AWS Step Functions makes orchestration visible, which creates a sense that failure is handled. The state machine diagram shows retries and catch blocks, so the workflow looks robust. What the diagram doesn't show is whether those policies match how the underlying tasks actually fail.

Add Failure Flags to the Lambda tasks behind your state machine and you can test each step independently. With Gremlin, you can:

• Find out which errors your catch blocks catch. Trigger an exception at a specific step and watch whether the workflow routes to the error branch or falls through.
• Catch tasks that hang rather than error. A stalled task may never trigger a catch block at all. Injecting latency reveals whether anything eventually times out.
• Check your retry policy against a real outage. Counts and backoff intervals chosen for a transient blip become an amplifier when a dependency is genuinely down.

Workflows that sit in a running state for hours don't alert anyone, because nothing errored. Testing turns the diagram into evidence.