I think both matter. You definitely need the latter to be able to follow ideas outside of what you’re currently reading − using both explicitly linked content and automatic backlinks. But for anything larger than a screeful of text (my totally arbitrary estimation) you probably want the former too. What an «organization» (or «structure») of the document gives you is its abstraction. When sections of a (long-ish) text are named properly, you can get a lot of value from just looking at the TOC − just like looking at something remote using just your eyes first before going straight to the binoculars gives you a better idea of the entire thing, without concentrating too much on individual details. If other words, good document organization lets you see the forest for the trees − something you can’t do with a «wall of text» without reading it in its entirety and processing it after reading.
Simply put, because not all text is equal.
A list of possible reasons why this might be handy:
- A named reference (consider how many people use «Sermon on the Mount» instead of «Mathew 5:1−7:27» to refer to the thing − despite having references to the individual chapter and verse!)
- «Note within a note»: you can have a smaller and relatively self-contained piece of text within a larger text, e.g. tangential stuff that can be safely skipped without losing track of what’s happening in the main text. Or the opposite of that: something important that should stand out. Arguably you can do that with formatting, but then you start diverging from the «plain text» paradigm and moving into «fonts and colors» domain of Microsoft Word et al.
- To mark an end of one thing and beginning of another. Consider plays of Shakespeare: without separate acts and scenes they wouldn’t make any sense. Arguably you can remove this by making each scene its own «note», and linking directly to the next one, but I fail to see how this purely navigational use of linking is much better than having the entire play or act as a single document. This is generally true for all documents that should be read sequentially in a specific order − linking doesn’t bring much improvement to the table when used as a navigation tool (as opposed to semantic linking).
This is just off the top of my head − not an exhaustive list.
Re-reading your initial post again, I see now that it can be taken as an argument against pages as artifical construct orthogonal to the document structure, and not as an argument against any and all structure. In that case the argument is somewhat legitimate. However, as you surely know from the use of paper-based maps, having orthogonal reference (as in «Town hall is in the sector H2 on this city map») can be very useful, especially if there’s no other way to refer to the thing you want.
To me, this boils down to the tradeoff between document length, the medium, and granularity of the links/refecences.
Given that there’s another way to reference (and thus link) pieces of text, you might consider pages irrelevant. But I think that the question you should ask yourself is: «How granular I want my refereces to be?» Absent block references, you can only refer to the entire note (i.e. a file). To make your reference useful/relevant in this case, the note has to be small enough, otherwise the link brings in a lot more context (potentially irrelevant!) than you initially wanted. If your notes aren’t always small, you would definitely want a way to make your references more granular. This is where headings/sections and blocks/paragraphs come into play. Obsidian has block references now. However, we returned to the start of my post where I defend the need for granular document structure: being able to structure the document using sections, chapters, blocks or anything of that sort is definitely useful, and even benefitial for linking as it allows referring to specific parts of the document, regardless of the medium.
For non-electronic documents of non-trivial length though, I don’t see how pages can be made irrelevant in the absense of another way to link to sections of other documents: one page is sufficiently small to make your reference relevant enough while still keeping it precise. In my layman understanding of references as used in scientific publishing, it might be enough to just name the cited paper properly without having to provide specific pages, but such reference is very generic and works not-too-terribly only for cited works with very low number of ideas (i.e. one thesis for the entire paper, and even in that case they still provide abstracts to allow getting the gist of the paper quickly!), or in cases where the referenced idea is very general and/or well known (i.e. referring to Marx’s «Das Kapital» as a whole book, not to its particular sections). In all other cases your reference pretty much forces you to read the whole cited document, which makes your reference way too general. Thus, we need pages if we want our dead-tree documents to have precise and relevant links.
So, that’s my 0.02$.
Please don’t consider my post as an argument against backlinks or graphs or anything like that − those are wonderful and powerful new mechanisms that help us more easily do what we consider important. What I suggest is that we don’t give up on things that served us well for very long tme just because we have some new shiny thing − consider those tools as supplementary, not as competing. Take the best of each!