practice

Conformance Automation

Encoding architectural standards as automated checks, so review effort is spent on novel design decisions rather than on verifying known rules.

Architecture review capacity is scarce and is largely consumed by checking things a machine could check: is this service using an approved base image, does it emit the required telemetry, is it registered in the catalogue with an owner, does it have a retention policy, does its API conform to the organisation's conventions.

Automating those checks does two things. It gives the team an immediate answer instead of a scheduled meeting, and it frees the review to address the questions that actually need judgement — is this the right boundary, is the consistency model appropriate, what happens when this dependency is slow.

The mechanisms are ordinary: policy-as-code at admission, fitness functions in the pipeline that assert on dependency direction or coupling metrics, scorecards in the service catalogue reporting conformance per service, and linting on API definitions.

The failure mode to avoid is automating a standard nobody agreed to, which produces a build failure and an argument. Each automated rule should trace to a written, agreed principle, and there must be an exception route with an owner and an expiry — because there will be a legitimate reason to deviate, and a rule with no exception path gets disabled rather than debated.