Design Lab Methodology¶
The Design Labs section uses a 16-section Architecture Decision and Validation Record (ADVR). The goal is to keep architecture work evidence-based, testable, and reusable across design, review, and operations.
ADVR lifecycle¶
flowchart LR
A[Frame decision] --> B[Compare options]
B --> C[Choose recommendation]
C --> D[Predict outcomes]
D --> E[Validate and falsify]
E --> F[Operate and revisit]
F --> A The 16 ADVR sections¶
- Decision Question — the architecture question that must be answered.
- Business Context — users, critical journeys, and business drivers. [Documented]
- Scope and Non-Goals — what the decision covers and what it intentionally excludes.
- Constraints — compliance, budget, geography, skill, latency, and lifecycle constraints.
- Quality Attribute Priorities — ordered priorities such as security, reliability, performance, cost, and operability.
- Candidate Options — realistic alternatives that could satisfy the need.
- Recommended Option — the selected pattern or baseline.
- Architecture Hypothesis — why the recommendation should fit best. [Inferred]
- Predicted Outcomes — expected benefits, costs, and operational consequences.
- Validation Plan — the tests, reviews, and measurements required before standardization.
- Falsification Criteria — the evidence that would prove the recommendation wrong.
- Evidence — Microsoft Learn references, measured results, diagrams, and review notes.
- Trade-offs and Risks — what the team gives up and the known failure modes.
- Guardrails and Operating Model — policy, ownership, alerting, runbooks, and release controls.
- Revisit Triggers — conditions that force redesign or revalidation.
- Takeaway — a one-page summary for decision-makers.
Evidence levels for architecture¶
| Tag | Meaning | Typical source |
|---|---|---|
| Documented | Explicitly stated in official guidance | Microsoft Learn, standards, approved ADRs |
| Observed | Seen in telemetry, deployments, or operations | Logs, metrics, platform behavior |
| Measured | Quantified result | Load test, cost export, recovery metric |
| Validated | Confirmed through experiment or drill | Failover test, threat model review |
| Correlated | Multiple signals support the conclusion | Combined logs, incidents, review notes |
| Inferred | Reasoned synthesis from evidence | Architecture analysis |
| Assumed | Working assumption pending proof | Early design workshop |
| Unknown | Missing or unresolved evidence | Gap to investigate |
How to write an ADVR record¶
- Write the decision question in one sentence.
- Capture business context before naming services.
- Rank quality attributes so trade-offs are explicit.
- Keep candidate options realistic and comparable.
- State the architecture hypothesis in falsifiable language.
- Predict outcomes with numbers where possible. [Inferred]
- Define validation steps that can be scheduled and owned.
- Record guardrails so the design survives beyond diagram approval.
Example ADVR stub¶
## Decision Question
What is the right baseline architecture for workload X on Azure?
## Business Context
- Users: ...
- Driver: ...
## Candidate Options
1. Option A
2. Option B
## Recommended Option
Choose Option B because ...
## Validation Plan
- Load test to ...
- Threat model to ...
- Cost review to ...
Writing guidance¶
- Prefer concise claims with evidence tags.
- Use diagrams to show request flow, dependency boundaries, and failure domains.
- Link the lab to the matching workload guide and infrastructure path when available.
- Mark unresolved points as [Assumed] or [Unknown] instead of hiding uncertainty.
Microsoft Learn references¶
- https://learn.microsoft.com/en-us/azure/architecture/patterns/
- https://learn.microsoft.com/en-us/azure/well-architected/