A wide “photometric” slit is certainly needed for absolute flux calibration of course but in my experience, provided precautions are taken I have found it is possible to consistently get “reasonable” (+-10% say above 4000A) broad continuum relative flux calibration with my setup using a slit width the same order as the star PSF.
(Except for the effect of changing focus on the original run and an example of the effects of scintillation on the latest run, the continuum errors my two runs using MILES stars I have documented here have been reasonably under control. You can also see some comparisons with professional spectra in what is perhaps a worse case scenario of faint targets using the modified ALPY 200 here.
Comparisons between observers on the same object posted on the ARAS forum using the reference star techniques are also usually (but with admittedly some exceptions) consistent.
Perhaps my success though is due to the not particularly good guiding performance which effectively scans the centre of the star image across the slit.
Speaking to a professional recently, this sort of thing is tough for them too and their approach seemed to be to specifically measure the instrument response using standard stars only occasionally on photometric nights, removing the effect of the atmosphere by measuring at different air masses and then measure just the atmospheric extinction during the observing session, typically making one standard star measurement and correcting for air mass using an atmospheric model generated for their particular local conditions. This gets round the problem of having to find a reference star close to the target but does not address the problem of selective wavelength sampling at the slit which may be what we are seeing here, though I believe that is yet to be proved. (David, perhaps you could try running with a wider 35um or even 50um slit to see if you get any improvement?).
Buil’s ideas on spectrophotometric measurements using a combination of wide and narrow slits are probably worth thinking about in this context too, though adopting the 4 spectra measurement every time would reduce throughput significantly and might still give extinction errors with our fickle weather.
Professionals though probably measure with the slit orientated at the parallactic angle and their CA and positioning and guiding errors are likely much lower than ours in general so perhaps selective sampling at the slit is not a big issue for them