MVCL: An AI-Enabled Multi-View Constraint Language for Formal Inconsistency Detection and Context-Aware Tolerance Reasoning

Multi-view engineering uses heterogeneous models to represent complementary system concerns, but independent model evolution can introduce cross-view inconsistencies. This paper presents MVCL (Multi-View Constraint Language), a domain-specific language for specifying cross-view constraints over heterogeneous model configurations. MVCL separates deterministic inconsistency detection from context-dependent tolerance classification: validated constraints are evaluated over composed model states, and violations are represented as typed inconsistency instances. The executable realization uses Xtext and EVL for language processing and validation, structured JSON for violation exchange, and gpt-5.6-luna as one downstream realization of the abstract tolerance layer. A controlled automotive benchmark evaluates 30 rules across three primary detection configurations and one larger stress configuration. Pooled detection results are 0.920 precision, 0.900 recall, and 0.910 F1, while 75 tolerance cases yield 0.867 accuracy and 0.853 repeated-run consistency. The results support the feasibility of separating formal inconsistency detection from context-dependent tolerance decisions while highlighting the need for broader industrial validation and systematic threshold calibration.