NF Measurement of Assembly with Multiple Frequency Conversions

Hi,

I'm having difficulty verifying my test for measuring Noise Figure of an assembly using a spectrum analyzer set for noise spectral density. Essentially I'm trying to measure noise figure by loading the input to 50ohms, having the device on (with LOs) and measuring the NSD at the output. I can measure NF by knowing the thermal noise at the input (-174dBm/Hz), the gain of the assembly and the measured NSD at the output of the device through...
Problems arise when the NSD outputted by the device is below the noise floor the Spectrum Analyzer (-168dBm/Hz). The first device has a single passive downconversion mixer and no gain in the device. The second has a double conversion (up then down convert, passive mixers again) but has an amplifier in the middle.

Because of these two scenarios it was determined we should add gain to our measurement setup. We added a known cable, filter and external amplifier preceding the spectrum analyzer to increase the noise floor further. So we measured the gain of the cascade using a VNA and the NF using the spectrum analyzer method. We will call this the 'Added Gain' Stage.

So now here's how I have been measuring the entire cascade.
Now knowing the Gain of the DUT and the 'Added Gain' Stage, the NF of the 'Added Gain' Stage and the NF of the entire cascade I can use the Friis Equation to find the NF of the DUT.

I'd like someone to tell me why this would be completely wrong and not feasible to measuring NF. Thanks!