chaos-resilience-testing

Define deterministic scenarios that inject specific faults such as timeouts, malformed payloads, or empty responses using controlled stubs and fake providers. For each scenario, specify the expected behavior, user message, audit evidence, and recovery path before implementation.

Fail-closed applies to security, approvals, payments, secrets, and tenant isolation, where the system must block the action entirely. Degrade-with-recovery applies to optional enrichment or advisory features that may fail but must show clear rationale and recovery guidance.

Start with local deterministic faults: fake providers, injected stores, controlled timers, AbortController, and fixture corruption. Use Playwright route stubs for web degraded states, and reserve Toxiproxy, WireMock, or k6 for integration-level network evidence.

Use Chaos Mesh or LitmusChaos only for Kubernetes-managed services requiring infrastructure-level fault injection. They are not appropriate for npm package or application-level MVP testing, where deterministic local stubs provide faster and more reliable evidence.

Bound every retry with an explicit timeout, retry count, backoff strategy, and budget policy, then capture evidence of the final state and emitted events. A passing command alone is insufficient; record the recovery artifact or blocked gate as proof.