The flux calibration of EDR2 is likely very poor. I looked at CALSPEC standard P330E, which is G2V, in visits=92557 and 92560. In the first visit, the standard star is on SM1, while it is on SM3 in the latter visit. The two figures here shows the CALSPEC star in blue calibrated using the FLUXSTD in the same visits. The black spectra are the CALSPEC spectra. The normalizations differ significantly between the two visits and also the spectral shapes of the two spectra in each plot differ significantly. This poor calibration is likely due to combination of reasons (see PIPE2D-1237 and PIPE2D-1239, and I am sure that there are more issues). The plots here are based on data from edr2-20230621/.

The following is a copy of my post to the drp-2d channel on October 13:

I looked at a CALSPEC visit from the July run (visit=97301, P330E). The top panel shows the entire spectrum and the green spectrum is the best-fit model. The shaded regions are the regions excluded from the spectral fits (we need to update these masked regions by the way). The bottom 3 panels on the left are blow-ups of a few lines. Thanks to the LSF fix, we can now fit the Halpha and CaII triplet lines well, indicating that we measure the temperature and metallicity reasonably well. The bottom-right panel is the radial velocity measurement. Most objects have a single peak. I did not do any quantitative tests, but the chi-squares-by-eye suggests the fits are overall reasonable. The Na doublet is not fit very well; the continuum normalization does not seem to work very well there and also this line was not used in the fit.

Then, this plot shows the scatter of the flux calibration vectors. If our stellar typing is good, the scatter should be very small. The red is the median and the pink region shows the 68% interval. A fractional error is ~10% at 500nm and 1000nm. Not very good, but not very bad either. The scatter inflates both at the blue edge and red edge. The blue edge is hard; Yamashita-kun's experiment a while ago showed that this region is hard to calibrate. I am not sure about the red edge, but let's not worry too much about the IR arm.
 

Finally, here is a comparison with CALSPEC. The blue is from PFS (P330E star flux-calibrated by the pipeline) and the black is the CALSPEC spectrum. there is a tilt at ~30% level. Due to the chromatic effect, we should not over-interpret this number, but this is what I am getting now. I will look at other CALSPEC stars.
 

I looked at other CALSPEC objects. Here are all the plots. It is clear  that some visits suffer from the bad extraction (due to bad detectorMap and/or bad profile)

Note that some CALSPEC objects have large proper motions. Moritani-san told me that they included proper motions of CALSPEC objects after July 24 2023. P330E here may be affected by its proper motion.

The proper motion will have chromatic effects as the PSF is larger at bluer wavelengths, so decentering matters more. The dependence is classically lambda^{-1/5}, but it's usually better described by c. lambda^{-0.25} probably due to the effects of the atmospheric outer scale.

The plots above are based on problematic reruns with bad calibs. I made a separate ticket to further address the flux calibration accuracy with a newer rerun. PIPE2D-1341. Closing this ticket.