Architecture diagrams fail by mixing zoom levels: a box is a company, a JVM, and a class. C4 (Simon Brown) fixes the zoom. Level 1 Context: your system and the people/systems around it. Level 2 Container: deployable/runnable units (SPA, API, database, queue). Level 3 Component: major modules inside one container. Level 4 Code: classes — usually skip this; the IDE is better.

What a container is

Not Docker-only. A container is a process or store you deploy or operate: a Spring Boot app, a mobile client, Redis, S3. Lines are relationships with protocol and purpose ("HTTPS/JSON, places orders"), not just arrows. C4-PlantUML and Structurizr generate diagrams from a model so the diagram does not rot independently — still a risk if nobody updates the model.

text
Person Customer → [Web App] → [Orders API] → [(PostgreSQL)]
                         ↘ [Stripe]

Context diagrams should be readable in five minutes on day one. If you need a legend for 30 colors, you drew components at context level.

Notation hygiene

Stick to boxes and arrows; C4 is notation-light on purpose. UML component diagrams are optional at level 3. Sequence diagrams complement C4 for runtime stories; C4 is static structure. Don't put every Kafka topic on the context diagram.

Multiple systems in an org: each has a context diagram. An enterprise "all of IT" poster is not C4.

Governance without bureaucracy

Store diagrams next to the code (docs/c4). Review them in ADRs when containers appear. A PR that adds a database should add a container box. That's enough process.

Read c4model.com's example diagrams. Then redraw your "architecture" slide as a context diagram and a container diagram only. If you cannot name the containers, you cannot operate them. The model is a zoom control. Use it to stop drawing classes on the all-hands slide.