practice

Architecture Principle

A durable agreed rule that rules options out, stated with rationale and implications rather than as a slogan.

principlesgovernancestandards

A principle is only real if it forbids something. "We value quality" forbids nothing and is therefore decoration. "Services own their data; no service reads another service's database" is a principle, because you can point at a design and say it violates the rule.

The standard form has four parts: statement, rationale, implications, exceptions and who grants them. The fourth part is the one usually missing, and its absence is why principles decay — an unexceptionable rule gets quietly broken instead of formally waived.

Industry example

Amazon's internal service-interface mandate is the canonical case: teams must expose data and functionality only through service interfaces, with no back-door reads of another team's store, and every interface must be designed as though it were externally exposed.

What makes it instructive is not the rule but its implications, which were accepted rather than discovered. Cross-team joins become network calls, so latency and failure handling get worse. Reporting that used to be one SQL query becomes a data pipeline. Teams must version and support interfaces for consumers they do not control. That cost was paid deliberately in exchange for independent evolution and the eventual ability to externalise those interfaces as products.

A principle whose implications have not been priced is a wish.

Failure scenarios

  • Principles nobody can name. If the team cannot recite them, they are not governing anything.
  • Principles with no waiver path. The first legitimate exception discredits the whole set.
  • Copied principles. "Cloud first" imported from another organisation's context, where it encoded a real constraint that does not apply here.

Trade-off

Strong principles buy consistency and fast decisions — most choices stop being debates. They cost you local optimality: some team, somewhere, is doing something more expensively than they need to because the principle says so. That is usually a good trade, and it is a trade.

Interview question

"Give me an architecture principle you have enforced, the implication that hurt most, and the occasion you granted an exception."