The pipeline by default interpolates across the bad columns in the REDL CCD when it creates the master bias, but does not so when it processes the science data. This inconsistency can create boxy emission-like features as seen in the figure below. You can get rid of them by re-processing your data using --clean_trap=FALSE also for the bias recipe. Below you find a list of wavelength ranges where the features can be seen in the standard REDL settings: