the problem might  be the same on this spectrum, but with a fainter amplitude). It happens to produce a wrong redshift solution for PIPE1D

The wiggles are present in the extraction, both for the science exposure and the quartz. There are some features in the fiber profiles that I suspect may be causing this.

more proof on this, looking at n1 quartzes.

I believe the cause of the wiggles in the extracted fluxes (which I've found are far more extensive than just a single fiber) are wiggles in the fiber profiles of r3, shown below. The r1 profiles show no such wiggles, and so I'm inclined to believe this is related to the optical misalignment of spectrograph=3 during the 2023 April run. I am trying to measure the profiles more carefully to suppress the wiggles.

The fiber profiles are constructed by iteratively measuring the profile and measuring the spectrum (because the profile affects the spectrum which affects the profile). If the measured profile has any nasty little features in it (like those wiggles, which might come as a result of the initial guess profile being a long way off, like when SM3 was mis-aligned) then those were being used to measure the spectrum, which could really slow down the convergence. Instead, using a Gaussian profile with the median profile width to measure the spectrum causes the process to converge much more rapidly.

Here are the same profiles as above, measured with this technique:

Merged to master.