Hi, qslazio,
Calculate and analyze what you and Hurst call return ratio and forget about what you and Hurst call loop gain. I would also suggest forgetting about all this mess with lumping one transmission into another; simply consider the following equation for closed-loop gain:
AF=G*AOL/(1+AOL*B)+G*BF
where G is the input transmission (Sin/Ss for aol=0), aol is the dependent source value, AOL is the open-loop gain (So/SinM for Sin=0, SinM means that the dependent source must remember the Sin value that it had seen before applying superposition), AOL*B is the loop-gain, or return ratio, whatever you like (Sin/Sinm for Sin=0), B is the feedback transmission of the feedback network (Sin/So for Sin=0), G*BF is the direct signal feedthrough (So/Ss for aol=0), and BF is feedforward transmission of the feedback network (So/Sin for aol=0).
Calculate all the above transmissions by applying superposition with DEPENDENT source, without ANY breaking the feedback. Just like this (attached):
https://users.ece.gatech.edu/~mleach/papers/superpos.pdf
This works very fine for circuits with a single dependent source (most practical cases).
Superposition is especially handy in noise analysis.
It is a fun to do noise analysis just 'in head' by applying superposition to many noise sources. You have to only calculate Gi and Gi*BFi for each noise source and then use one and the same AOL/(1+AOL*B) for all the sources. Try it out!
Do not be too formal! Down with cook books.
PS: Nyquist criterion is one of many stability criterions. It is not usually used to analyze phase or gain margins.