[DSRP Evidence](https://dsrpevidence.org/) # Is UML Made of DSRP? ## Details **Authors** Cabrera, Derek **Year** 2026 **Publisher** Journal of Systems Thinking **Kind of work** article **Discipline** Computer Science & AI [Read it at the publisher](https://doi.org/10.54120/jost.v26i1.dc6umdd6) 10.54120/jost.v26i1.dc6umdd6 ## In authors' words ### Abstract Unified Modeling Language (UML) is a standardized modeling language whose vocabulary spans structure, behavior, interaction, deployment, and extensibility. This paper asks whether UML introduces any organizational structure not already expressible by DSRP-484. Following the structural-remainder protocol used in companion analyses of system dynamics and network science, we separate information (I) from organization (O), and we replace the earlier practice of nominating constructs by judgment with a pre-registered extraction rule applied to the OMG UML 2.5.1 normative Abstract Syntax Metamodel. The rule yields 242 metaclasses, of which 100 normalize into their ancestors under a purely structural criterion, leaving a closed candidate set of 142. The inclusion filter excludes none of them, because a normative abstract syntax has already been filtered by the standards body. The entire metamodel, 6,785 elements, is assembled from eighteen element types, of which eight are plainly informational. Ruling the remaining construction kit therefore rules UML: Class is identity and part-whole under two undeclared perspectives; Property performs three jobs on one organization, selected by a flag; Association carries two ordered action and reaction ends; Generalization is Distinction and Relationship simultaneously on one link. Twenty further constructs were selected before ruling on the criterion of whether each could break the claim, including conditionality, the temporal metaclasses, ordering, negation, templates, and the metalevel. A separately counted adversarial block was drawn from the Unified Foundational Ontology, whose authors hold that UML requires more primitives rather than fewer; its four load-bearing challenges are rigidity, existential dependence, four irreducible parthood relations, and relators. No residual organizational primitive X was identified anywhere. A second result concerns what the omissions cost: UML has no way to express negation and builds it eleven times, no way to name a point and hard-codes framing in sixty-nine Boolean flags, no Link construct and builds a surrogate for one, and no loop construct while ninety of its own metaclasses sit inside cycles. The result does not make UML unnecessary. It locates UML as a domain-specific modeling language whose higher-order constructs are generated from a more elemental organizational calculus, and whose recurring workarounds are the measurable cost of the structures it declines to name. ### What they set out to do (purpose) Third run in the structural-containment series after system dynamics and network science. Asks whether UML introduces any organizational structure outside the closure generated by DSRP-484, using a pre-registered extraction rule against the normative machine-readable metamodel rather than constructs nominated by the analyst. ### What they found (results) Twenty-eight constructs were ruled individually – nine from which the whole normative metamodel is assembled, plus twenty hard cases selected in advance for their capacity to break the claim, Generalization counted once in both groups – together with a separately counted adversarial block from the Unified Foundational Ontology. No residual organizational primitive X was identified in any of them. Across the closed 142-construct candidate set there are 271 novel structural contributions, and every one is a relationship end (160), a primitive value (96), or an enumeration value (15); there is no fourth kind. Cabrera reports a second, independent result as a census of workarounds: 83 derived properties plus 32 derived unions recomputing severed paths; four metaclasses standing in for the Link the metamodel does not contain; negation hand-built eleven times; framing hard-coded in 69 Boolean flags, 13 enumerations and 674 OCL constraints; and one case where the specification concedes its own formalism cannot express one of its operations. The conclusion is containment, not replacement: UML is located as a domain-specific modeling language composed from the same organizational grammar found in the system dynamics and network runs. ## Commentary ### In short Our reading: this run's contribution over the SD and network papers is methodological. The candidate set is generated by script from a public normative file rather than curated by the analyst, the ruling on each contribution is decided by OMG's own declared types rather than by interpretation, and the falsifier is a count that a critic can re-run. That converts the containment claim from a sample into a census. The workaround census is the second move worth keeping: it makes formalism bias measurable in expenditures rather than arguing it, and Cabrera reads the eleven hand-built negations as a Love-Reality Loop that ran and could not close because the missing structure had no name. ### In more detail Primary nominator: OMG UML 2.5.1 normative specification cross-checked against UML.xmi. Normalization criterion: a metaclass enters the candidate set exactly when it introduces at least one owned attribute or association end that neither redefines nor subsets an inherited property – 242 metaclasses, 100 normalize, 142 remain, inclusion filter excludes none. Two further nominators: Selic's 1998 account of modeling complex real-time systems (retrospective containment check – every construct he names is already in the candidate set) and the UFO/OntoUML adversarial block. Decision rule carried over from the companion papers: a feature is organizational if changing or removing it changes what entities or relations exist or how they are bounded, nested, or ordered; informational if it changes only content or value. Falsifier stated as a procedure: subtract the informational content, and if what remains requires a new element to hold it, that is X – a fifth mediation that is not exists/negates, belongs/contains, affects/effects or frames/is-framed. Measured compression: UML declares 950 features across 242 metaclasses, of which 836 are typed by another metaclass and 114 by a primitive or enumeration – 88 per cent relational vocabulary, 12 per cent information. Appendix A lists all 142 candidates with their contributions and rulings; Appendix B lists the 100 that normalize. **Patterns it shows** D, S, R, P Shows the structure is there, Shows it makes a difference **How to cite this** Cabrera, Derek (2026). Is UML Made of DSRP?. Journal of Systems Thinking.