Catch Me If You Can: Detecting Model-Data Inconsistencies in Low-Code Applications

Abstract

Low-Code Development Platforms (LCDPs) offer the benefit of rapid application development, but they sometimes result in inconsistencies while the generated application is in operation. Such inconsistencies often occur despite passing technical validations, indicating that the generated application functions properly without errors. However, issues arise due to semantic discrepancies, leading to conflicting stakeholder perspectives on shared data. The inconsistencies can emerge from model and data co-evolution, but existing inconsistency management techniques, e.g., in databases, multi-view and multi-paradigm modeling, are not well suited to the particular challenges in LCDPs. These approaches are inadequate in this context as they rely on relationships and adherence, such as conformance, which are not applicable in LCDPs. We present a technique and formalization for detecting inconsistencies between various artifacts based on their corresponding rules in low-code applications. We evaluate the correctness of our approach on a domain-specific low-code platform, and assess its scalability, sensitivity to rule mapping complexity, and efficiency with experiments using synthetic data. The results show that the proposed approach is capable of detecting inconsistencies while maintaining a desirable level of efficiency, scalability, and sensitivity.