Chaos Engineering
Deliberately injecting failures into a production-like system to verify it actually survives them, instead of hoping it would.
Chaos engineering (popularized by Netflix’s Chaos Monkey) runs controlled experiments that deliberately break things in production or a production-like environment — killing an instance, adding network latency, cutting off a dependency — to verify the system’s failure-handling actually works, rather than trusting an architecture diagram that says it should. It converts "we assume this is resilient" into a tested, repeatable claim.
Worked example: deliberately injecting failures into production — killing instances, adding latency, dropping network — to verify the system degrades gracefully and find weaknesses before a real outage does (Netflix’s Chaos Monkey being the canonical tool). Gotcha: do it with a hypothesis and a blast-radius limit (start small, with a kill switch) — chaos without a steady-state metric to check against and a way to abort is just causing outages; the goal is to LEARN, so an experiment that surfaces no information was run wrong.