Multi-Source Evidence-Guided Test Generation

This dataset is constructed for multi-source program evidence-guided LLM-based test generation. It contains 12 Java subject projects. For each project, the dataset provides the original project files, MuJava-generated mutants, structured evidence files for test generation, LLM-generated JUnit tests, compiled test classes, test execution reports, and mutant-killing results. The main goal of this dataset is to evaluate whether, given a target mutant and its structured program evidence, a large language model can generate a test that reaches the mutation point, triggers behavioral differences, observes the differences, and compiles and executes successfully in the original project environment.