TestDecision: RL-Guided Sequential Test Suite Generation Addresses Open-Source LLM Limitations

arXiv·medium signal

TestDecision uses greedy optimization and reinforcement learning to guide open-source LLMs in building comprehensive test suites iteratively — addressing the observed failure of smaller models to generate diverse, high-coverage tests in a single pass. The sequential approach strategically selects each next test case to maximize coverage gain, making open-source LLMs competitive with proprietary models for automated test generation.