When a graph starts helping
Honest answer: not at twenty notes, and not as a picture. Here is what changes, and roughly when.
The graph is oversold, including on this website
A knowledge graph makes a beautiful screenshot and a poor first impression. At 20 notes and 26 links — the size of the example on the homepage — it is legible and pretty and it tells you nothing you did not already know, because you wrote all of it this month.
At several thousand notes it becomes a hairball: a dense grey mass where every node touches every other and no shape is visible at all. Both ends of that range are common, and neither is the case that justifies the feature.
What actually changes, and when
The useful moment is the one where you have forgotten some of your own notes. That is not a note count so much as a time span — typically a few hundred notes written over several months, because forgetting needs time more than it needs volume.
Three specific things start happening around then:
- A hub appears that you did not plan. One note accumulates links from directions that have nothing to do with each other. That note is usually an idea you have been circling without naming.
- Two clusters touch through a single note. On the homepage example, a book about how states simplify what they govern connects to a work note about which numbers a team reports — through one idea note between them. Those bridges are where the actual thinking is.
- An island shows up. A cluster with no link to anything else is usually a project you abandoned, or a subject you never integrated with the rest of what you know.
Look at the local graph, not the global one
The global view is a poster. The view that earns its place is the local one: this note, its neighbours, and their neighbours. That stays readable at any vault size, because its size depends on one note's degree rather than the whole collection.
The same is true of the backlinks list, which is a local graph rendered as text and is genuinely useful from the very first week.
What a graph will not do for you
- It will not create connections. It renders links you wrote. A sparse vault produces a sparse graph, and no amount of layout algorithm changes that.
- It will not tell you what is important. Degree is not significance. The most-linked note in most vaults is something structural like a daily index.
- It is not a search interface. Finding a specific note by clicking around a graph is slower than typing three letters, every time.
A note on performance, since it is the question that follows: I do not have a vault size I would publish a number for. The graph is the part of Lattice that gets slow first and its optimisation work is still open. Quoting an unmeasured limit is how a benchmark becomes a support ticket.