We’re integrating human fallback and want to test that escalation triggers fire correctly.

Simulating failure cases manually is slow and inconsistent.

Anyone found a scalable way to validate fallback logic?