Definition
When a system fractures under ordinary variance (unexpected inputs, refusals, or edge cases) and people have to absorb the impact. It usually means the system has no cushioning at its boundaries, no controlled way to shed load, and little tolerance for people who decline or make mistakes.
Diagnostic
Tests that show it is in place
- Ordinary input variance produces safe fallbacks, not hard failures.
- Operators who patch failures have tooling support and documented playbooks.
Hypothetical example
- A flight booking bot fails on hyphenated surnames, forcing agents to rebook travelers at the counter.