Statistical Constraints and Verification

• Authors: John Knapman
• Venue: Object Modeling with the OCL, The Rationale behind the Object Constraint Language
• Year: 2002

Statistical constraints have been introduced to UML models to describe their most salient aspects, allowing a natural expression of what is usually the case while tolerating exceptions. They are defined using well-known statistical constructs in terms of OCL collections. They offer more freedom and flexibility than the standard logical quantifiers ('exists' and 'forAll'). This is achieved in a way that is mathematically well formed so that such constraints can be interpreted and verified at run time when a system (represented by a UML model) has been deployed. To make the constraints intelligible to non-IT people, a grammar has been defined that supports more than one syntactic style. The syntax of OCL is supported in addition to other styles that are more accessible to persons without mathematical or computer-science training, and the styles can be mixed. The paper shows examples of statistical constraints, particularly in cases where application systems are being extended by the addition of new capabilities and a new software package for business-to-business (B2B) trading on the Internet. The scenarios involve setting up routing and transformations using a message broker with verification on both extracted and sample data. There are also examples of limiting the complexity of transformations by constraining their definitions.