Interaction-Aware Patch Assessment for Multi-Fault Automated Program Repair

Patch overfitting remains a persistent challenge in automated program repair (APR), especially when validation depends on incomplete test suites. We argue that this problem is significantly exacerbated by the overlooked presence of multiple interacting faults, a common yet under-addressed reality in real-world software. Conventional APR tools typically treat faults in isolation, neglecting subtle interactions that can mask faults or introduce regressions. To address this, we develop a taxonomy of five fault interaction-aware patch assessment strategies, supported by a formal model that identifies when and how each should be applied. Our framework guides robust multi-fault repair and exposes how fault interactions critically influence patch outcomes. To our knowledge, this is the first formal treatment of patch assessment and overfitting in multi-fault settings, offering a foundation for more reliable and practical APR.