MIRAGE: Runtime LLM Simulation Replaces Static Stubs for Microservice Dependency Testing

arXiv·medium signal

Zhang proposes online LLM simulation where the model directly answers dependency calls at runtime, replacing record-replay, pattern-mining, and specification-driven stubs that generate static artifacts before test execution. This is a paradigm shift for microservice testing — instead of maintaining brittle mock configurations, the LLM generates realistic dependency responses on the fly. Directly applicable to any team struggling with integration test infrastructure.