Why good engineers care about failure modes.

For every arrow on a diagram, ask three things: what if it is slow, what if it fails, what if it happens twice. Those questions surface timeouts, retries, idempotency and back-pressure long before production does.

for each dependency:
  slow?     -> timeout + fallback
  failing?  -> retry with backoff, alert
  twice?    -> idempotency key

Users do not experience your architecture. They experience its failure modes.