MR-Coupler: Automated Metamorphic Test Generation via Functional Coupling Analysis

A research team from multiple universities has developed MR-Coupler, a novel AI system that automates the generation of metamorphic tests—a crucial but challenging software testing technique. The tool addresses the "oracle problem" in testing by leveraging functional coupling analysis between methods in source code, then employing large language models to generate candidate test cases. Unlike traditional approaches that require expensive enumeration of method pairs, MR-Coupler uses three functional coupling features to efficiently identify relationships, followed by a validation mechanism that reduces false alarms by 36.56% compared to baseline methods.

In evaluations against 100 human-written test cases and 50 real-world bugs, MR-Coupler demonstrated remarkable effectiveness. It generated valid metamorphic test cases for over 90% of tasks, representing a 64.90% improvement in valid test generation over existing approaches. Most impressively, the automatically generated tests detected 44% of the real bugs in the evaluation set. The researchers have released both the tool and experimental data publicly, supporting broader adoption of metamorphic testing in software development workflows.

The system's architecture combines static analysis with AI generation: first identifying functionally coupled method pairs from source code, then using LLMs to generate candidate test cases, and finally validating them through test amplification and mutation analysis. This approach makes metamorphic testing—previously limited by the need for domain expertise—accessible to developers without specialized knowledge. The research was accepted at ACM FSE 2026, indicating its significance in the software engineering community.