Mutation testing evaluates test suite quality by injecting intentional faults into source code to verify if tests detect them. Existing engines typically generate mutants by exhaustively applying mutation operators. This creates an implicit blind spot where trivial syntax changes and complex semantic drifts are treated as equally significant. This lack of prioritization wastes computational resources on "easy-to-catch" issues. As a result, the assessment often fails to show whether the test suite is actually capable of catching the most dangerous, subtle bugs that actually matter. We present Argus, a mutation engine that enhances exhaustive syntactic perturbation with intentional semantic exploration. Our key innovation is modeling mutation as a seed-controlled walk over AST locations, enabling the generation of compositional mutation sequences (i.e., structured chains or trees of dependent edits) that simulate complex regressions rather than isolated, trivial shifts. Argus implements this through three core mechanisms: (1) mutations guided by syntax coverage and semantic properties in tandem, (2) dependency-aware steering toward critical APIs, and (3) an interactive viewer for step-wise traces and mutation trajectories. By allowing developers to interpret and reproduce complex mutation paths, Argus provides a transparent assessment of a test suite's capability to detect deep semantic drifts that existing tools overlook.

@inproceedings{YangYWZ26,
  title = {Argus: A Guided and Traceable Mutation Testing Engine},
  author = {Zi Yang and Zhaorui Yang and Jiyuan Wang and Qian Zhang},
  booktitle = {FSE},
  year = {2026},
  doi = {10.1145/3803437.3806410}
}