Main idea

The size of a note is judged by what it does to its neighbourhood. A note that talks about too many things cannot be connected cleanly: you no longer know whether the link bears on the definition it holds, on the example, or on the conclusion — and the relationship you meant to name becomes impossible to name.

The opposite excess produces a symmetrical and less expected damage. Cut into micro-ideas, each one demands its own link just to recover what the whole note was saying; the number of edges explodes, and the network becomes a chore to read at the very moment it looks best built.

The right granularity is therefore sought between these two bounds, and neither of them is a matter of length as such: it is the readability of the relationships that sets the interval.

Why it matters

This gives a splitting criterion that does not depend on taste: you do not ask "is this note too long?" but "could I say exactly what a link to it connects?". The question comes up at the moment of linking, which is to say at the moment the answer is observable.

It also corrects the belief that fragmenting is always safer. Splitting is reversible, but splitting endlessly does not make a base finer: it makes it unreadable from the other end.

Nuances and limits

The interval is not the same for everyone, nor for every kind of material. A base kept by one person tolerates broader notes than a shared base, where every reader has to work out what the link was pointing at.

And the fragmentation fault only shows at scale: a note that is too fine looks impeccable taken on its own, and the cost appears only once the network is traversed.

Open questions

• Can over-fragmentation be spotted other than by noticing after the fact that the network has become unreadable?