Architecture Resilience Scorecard

0 of 12 answered

Integration boundaries

How many services or teams call each third-party system (CRM, payments, ERP) directly?

Several, and nobody clearly owns the integrationMostly one, with a few exceptionsOne owning service per vendor; everyone else goes through it

A provider renames a field or changes a contract. What happens?

Several features break and we hear about it from usersWe notice quickly but must change code in several placesOne mapping changes and nothing downstream notices

Do you keep your own stable internal contract for each vendor?

No, vendor types and field names flow through our codeFor some vendorsYes, vendor-specific types stop at the boundary

Reliability under failure

A message or request is delivered twice. What happens?

Duplicates can reach customers, or we are not sureSome handlers are protected, others are notHandlers are idempotent by design and tested for it

A vendor gets slow or goes down. What happens to the rest of the platform?

The slowness spreads and unrelated features sufferWe have timeouts or retries, but no defined degradationOne feature degrades; circuit breakers and timeouts protect the rest

What happens to messages that cannot be processed?

They retry forever or quietly disappearThey reach a dead-letter queue that nobody watches closelyDead-letter alerts, a named owner and a safe replay process

Observability

Can you follow one request across every service and into the vendor call?

No, we cross-reference logs by handPartly; some services share an IDYes, one correlation ID end to end

How do you find out that a provider is failing?

From customers or supportFrom a dashboard someone has to remember to checkFrom alerts on error rate and latency, before customers notice

How long does it take to answer “why did this order or request fail?”

Hours of log digging, or we cannotUnder a dayMinutes

Change safety

How do you move a live integration to a new implementation?

One cutover, and we hopeStaged rollout, but no side-by-side comparisonShadow first, then reads, then writes, with a fallback

How often does a change in one area break something in another?

RegularlyOccasionallyRarely; boundaries are clear and tested

Can you roll back an integration change quickly?

No, it needs a redeploy or downtimeYes, but it is slow or manualYes, there is an instant fallback path

Answer 12 more to see your score