The flux calibration vectors for EDR2 are often jagged. I mistakenly thought that this was due to bad extraction before, but I confirmed that this is due to bad flux calibration vectors. An example spectrum is attached here. This is a flux-calibrated CALSPEC star. It is a very bright star and the S/N is supposed to be very high. The jagged spectrum is due to bad flux calibration. In each exposure, we have 40-100 FLUXSTD, each of which is slightly off from the fiber center (and the amount of offset is different from fiber to fiber). This results in flux cabration vectors with varying normalizations. When we combine them into the master flux calibration vector, the pipeline applies clipping at each wavelength, making the master vector look jagged. The solution is to normalize the flux calibration vectors to the best one (i.e., the one closest to the fiber center) before combining them into the master vector.