Discoverability. I can't count the number of times I've had to ask phone support people to please read the notes that the last agent took. If you have a paper form, someone marking in the margins is immediately apparent.
Hypermedia didn't go far enough, in the sense that it overlooks how people form relationships with specific documents. In case of margin notes, individuals take an instance of a document template and mark it up to customize it. They recognized that the shape of the form was not the only shape the document could take and they knew that the form was an extension of their authority. So if it didn't make sense they changed it.
You can't really reproduce that experience with hypertext because there's no concept of "this document doesn't apply to this situation let me, the individual user, change it to communicate something unique." And because humans aren't reviewing any of these documents, there's nobody there to interpret it correctly even if you could. Essentially the Internet and web form design and replacement of humans with machines has involved a great abdication of authority from the front-line employees who process these forms. I would argue that it's becoming the downfall of our modern society. If a programmer or analyst couldn't envision your situation, nobody works at the company who can do anything about it.
Working for a company that has mixed paper/electronic documentation. I can’t overstate how useful it is to highlight, circle, and/or cross-out sections of a form. Also to see /what/ was crossed out to know the history of the document.
You might say, “well we can add these formatting features and make the form elements removable and…”. But if you know you know. Asking “normal people” to do all this in markup is an insta-fail. Most of us are caterers, not programmers.
AI can handle the mixed form... Perhaps now with LLMs we can just scan all the stuff with crossed out words and cicled conditions and have the ai make it searchable.
Then again we should always ask ourselves if a drug dealer would ever recommend someone stop doing drugs.
LLMs certainly can cope with disordered notes scribbled into the metaphorical margins, and OCR can usually turn literal scribbles in literal margins into something approximating what was written, buuuuuuuuut the AI we currently have is just good enough to be dangerous if you tried using it this way.
On the other hand, the last 15 months have been "if you don't like the state of the art AI, wait a week", so I may be out of date with that assessment.
