Graph-based Patterns in Inconsistent Declarative Process Models

Declarative process models are widely used in process mining to describe flexible process behavior through sets of constraints. However, models discovered automatically from event logs may contain inconsistent constraints, which can make them difficult to interpret and unusable for execution, conformance checking, or further analysis. Existing methods for consistency analysis either rely on automata-based constructions with high worst-case time complexity or use heuristics based on MIS (minimal inconsistent subsets) that do not provide a full formal characterization of the inconsistency patterns they detect. In this paper, we propose a graph-based approach to the inconsistency analysis for a restricted fragment of Declare process modeling language. We represent dependencies between constraints through the task entailment graph and characterize inconsistency by means of three structural witness types. Based on this characterization, we first detect candidate inconsistent subsets and then verify whether a candidate is a minimal inconsistent subset by dedicated verification procedures. In contrast to automata-based approaches, the proposed method avoids explicit automata products and relies instead on graph-based analysis and constructive trace arguments. We implement the proposed approach and evaluate it on real-life event logs, showing that it is practically feasible and achieves competitive runtime.