input impedance.

You need to make sure the noise measurement BW is well defined.

Strictly, the BW should be a brickwall. Since it's hardly possible in practice, you need to know the equivalent noise BW of your measurement test set. For example, a single-pole (-6dB/octave) response ar 15.7kHz is equivalent to a 20kHz brickwall.

The 22kHz filters in the AP and HP are sharp enough to introduce little error, but the measurements need a better level of accuracy.

In order to avoid fluctuations in the reading (due in particular to 1/f noise, but also to the LF content of Johnson noise) it is safe to use the 400Hz HPF and adjust the math in consequence.