I think that returning nu/nuFnu for each of the components is more general, easier to plot, and would allow for faster computation of the luminosities. Returning a blob with the parameters initialized with best-fit values could be useful for theory considerations, but one could also just initialize that manually after the fact.

Hi,

In principle it can be done already by running the individual processes with the parameters obtained from the fit.
Returning a table of frequencies and SEDs is not a good solution in my opinion, because fits can be done with different processes, so the behaviour of the function will be difficult to describe (e.g. the sequence of SEDs), if you want to go in this direction the best would be to return a dictionary with names of every SED, and then also the blob can be returned, so something like:
blob, table_of_nu, {'Synchrotron': SED_synch, "SSC": SED_SSC, "EC_BLR": SED_BLR, ...}