It's possible that {{{they}}} use their own system cooked-up by the C_A for use by their own spies and that PixelKnot was only a generic example of steganography.
Wait a sec… all you are checking for is that they are either missing the normal JFIF header, or have the normal header and are encoded with a 94% quantification table, like the SS pic. Then you check to see that they have the standard Huffman tables from the JPEG spec that is used by 99.9% of all the color JPEGs in existence.
But PK is hardcoded to always encode at 90%. And 4chan's JPEG recombobulator does not change the compression quality.
94% is not a number that a developer would hardcode as a default. That is a number from someone moving a GUI slider when exporting an image from Photoshop or GIMP. So if there is stego in the SS pic then it was done with a program that does not change the quality level.
You are forcing a match on the Sarah Silverman pic without explaining why that DQT is indicative of PixelKnot.
We can't tell with the naked eye. Re-encoding for whatever reason would do that. Have you checked that ONLY the non-zero AC coefficients have changed? If any DC coeff is different or if any AC coeff was was zero is non-zero, or vise versa, then you are looking at a false positive.
I'd take up the torch if I still believed this was feasible. But we have too many unknowns to solve for.
We can test a known password against millions of pics from image boards. Or we can try a billion passwords against a (confidently) known target image. But when trying to solve both unknowns the problem size increases beyond what is feasible for two guys with high-end desktops.