Chapter 28

What Density and Layout Show

Chapter 27 ended by asking what it means when Nodes crowd together in a projection. Lay a graph out on screen and you see not only where Nodes sit but how tightly they gather. What is that crowding?

There is no single answer. It may be the size of a topic, a region where relations concentrate, or an artifact of the layout algorithm. So density and layout have to be understood separately.

1. Position is not meaning

Node A sitting near Node B does not mean the two are close in meaning. Nearness arises from similarity of meaning, from topic grouping, from graph connectivity, and from layout optimization alike. On-screen distance alone must not be read as a semantic relation.

The default layout of Chapter 25 does exactly this. The hub preset reads the topic string as a path and extends the Nodes of one topic along one direction. So Nodes of the same topic being near each other is not a discovery but a construction. What that nearness says is “under the same topic”, and nothing more.

2. Density is a property of a collection

Density is not a property of one Node but a measure of how many Nodes and Edges sit in a region of space. A dense region shows many Nodes or Edges in one place. Why they are there is not something density says.

3. Topics make density

The Domain nodes of Chapter 10 build the tree of topics, and the layout reads that tree as direction. So the graph shows density by topic. This is different information from the semantic similarity of Chapter 17: not what search found close, but what a person filed under one topic, placed together in space.

4. What the Gaussian research left behind

There was one study. Treat the distribution of Nodes within each topic as a Gaussian, place the Nodes around that distribution, and perhaps natural clusters and density would come out together.

The verdict split three ways. Using it for placement was rejected: it would undo the regularity and size-aware spacing fixed that August and blur the boundaries between topics. Using it as a radial ramp was skipped: what a Gaussian would say, that membership is a matter of degree, the radial rank already says, with the better-connected Nodes nearer the topic's axis.

One direction remained: not a Gaussian as the algorithm that places Nodes, but a density layer drawn over the Nodes already placed. The research went as far as naming blended screen-space Gaussians as the right primitive for that layer. The layer is not yet drawn. In the verdict order of Chapter 27, causal order and cones came first, and density is next.

5. Density is a layer, not the layout

Layout is where a Node goes, and the sidecar decides it. Density is where Nodes and Edges gather, an auxiliary representation laid over the layout. The two need not run through one algorithm. The base coordinates come from the sidecar as in Chapter 25 §6, and a density layer draws the concentration around them. A Node remains an object of the graph; density is an auxiliary representation of how that graph is spread through space. Merging the role of layout with the role of density was rejected because it would blur the very structure the View is trying to keep.

6. Density does not explain its cause

Suppose a dense region exists. Looking at it alone cannot tell you which of these it is: the same topic, many relations, a concentration of search results, or the layout algorithm.

So density is a clue for interpretation, not the interpretation. To give it meaning you have to look at the Nodes and Edges that make up the region. Nodes of one topic crowding one side of the screen does not prove that topic is discussed more than others; they may have arrived recently, all at once, and not yet spread.

7. Layout has to be a function

The direction of the View design is to treat coordinates as a function computed from the graph's state rather than as arbitrary stored state. A coordinate is a function of the graph, the scope, the viewpoint and the layout preset.

The same input should give the same layout. That is why, in Chapter 25 §6, the sidecar's pure functions compute the coordinates and the client runs no force simulation. It is what lets the same space be reproduced when a View is reopened or drawn on another machine, and what lets two people opening the same pack see the same picture.

8. Proximity is not evidence, and density is not importance

A Node sitting near another does not give their relation any evidence. Evidence, as Chapter 23 showed, comes from a sentence and is written onto the relation. Proximity in space is not evidence.

Density is not importance either. A dense region does not make its topic more important. This is why visual representation and graph meaning are kept apart.

9. Space is the result of several layers

Vinc's space is not explained by one layout algorithm.

Each part of the graph feeds one visual channel; together they pass through the View model to coordinates, an optional density layer and the renderer kind of Node shape kind of Edge line kind and ink density topic direction and grouping time the Timeline axis causal order rank layers, cones evidence ink density of a relation View model coordinates and a density layer, not yet drawn renderer

Layout shows order and grouping, density shows concentration, and relations decide meaning. A good graph View does not explain everything through one rule like “close means similar.” Hold to that and space stops distorting what the graph means and becomes an interface for exploring it.

This closes Part 3. Where memory lives was seen through time, shape, space, causality and density. One question remains: was this design actually right? Part 4 is the story of measuring that, and its first chapter begins with the fact that measurement needs conditions.

Continue with Chapter 29, Measurement Needs Conditions.