Architecture traceability is sometimes presented as a choice between exhaustive relationship mapping and a lightweight practice that records almost nothing. Neither extreme is especially useful. An architecture repository should be able to explain why a decision exists, what it affects and how an organisation will know that it has been implemented — without asking architects to complete a compliance questionnaire every time they describe a service.
Model comprehensively; capture progressively
The underlying information model can be comprehensive even when the first interaction is deliberately small. An architect may begin by recording a requirement, a concise description, an owner and a project context. Later, as the design matures, the architect can link the requirement to a business goal, relevant principles, an architectural solution, a decision and verification evidence.
These are different concepts, and the relationships between them carry meaning. A requirement is not the same as a standard; a decision is not a delivery task; evidence that something was tested is not proof that the original business objective has been achieved.
A good tool should support all these distinctions without demanding that every record arrive fully populated.
The fictional example: reducing duplicate customer records
Imagine Northbridge Services, a fictional organisation whose customer records are duplicated across three applications. A new programme proposes a shared customer-reference capability. A first architecture entry might contain only:
- Requirement: Applications must resolve customers against a single governed identity.
- Driver: Reduce conflicting customer records across business processes.
- Accountable owner: Customer Operations.
- Context: Customer Data Modernisation project.
That is already a useful architectural fact. It should not be rejected simply because nobody has yet chosen an API pattern or agreed a verification method.
As the proposal matures, explicit links are added: the requirement supports a data-quality goal; it is constrained by a privacy principle; the proposed customer-reference API addresses the requirement; an Architecture Decision Record records why synchronous lookup was preferred to replicated reference data; and an integration test provides evidence against the agreed acceptance criteria.
Each link answers a different question. The repository becomes an explanation of the change, not just a collection of records.
Not every relationship should be mandatory
A useful approach distinguishes three levels of information:
- Essential at capture: enough to identify, describe and contextualise the requirement.
- Expected before approval: links required to justify a significant architectural decision, such as impacted services, applicable constraints and rationale.
- Expected before assurance or closure: evidence of implementation, testing, operational acceptance or justified exceptions.
The precise thresholds depend on risk and significance. A change to a minor report should not face the same assurance burden as a cross-enterprise identity service.
Make the traceability observable
Traceability is only valuable if someone can use it. Consider questions such as: Which proposed projects depend on a particular standard? Which decisions would need reconsideration if a business goal changed? Which requirements remain without an agreed solution? Which accepted designs lack verification evidence?
These should be natural questions the model can answer, rather than specialist spreadsheet exercises. The information model is doing its job when the answer remains clear even after the original architect moves on.
What to resist
Avoid treating every possible relationship as a compulsory field. Avoid repeating an entire external standard inside each requirement. Avoid using delivery-ticket references as substitutes for architecture rationale. And avoid confusing a high percentage of completed metadata fields with trustworthy architecture knowledge.
The MétiArch perspective: Model comprehensively. Capture progressively. Govern proportionately. Rich traceability should be available when needed, not imposed as bureaucracy at the point of first capture.
Read more: Progressive Architecture Traceability: A Practical Information Model for Enterprise Architecture.