This is a great observation. I hadn't made the connection, but this is one reason why continuous-scroll digital reading is uncomfortable. notes.andymatuschak.org/z6cxCDMXRWBrit…

One challenge here is that the semantic interpretation of scroll position as application state is interface-specific. So persisting this state can’t take place at the layer which would ordinarily be its model. (more discussion in note)

Feel similarly about page boundaries, which are arbitrary. Not quite as far from the model as scrolling position, but far still. Something closer would be section or chapter titles, which are — if the author chose well — part of the ”model” and help anchoring to doc’s structure.

That’s why I appreciate excellent navigation as you designed for quantum.country. Also related to addressing: references with page numbers feel somewhat broken, as they bake in the layout into the addressing mechanism. Works as long as you stick to (simulated) paper.

Smaller units (atoms?) of text with address based on document structure would solve this, similar to how legal documents work. I believe this also ties into what makes your notes system work — you operate on that smaller unit level and everything is easily addressable.

Right. It's tricky… there's no great universal reference chunk. Pages, paragraphs, text ranges, subsections—all have issues! It's a rough issue for hypertext systems in general.

That's a great Note. An even stronger claim/title: Transclusion Only Works with Context-Free Nodes, of which There Are None.

One reason I read all non-books in Instapaper is because it gives me that per-page state stickiness, even across devices.