Some of the methods mentioned that were used to discover it are pretty simple - prior compression artifacts showing borders around sub areas of a large graph, indicating cutting and pasting of elements, etc.
What I don’t get, is why the underlying data doesn’t get included and any graphs just generated from that? Why have an actual embedded image and that’s all good?
Creating those blots is a grind. It takes hundreds of hours to set up those experiments and they often don't work, or go wrong, and you have to do it all over again. The people who are good at making these are usually the people who aren't really up to date with the whole 'computers' thing.
Yes, there would a hundred ways to easily verify the provenance of data from an image of a blot all the way to publication. However, this area of research is very slow to incorporate the latest computer tech.
(Sorry if this is a super naive HN-esque comment – the topic is far outside my field, and I don't mean to actually suggest this as a solution, I'm just curious.)
These sort of images are typically re-arranged to fit nicely in a figure for publication. It's now more common that any figure that has a western blot within it, you also have to provide the raw unaltered image. Though this wasn't standard practice at the time.
As someone in the field, western blots are terrible for a number of reasons. Even if you get nice bands and you're honest about what you show. The companies that make the antibodies for these commonly sell people duds that will bind to a number of unknown things, and people blindly trust that since the website said it's specific for protein-XYZ, that's what they're measuring.