A tree diagram visualises a strictly hierarchical structure — a root node, internal nodes (branches), and leaf nodes — connected by edges that represent parent-child relationships. The perceptual mechanism is spatial containment implied by path-following: the viewer traces branches from the root outward to understand how a whole decomposes into its parts.
Unlike a network diagram, a tree has exactly one parent per node and no cycles. Unlike a treemap, the tree diagram preserves explicit parent-child structure in the layout — which path a node occupies encodes its lineage, not just its membership in a category.
A tree diagram fails the moment a node has more than one parent. In this taxonomy, "Machine Translation" could plausibly belong to both NLP and Coordination Platforms — but placing it in two locations would create a DAG, not a tree, and would require a network diagram or a Venn diagram instead. The tree structure is always a simplification of a reality that is rarely so clean.
The tree also carries no quantitative information. Node size does not encode any data variable — nodes are sized by depth only. If the quantity of capabilities per domain mattered (it does), a treemap would show it; this chart does not.
This taxonomy has 46 nodes. Fully expanded, a static tree at this scale becomes illegible — branches overlap, labels truncate, and the viewer cannot orient themselves in the structure. Collapsibility solves the scale problem by letting the viewer navigate the hierarchy incrementally, expanding only the branches they need to explore.
Collapsibility also changes the reading mode from "overview" to "exploration" — the viewer constructs the mental model of the hierarchy through interaction, which produces deeper comprehension than a static overview. This is the primary reason to choose D3 over a static tree tool when the hierarchy is deep.
Use a treemap when node size encodes a value (budget, headcount, coverage) and you need to fill a rectangular space efficiently. Use circle packing when hierarchical containment is more important than space efficiency. Use a dendrogram when the tree is produced by clustering algorithm output — the node positions carry statistical meaning (similarity distance), not just parent-child structure.
Use a network diagram when relationships are non-hierarchical — when nodes can have multiple parents or when the graph has cycles. The tree diagram is the simplest member of the hierarchy family, and therefore the right choice when the data is actually a tree.
FT Visual Vocabulary: Hierarchy — Part-to-whole structure. The one design decision worth knowing: the blood-red warning flag on ethically sensitive nodes (Facial Recognition) is redundant encoding — it uses both color (p3 blood-red, the same as domain-level nodes) and a dashed stroke to distinguish it from collapsed category nodes. Color alone would be ambiguous; the dashed stroke is the primary differentiator. This satisfies the hard no against color as the sole categorical encoding, while keeping the warning visually prominent without requiring a separate legend entry.