An application-performance management (APM) system manages the performance of a service on a highly redundant high-availability platform that configures duplicate instances of hardware, software, or infrastructure components. The APM system verifies the resilience of the platform by conducting a series of tests that each measure the performance of platform components while simulating a failure path comprising one or more distinct combinations of component failures. If the service is not highly critical, the APM system simulates single-failure failure paths, but more critical services are tested by simulating multiple concurrent failures. Self-learning cognitive modules of the APM system select the failure paths to be tested and then infer from the measurements which failure paths present an unacceptably high risk of service outage. The APM system then directs downstream systems to revise the architecture to mitigate any such risks.