Hi Tony,
Except for a mismatch in the strength of the lines, probably due to a mismatch in resolution, the disparity at the blue end does not look that obvious to me from here. What does the fractional error look like if you divide one by the other and smooth the result?
I (and others) have also found getting a good flux calibration below 4000A tough. Most of the sources of error were probably been covered in this thread.
https://britastro.org/node/9700
and Chrisitan Buil’s site covers chromatic aberration, atmospheric absorption and the effect of position on the slit
http://www.astrosurf.com/buil/dispersion/atmo.htm
I suspect the best (though time consuming) solution when after highest accuracy in the continuum shape at the blue end is to use a spectrophotometric technique using a very wide photometric slit to determine the shape of the continuum and combine that with the resolution of a narrow slit spectrum eg
http://www.astrosurf.com/buil/calibration2/absolute_calibration_en.htm
David found using a wider slit gave better results, though the reason for the improvement was not clear in this case
https://britastro.org/node/9199
Cheers
Robin