We’ve kept the entire packet header unencrypted so the effect is more visible. The final implementation will just leave the minimum required bytes (1 or 2) of the header unencrypted.

Interesting, thanks for the information. And also thank you for working on this important technology.

I think, I haven't looked at their implementation; they'll be an artifact of piping the encrypted/mis-decrypted stream into a video decoder. A lot of video coding techniques use variable-size blocks to describe changes to areas of the image, so it's reasonable that piping pseudo-random data into the decoder would produce some noticeable block shapes.

I think it's more a side effect of how video codecs try to make sense of bad data.