Deterministic Workflow Auditing via Dependency–State Coupled Verification and Explainable Conflict Tracing
Jixin Xu, Xingxin Li*, Senlin Zhu, Qingqing Song
Shijiazhuang Campus, Army Engineering University of PLA, Shijiazhuang, China
* Corresponding Author: Xingxin Li. Email: 
Computers, Materials & Continua https://doi.org/10.32604/cmc.2026.083290
Received 01 April 2026; Accepted 05 June 2026; Published online 09 July 2026
Abstract
Complex procedural workflows in maintenance and parametric design often fail due to violated long-range dependencies, unsatisfied preconditions, and inconsistent parameter bindings. In such workflows, an early structural deviation may propagate silently and eventually induce global failure. Existing workflow auditing methods, particularly those based on large language models (LLMs) or sequence matching, often rely on implicit reasoning, exhibit unstable outputs, and provide limited support for reproducible error localization. To address these limitations, we propose a verification-centric auditing framework that couples an explicit typed dependency graph with an executable state transition system. The dependency graph represents data dependencies, precondition dependencies, parameter couplings, and mutex/resource conflicts among workflow steps, while the transition system provides deterministic execution semantics for formal consistency checking. Given a structured audit instance parsed from raw or multimodal evidence, the framework verifies formal consistency through forward execution with incremental satisfiability checking over stepwise constraint stores. When an inconsistency is detected, the framework produces compact and reproducible explanations by combining provenance-guided dependency slicing, UNSAT core or near-minimal conflict-core extraction, and type-prioritized propagation-path tracing. This process identifies both conflict sources and their long-range effects. We evaluate the proposed method on two benchmarks, MyFixit-Audit (text and image, 25,000 samples) and Fusion360-Audit (text, image, and structure, 12,000 samples), covering binary consistency auditing, inconsistency typing, error localization, and structural trace recovery. Experimental results show that the proposed verifier outperforms multimodal fusion and LLM-based baselines, achieving accuracy values of 0.953 on MyFixit-Audit and 0.951 on Fusion360-Audit. In addition, it provides fully deterministic decisions, interpretable explanations, and efficient execution for reliable workflow auditing, with a median runtime of 123 ms per sample.
Keywords
Workflow auditing; formal consistency checking; dependency graph; constraint solving; explainable AI; multimodal understanding
Open Access
Login
Register
Submit a Paper
Propose a Special lssue
Download PDF
Downloads
Citation Tools
