Chaos & Fault Injection

Goal

Most systems are tested only on the happy path; they fail in production when a dependency does something rude. Fault injection deliberately causes the rude thing — drops the connection, corrupts the write, slows the response — and asserts the system reacts the way it is supposed to: escalate, hard-stop, or roll back. The deliverable is a fault-injection matrix mapping each fault to its expected recovery and a concrete assertion that proves the recovery happened. This skill covers both deterministic in-test injection (the default — reproducible, in CI) and production chaos (live, opt-in, blast-radius-controlled), and is explicit that the latter is risky and must never run unbounded.