Background: The development of modern organization emphasizes the importance of accurate and comprehensive business process models (BPMs). BPMs serves to provide clear work standards for business actors. Business Process Model and Notation (BPMN) is widely used to model and analyse business processes. However, BPM models in practice often contain missing or inconsistent control-flow links, which reduce model correctness and limit effective analysis. Existing BPM retrieval approaches mainly focus on similarity measurement and provide limited support for explicit missing-link reconstruction. Objective: This study aims to propose a repository-driven approach to detect and reconstruct missing control-flow links in BPMN models while preserving computational efficiency and explainability. Methods: This study employs a quantitative experimental methodology on the use of an application called Repolink., a graph-based technique that transforms BPMN models into directed graphs and computes structural similarity values using Graph Edit Distance combined with semantic weighting. A query BPMN model is compared against a repository of reference BPMN models to identify structural inconsistencies. Missing links are detected using adjacency comparison supported by forward and reverse mappings. Results: The results show that Repolink can detect and reconstruct missing control-flow links in various BPMN structures, including branching and loop-related patterns. It is also able to significantly generate efficient retrieval with an overall time complexity of , where is the number of nodes and is the number of repository models. Compared to existing methods, Repolink provides higher explainability by explicitly reporting missing edges. Conclusion: Repolink effectively supports missing-link reconstruction in BPMN models through a repository-driven and explainable approaches. While the method focuses on structural analysis rather than full behavioural semantics, it offers a practical solution for BPMN conformance checking and model debugging. Keywords: Information Retrieval, Diagram Similarity, Structural Semantic, Graph Edit Distance, Greedy Algorithm