A network diagram (or node-link diagram) encodes relationships between entities as nodes connected by link lines. The perceptual mechanism is spatial proximity — the force-directed layout clusters well-connected nodes together, making dense sub-graphs (communities) visible without explicit labelling. The viewer's eye traces paths between nodes to understand chains of relationship.
Force-directed layouts use a physics simulation: links act as springs pulling nodes together, while a global repulsion force pushes all nodes apart. The resulting position is an energy minimum, not a meaningful coordinate system. Node position carries no information — only topology (who is connected to whom) and degree (how many connections a node has) are encoded.
The data structure is a tripartite network: three node groups (organisations, technologies, domains) connected by directed and undirected edges of three semantic types. The message is structural — which entities are most connected, which technologies bridge the most domains, and which organisations cluster around the same capabilities.
No matrix, bar chart, or hierarchy encodes this. A chord diagram could show pairwise flows but would lose the tripartite structure. A tree would impose a hierarchy that doesn't exist in the data. The network diagram is the only chart type that lets the clustering emerge from the data structure itself.
Network diagrams degrade rapidly with scale. At ~60–80 nodes and high link density, the layout becomes a "hairball" — visually dense, analytically useless. This dataset sits at 21 nodes and 49 links (average degree ~4.7), which is within the legible range.
If your network grows beyond this, the correct alternatives are: an adjacency matrix (scales well, loses spatial intuition), a chord diagram (shows flow magnitudes between groups), or an arc diagram (preserves ordering, reveals clusters through arc density). The network diagram should be the last resort, not the first, for large networks.
Node size by degree. Nodes with more connections are larger, making hubs immediately legible without requiring the viewer to count edges. Data Integration and Crisis Mapping visually dominate because they have the highest degree.
Link color by type. Three semantic edge types (uses, enables, operates in) are color-coded and styled — directed edges carry arrowheads; "operates in" edges are dashed. Color encodes the relationship type redundantly with style, supporting color-blind users.
Group color on nodes. Node fill color encodes group membership (organisation, technology, domain) — the most important categorical distinction in this network.
FT Visual Vocabulary category: Relationships — Connection. The Tufte principle most relevant here is data density without clutter — 21 nodes, 49 links, 3 node types, 3 edge types, and degree-proportional sizing all coexist without supplementary ink. The one design decision most worth knowing: node labels are rendered inside the circles at the cost of clipping long names. The alternative — external labels with leader lines — adds ink without improving legibility in a force layout, where the label position changes with every simulation tick.