A poster named gsgs observed that two sides of each foreground layer aligned with 8×8 bit boundaries in the 300 PPI resolution, which is the resolution for the foreground layers. Realizing that the image was likely scanned at 600×600 PPI and then the JPEG was padded to align with 16×16 bits (MCU) in this resolution, you may understand why these become 8×8 bit alignments in the 300 PPI resolution.
The WH Long Form Birth Certificate indeed showed that 2 of its sides aligned with these 8×8 bit boundaries, the other two sides aligned with the inside of the image, in the sense that these lines were touching at least one pixel of the image.
The present explanation is somewhat preliminary but given the evidence, the following workflow makes sense:
The scanner scans the picture at 600 dpi and separates it into a background and multiple foreground images.This is done by analyzing a 16×16 pixel element
Before converting the background to JPEG, the software pads the image to a multiple of the (MCU) Minimal Coded Unit, which for this particular situation is 16×16 as both the Cb and Cr channels are subsampled by a factor of 2.