Checkpoint: Can You Read the System as a Whole?

Diagram Design becomes useful only when its parts cooperate: a reader's question determines the relationship, the relationship determines a pattern or layout grammar, and the chosen view is then constrained by hierarchy, evidence, and delivery context. This checkpoint connects those ideas into one judgment. The test is practical: can a person diagnose a mismatched diagram before worrying about its colors, and can they tell a documented rule from a claim about proven effectiveness?

The choice can be stated in one sentence

A strong diagram decision can usually be expressed as a sentence with a subject and a relationship: “This figure shows how requests enter the order platform and reach the data store,” or “This figure shows why two policy evaluations diverge,” or “This figure shows how a starting total changes through signed contributions.” If the sentence is only “This figure shows the system,” the subject is present but the relationship is not. Without that relationship, type selection becomes a search for a familiar shape.

The selection loop is therefore short but ordered:

flowchart LR N01["Reader question"] --> N02["Dominant relationship"] N02 --> N03["Pattern or visual type"] N03 --> N04{"Within budget and evidence?"} N04 -->|"Yes"| N05["Choose focal signal"] N05 --> N06["Redraw or author the figure"] N04 -->|"No"| N07["Use prose, table, or split view"]

The first box keeps the primary subject in view. Diagram Design is not asking an author to replace an order platform with the abstract idea of “a secure system,” or a database with “structured data.” The system's patterns and grammars are lenses through which that subject becomes understandable. They do not become the subject themselves.

The second box names the load-bearing relationship. Is the reader following a path, a state change, a hierarchy, a quantity, a physical placement, a dependency, an ownership boundary, or a user's experience? Several of these relationships can exist in one system. The choice is not to deny the others; it is to decide which one gets the main figure. If two relationships both require full treatment, separate views are more honest than one canvas with competing grammars.

The third box separates behavioral meaning from arrangement. A queue or trust boundary may require a semantic pattern, after which Data Flow or Architecture supplies the layout. A plain hierarchy may need only Tree. A quantitative request may go directly to a chart grammar. The result is a choice that can be defended in terms of what the reader must see, not just what the author happened to draw first.

The fourth box is where editorial taste and evidence meet. A figure can be within a node limit and still be wrong for its claim. A chart can have elegant spacing and still misstate a value. A policy trace can show two colors and still fail to mark the first divergence. “Within budget” means that the selected view has enough room to preserve the required semantics, not merely that the number of boxes is acceptable.