I thought it might be a seam line as well, but there is only one seam line that runs vertical through that polygon. The blending mode is mosaic; would that be doing blending well away from the seam line?
Unfortunately I had to quickly redo this project a different way, so it would take me a bit to recreate to show the seam lines.
This was showing up because one of the multispectral channels had a very low signal (pixel values of ~20-30 of 255), and the other was very high. The low signal channel was set as the primary channel had its brightness settings turned up to 400% in order to process. This showed up on 3 different processing runs of different areas, but using data from the same flight where one of the channels had low pixel values.
I did a test where we brought that specific channel up (histogram expanded it) on the images in the area, and the issue has gone away. That sort of correction however can't be used for NDVI though, as it would produce non-radiometric changes. Although the channel is dark, the signal is well above the noise.
Unless mosaic blending is creating a blocky area like that from two images (seems unlikely), I'm fairly positive we're looking at jpeg artifacts.