The radial velcoity of each FLUXSTD is measured using the standard cross-correlation technique. It occasionally fails even for bright stars in which strong absorption lines are observed.  This has to be robustified because the failure in the radial velocity measurement means the failure in the flux calibration vector from that star.

The attached plot is an example from the Apr 2023 run (visit=92114).The top panel shows the normalized spectra: the black is observed and the green is model. The shaded regions are affected by telluric features and are not used in the cross-correlation.  The first 3 bottom panels are the blow-ups of some of the strongest lines and the bottom-right panel shows the cross-correlation signal.

Visual inspections these cases suggest at least two problems:
1: the LSF is broader than the observation. NaD is resolved in the data (not very obvious in the attached plot, though), while not in the LSF-broadened model. Also, model absorption line depths are often shallower than the observation.
2: the metalic lines in the model are too weak as in the attached example. We used the best-fit model from the broad-band fit, but the broad-band colors are sensitive only to Teff, not to Z nor [alpha/Fe]. We may have to adopt a set of fiducial values to Z and [alpha/Fe] (and perpahs also logg) regardless of the broad-band fit for the cross-correlation template.