Main idea

A ten-page synthesis on a subject may hold forty ideas. They are true in it, well sequenced, and collectively out of reach: not one of them can be quoted, compared or linked without dragging the other thirty-nine along. The synthesis is an indivisible unit, and its grain is the document.

Writing the same material as short notes, each well titled and explicitly linked to the others, changes the nature of what you own. Each note is a point; the network of links between those points is a map. What you can then do is not merely find an idea again: it is to hold two side by side, coming from two places that would never have met in a continuous text.

This holds for a human reader and for an assistant, and it is the latter case that made the constraint visible — a memory system for agents such as A-MEM (Xu et al., 2025) is built directly on this principle: structured notes, indexed, linked, which the agent navigates instead of loading everything in one block.

Why it matters

This gives a reason to fragment that rests neither on reading comfort nor on the Zettelkasten fashion: it is the grain of the writing that decides what can be done with it afterwards, and it is chosen once and for all at the moment of writing.

It also explains why a base of long syntheses is not reused, however well filed and well indexed. Finding the right document does not give access to the idea it contains.

Nuances and limits

The synthesis keeps its own function: holding a line of reasoning from end to end, with its order and its transitions, is what a collection of notes does not do. The point is not to replace one with the other, it is not to mistake one for the other.

And fragmentation has its symmetrical excess: notes that are too fine each demand a link, and the network becomes as impractical as the monolithic document.

Open questions

• At what grain does the splitting stop, given that the signal of an excess — too many links to make — only appears once the base is already written?