There is currently code in ip_isr that allows an amplifier to be split into two segments for the purposes of overscan correction. This was originally added for DECam, which has amplifiers that show a bias jump, where the overscan level is discontinuous. When using a simple correction method (full overscan median as an example), ignoring this jump results in a measurement that is bad everywhere.
However, I believe that obs_decam is now using the MEDIAN_PER_ROW algorithm universally, which solves the problem naturally, while also better fitting smaller features along the y-axis. If this code isn't expected to be used, it would help simplify the overscan refactor (