A list teaches you that things exist; a graph teaches you what to learn next. The registry’s resources are connected by a small set of explicit relationships, emitted as graph.json at build time — 204 nodes and over 600 edges at current build.
The edge types
Five relationship kinds, all declared by curators in the source records and flattened during normalization:
in_category— the backbone: every resource belongs to exactly one category, every category to exactly one stage. This is what makes the roadmap, the taxonomy and search agree.prerequisite_of— “learn/run this first.” The dbt model-review skill is a prerequisite of the dbt test-writer skill; following these edges backwards from any advanced resource gives you a learning path.uses— links a resource to the technologies (dbt, Spark, Kafka, …) it actually exercises, from the first-class technology list.related_to— lateral “see also” edges: a skill related to its category, a pattern related to the quality-gate skills that implement it.implements— the concrete-to-abstract edge: starter projects and skills that implement an architecture pattern. Read the pattern for the why, then the implementation for the working copy.
What the graph is for
For you: detail pages render their neighbourhood — prerequisites above, related entries beside, patterns below — so no entry is a dead end. The tutorials page is effectively the graph sorted into a queue.
For agents: the graph is machine-readable. An agent asked to “teach me data contracts” can walk in_category → the data-contracts category → prerequisite_of chains → implements edges to a runnable project, instead of guessing.
For curators: edges are validated like data. A relationship pointing at a slug that doesn’t exist is a build error, not a 404 discovered later — referential integrity is enforced by the same pipeline stage that validates the schema.
Why so few edge types
Deliberately. Five relationships are few enough to stay meaningful at 135 resources and dense enough to express learning order. If a relationship can’t be said as prerequisite, uses, related, or implements, it probably belongs in prose, not in the graph.