Architecture diagram
See components, boundaries, and the main path.
Explore 28 diagram types. Sequence is our first evidence-backed pilot: a human field guide, an AI contract, and checks that can actually fail.
See components, boundaries, and the main path.
Understand an end-to-end stack at a glance.
See levels of abstraction and dependency direction.
Understand hierarchy through containment.
Expose legacy pressure and modernization targets.
Follow actions, decisions, and branches.
See a multi-stage workflow and its outputs.
Trace work across owners and handoffs.
Read events along a time axis.
Plan tasks, duration, and dependencies.
Follow messages between actors over time.
See an object's valid states and transitions.
Understand entities, keys, and cardinality.
Track information from source to destination.
See storage zones and data maturity.
Follow data through bronze, silver, and gold tiers.
Connect sources, a data core, and consumers.
Compare access permissions by role and asset.
See reporting, ownership, and routing.
Navigate parent–child relationships.
Understand a reinforcing cycle.
See rank, hierarchy, or drop-off.
Position items across two dimensions.
Reason about set membership and overlap.
Compare magnitude across categories.
Read trend and change over a continuous axis.
Inspect distribution, clusters, and correlation.
Compare profiles across several common measures.
An interaction model where actors sit across the top, time moves downward, and horizontal messages explain a scenario.
Documenting API calls, protocols, authentication, or request/response behavior.
The goal is a static component inventory.
Notation semantics are separated from Diagram Atlas editorial limits. Every executable rule points to evidence.
A sequence diagram explains one interaction scenario. It privileges message order and responsibility—not system inventory or measured duration.
The vertical trace of one participant. Occurrence order increases from top to bottom; distance is not literal duration. [1]
A directed communication between lifelines, labeled with an action, request, signal, or result. [1]
A framed region for alternatives, optional behavior, loops, or parallel behavior, with guards in square brackets. [1]
This is the Diagram Atlas visual profile. UML defines message semantics; Mermaid and Diagram Design make the rendering conventions executable.
These limits are editorial guardrails, not restrictions in UML. Exceeding them produces a warning, not an invalid diagram.
This specimen runs the same validator used by the drawing workspace. Introduce three semantic problems, then repair them from the knowledge model.
SEQ-001NormativeRequiredRead event order from top to bottom on each lifeline. Never use vertical spacing as a promise of elapsed time. [1]
Message order must be explicit; reject reverse-time routing.SEQ-002NormativeRequiredEvery message endpoint must resolve to a participant declared once across the top row. [1]
from and to must match participant IDs.SEQ-003NormativeRequiredPlace alternatives, optional paths, and loops inside a labeled combined fragment; put guards in square brackets. [1]
A conditioned message must reference an alt, opt, or loop fragment.SEQ-004EditorialRecommendedSEQ-005EditorialRecommendedStart with at most five participants and twelve messages. Treat these as editorial defaults, not UML limits. [3]
Warn above 5 participants or 12 messages.SEQ-006EditorialRecommendedUse concise verb phrases, operations, or concrete results. Avoid labels such as data, call, or response without context. [3]
Warn on missing or placeholder labels.SEQ-003Two arrows diverge without a frame or guard.
Fix · Use an alt or opt fragment and label each guard.SEQ-001A larger vertical gap is presented as a longer duration.
Fix · Add an explicit timing constraint or use a timing diagram.SEQ-005Six or more lifelines and a wall of crossing arrows.
Fix · Extract a sub-scenario and link the two diagrams.SEQ-004Calls, replies, and fire-and-forget signals all look identical.
Fix · Apply one documented arrow profile consistently.The repository includes a typed validator and automated tests. Issues return stable rule IDs, severity, and the failing data path.
4 participants · 5 messages · endpoints resolve · reply style valid
SEQ-002 · message endpoint is not declared
SEQ-003 · conditioned message has no loop fragment
SEQ-005 · 7 participants and 16 messages exceed the editorial budget
We label the role of each source so an opinion never masquerades as a standard.
Object Management Group · Interaction, lifeline, message, occurrence-order, guard, and combined-fragment semantics.
Mermaid · Executable syntax for participants, arrows, activations, notes, loops, alternatives, and sequence numbers.
Diagram Design · Cathryn Lavery · Visual defaults, density limits, anti-patterns, and the specimen style. MIT licensed.
id: sequence
version: 1
status: evidence_backed_pilot
intent: ordered_message_exchange
input:
participants: { min: 2, recommended_max: 5 }
messages: { min: 1, recommended_max: 12 }
message_kinds: [sync, reply, async, self]
fragments: [alt, opt, loop]
profile:
sync: { line: solid, arrowhead: filled }
reply: { line: dashed, arrowhead: filled }
async: { line: solid, arrowhead: open }
checks:
- id: SEQ-001
severity: required
basis: normative
- id: SEQ-002
severity: required
basis: normative
- id: SEQ-003
severity: required
basis: normative
- id: SEQ-004
severity: recommended
basis: editorial
- id: SEQ-005
severity: recommended
basis: editorial
- id: SEQ-006
severity: recommended
basis: editorial
provenance:
- omg-uml-251
- mermaid-sequence
- diagram-design-sequence03 ONE KNOWLEDGE MODEL
The Sequence pilot separates standard semantics, tool implementation, and editorial judgment. Its definitions, rules, evidence, tests, and specimen can become an agent skill without losing provenance.