# ===================================================================== # Decision Model Ontology (dm:) # A vendor-neutral vocabulary for typed decisions and their execution. # # Layer 1 of 3: # decision-making.ttl general semantics of decisions <- THIS FILE # jev.ttl native Jev / System One vocabulary # inference.ttl semantic bridge between the two # # This file MUST NOT reference any provider vocabulary. It imports # nothing and is usable on its own. # ===================================================================== @prefix dm: . @prefix rdf: . @prefix rdfs: . @prefix owl: . @prefix xsd: . @prefix skos: . @prefix dcterms: . @prefix prov: . @prefix sh: . @prefix vann: . # --------------------------------------------------------------------- # Ontology header # --------------------------------------------------------------------- a owl:Ontology ; owl:versionIRI ; dcterms:title "Decision Model Ontology"@en ; dcterms:description """A vendor-neutral RDF/OWL vocabulary for describing typed decisions -- classifications, ratings, determinations, estimates, extractions and rankings -- independently of the inference backend that executes them; The ontology separates three things that are routinely conflated in AI/ML tooling: 1. WHAT is being decided (dm:Decision and its subclasses) 2. HOW it is executed (dm:DecisionBinding, dm:Executor, dm:Model, dm:Provider) 3. WHAT was decided (dm:DecisionResult) Keeping these apart is what allows one decision definition to be executed by several backends, compared across them, and audited over time without rewriting the decision itself."""@en ; dcterms:creator , ; dcterms:created "2026-09-20"^^xsd:date ; dcterms:license ; vann:preferredNamespacePrefix "dm" ; vann:preferredNamespaceUri "https://www.openlinksw.com/ontology/decision-model#" ; rdfs:seeAlso ; skos:scopeNote """Design rules observed throughout this file: * rdfs:range is declared ONLY on properties whose values are structural nodes minted by this ontology. Properties that point into a user's own ontology (dm:option, dm:about, dm:selectedValue, dm:input, dm:context, dm:inputResource, dm:scoredValue, dm:generatedBy) carry NO range, so that dragging an arbitrary skos:Concept, owl:Class or owl:NamedIndividual into a decision never rewrites the semantics of the term being dragged. * rdfs:domain is declared where the subject is necessarily a class of this ontology, because that entailment is intended and additive. * Only three OWL 2 constructs beyond RDFS are used: owl:disjointWith, owl:propertyChainAxiom and owl:AnnotationProperty. Every mapping that carries reasoning weight stays inside the RDFS-Plus subset supported by mainstream stores, including OpenLink Virtuoso's rule sets."""@en . # ===================================================================== # 1. DECISIONS -- what is being asked # ===================================================================== dm:Decision a owl:Class ; rdfs:label "Decision"@en ; skos:altLabel "decision task"@en ; rdfs:comment """A bounded decision, prediction, classification, estimate or determination that an inference system is asked to make. An instance of dm:Decision is the SPECIFICATION of a question -- not the act of answering it and not the answer. The specification is stable, citable and reusable: it is the unit that is version-controlled, reviewed by domain experts, bound to one or more backends, and compared across them."""@en ; skos:scopeNote "Instances are usually named IRIs in an application namespace, e.g. :ticketCategory. The answer is a dm:DecisionResult, which is disjoint from dm:Decision."@en ; owl:disjointWith dm:DecisionResult . dm:CategoricalDecision a owl:Class ; rdfs:subClassOf dm:Decision ; rdfs:label "Categorical Decision"@en ; rdfs:comment "A decision that selects exactly one item from a finite set of unordered alternatives. The admissible alternatives are given by dm:option, or by dm:optionsFrom when the set is large enough to be better named than enumerated."@en . dm:OrdinalDecision a owl:Class ; rdfs:subClassOf dm:Decision ; rdfs:label "Ordinal Decision"@en ; rdfs:comment "A decision that places its subject on an ordered scale, by selecting a scale value or by estimating a position between scale values. The scale is given by dm:scale (reusable) or by dm:scalePoint (inline)."@en ; skos:scopeNote "Ordinal is not categorical-with-an-order: the answer space of an ordinal decision is an interval of the scale's index line, so a result may legitimately fall between two named levels (e.g. 1.43 on a 0..2 scale). That is why the two classes are declared disjoint."@en . dm:BinaryDecision a owl:Class ; rdfs:subClassOf dm:Decision ; rdfs:label "Binary Decision"@en ; rdfs:comment "A decision that determines a two-valued proposition, preferably reported with the probability that the proposition holds rather than as a bare boolean."@en ; skos:scopeNote """Deliberately NOT modelled as a categorical decision over {true,false}. The useful output of a binary decision is the probability mass on 'true' (dm:probability on the result), which the consuming application thresholds according to the cost of each kind of error. A categorical decision's useful output is the identity of the selected resource."""@en . dm:NumericDecision a owl:Class ; rdfs:subClassOf dm:Decision ; rdfs:label "Numeric Decision"@en ; rdfs:comment "A decision that produces a numeric value, optionally bounded by dm:minValue and dm:maxValue and typed by dm:outputDatatype."@en . dm:MultiLabelDecision a owl:Class ; rdfs:subClassOf dm:Decision ; rdfs:label "Multi-Label Decision"@en ; rdfs:comment "A decision that selects zero or more values from a set of alternatives. Its result may carry several dm:selectedValue statements."@en . dm:RankingDecision a owl:Class ; rdfs:subClassOf dm:Decision ; rdfs:label "Ranking Decision"@en ; rdfs:comment "A decision that orders candidate values. Its result is normally expressed as a set of dm:outcomeScore nodes whose dm:probability or score induces the order."@en . dm:ExtractionDecision a owl:Class ; rdfs:subClassOf dm:Decision ; rdfs:label "Extraction Decision"@en ; rdfs:comment "A decision that identifies or extracts entities or values from the input data rather than choosing among alternatives supplied in advance."@en . # Disjointness is asserted only among the three core types, whose answer # spaces are provably different in kind. MultiLabel, Ranking, Extraction and # Numeric are left non-disjoint on purpose: real tasks legitimately straddle # them (an extraction that is also multi-label; a ranking scored numerically). dm:CategoricalDecision owl:disjointWith dm:OrdinalDecision, dm:BinaryDecision . dm:OrdinalDecision owl:disjointWith dm:BinaryDecision . # --------------------------------------------------------------------- # 1.1 Decision specification properties # --------------------------------------------------------------------- dm:question a owl:DatatypeProperty ; rdfs:label "question"@en ; rdfs:domain dm:Decision ; rdfs:comment "The human-readable question or instruction put to the inference system."@en ; skos:scopeNote "The value is a literal. A plain or language-tagged string suits most decisions; a backend that accepts structured instructions may carry an rdf:JSON literal here."@en . dm:about a owl:ObjectProperty ; rdfs:label "about"@en ; rdfs:domain dm:Decision ; rdfs:comment "The RDF class, property or concept that the decision is about -- the subject matter, not the input instance."@en ; skos:scopeNote "No rdfs:range: the value is a term from the user's own ontology (an owl:Class, an rdf:Property, a skos:Concept or any other resource) and must not acquire a type from being referenced here."@en . dm:option a owl:ObjectProperty ; rdfs:label "option"@en ; rdfs:domain dm:Decision ; rdfs:comment "An admissible alternative for this decision, identified by IRI so that the answer keeps its RDF identity."@en ; skos:scopeNote """No rdfs:range, by design. A dm:option may point at a skos:Concept, an owl:NamedIndividual, an owl:Class or any other resource, which is what lets a user drag terms out of an arbitrary loaded ontology and drop them into a decision. The label and definition of an option are NOT restated here: they are read from the option's own annotations (skos:prefLabel / skos:definition / rdfs:label / rdfs:comment) in whichever ontology governs it. One curated definition therefore serves every decision, every backend and every request built from it."""@en . dm:optionsFrom a owl:ObjectProperty ; rdfs:label "options from"@en ; rdfs:domain dm:Decision ; rdfs:comment "A resource that denotes the set from which the options are drawn -- typically a skos:ConceptScheme, a skos:Collection or an owl:Class -- used when enumerating each option with dm:option would be impractical."@en ; skos:scopeNote "A client expands this to concrete options with a SPARQL query before execution. Enumeration (dm:option) and intension (dm:optionsFrom) may be combined."@en . dm:scale a owl:ObjectProperty ; rdfs:label "scale"@en ; rdfs:domain dm:Decision ; rdfs:range dm:Scale ; rdfs:comment "The reusable ordered scale that this decision is measured against."@en . dm:scalePoint a owl:ObjectProperty ; rdfs:label "scale point"@en ; rdfs:domain dm:Decision ; rdfs:range dm:ScaleValue ; rdfs:comment "A scale value admissible for this decision, stated directly on the decision."@en ; owl:propertyChainAxiom ( dm:scale dm:scaleValue ) ; skos:scopeNote """Shortcut. The property chain makes every value of a referenced dm:Scale a dm:scalePoint of the decision, so the reusable form (dm:scale) and the inline form (dm:scalePoint) answer the same query. The chain is the only inference in this ontology that needs an OWL 2 reasoner; it is a convenience, not a load-bearing mapping. Stores limited to RDFS-Plus can materialise it with: CONSTRUCT { ?d dm:scalePoint ?v } WHERE { ?d dm:scale/dm:scaleValue ?v }"""@en . dm:minValue a owl:DatatypeProperty ; rdfs:label "minimum value"@en ; rdfs:domain dm:Decision ; rdfs:comment "The lowest value the decision may produce, inclusive."@en . dm:maxValue a owl:DatatypeProperty ; rdfs:label "maximum value"@en ; rdfs:domain dm:Decision ; rdfs:comment "The highest value the decision may produce, inclusive."@en . dm:outputDatatype a owl:ObjectProperty ; rdfs:label "output datatype"@en ; rdfs:domain dm:Decision ; rdfs:range rdfs:Datatype ; rdfs:comment "The datatype expected of a literal-valued result, e.g. xsd:decimal or xsd:integer."@en ; skos:scopeNote "Applies to decisions whose answer is a literal (numeric, ordinal position). Decisions whose answer is a resource describe their output semantics with dm:option, dm:optionsFrom or dm:scale instead."@en . dm:input a owl:ObjectProperty ; rdfs:label "input"@en ; rdfs:domain dm:Decision ; rdfs:comment "A description of an input the decision requires: typically an owl:Class whose instances are admissible inputs, or a SHACL node shape they must satisfy."@en ; skos:scopeNote "This is the input TYPE required by the specification. The individual input actually judged is recorded on the result with dm:inputResource."@en . dm:context a owl:ObjectProperty ; rdfs:label "context"@en ; rdfs:domain dm:Decision ; rdfs:comment "An additional resource -- a policy, a named graph, a reference dataset -- that must be supplied alongside the input for the decision to be answerable."@en . # ===================================================================== # 2. ORDERED SCALES -- reusable ordinal semantics # ===================================================================== dm:Scale a owl:Class ; rdfs:label "Scale"@en ; rdfs:comment "An ordered set of named values against which ordinal decisions are measured, e.g. Informational < Low < Medium < High < Critical."@en ; skos:scopeNote "A scale is a first-class, citable resource so that one organisational definition of severity, risk or priority can be shared by every decision, every backend and every report that uses it."@en . dm:ScaleValue a owl:Class ; rdfs:label "Scale Value"@en ; rdfs:comment "One named point on a dm:Scale, carrying its rank via dm:position."@en ; skos:scopeNote "A scale value is frequently also a skos:Concept in a domain scheme; nothing here prevents that, and the two typings are complementary."@en . dm:scaleValue a owl:ObjectProperty ; rdfs:label "scale value"@en ; rdfs:domain dm:Scale ; rdfs:range dm:ScaleValue ; rdfs:comment "A value belonging to this scale."@en . dm:position a owl:DatatypeProperty ; rdfs:label "position"@en ; rdfs:domain dm:ScaleValue ; rdfs:range xsd:integer ; rdfs:comment "The rank of this value within its scale, ascending, conventionally zero-based."@en ; skos:scopeNote "The integer is the ordering key, not a measurement. It is what lets a client serialise the scale in the right order for a backend, and what lets SPARQL sort, threshold and compare results across backends."@en . # ===================================================================== # 3. EXECUTION -- how a decision is carried out # Nothing in this section is part of the intrinsic semantics of a # decision. A decision may have zero, one or many bindings. # ===================================================================== dm:DecisionBinding a owl:Class ; rdfs:label "Decision Binding"@en ; rdfs:comment "An association between one dm:Decision and one execution backend able to answer it. A decision may have any number of alternative bindings; adding, removing or re-pointing a binding leaves the decision definition untouched."@en ; skos:scopeNote "Bindings are the unit of comparison in an evaluation harness: the same decision executed through two bindings yields two dm:DecisionResult sets over the same dm:inputResource, which is exactly the shape a benchmark, a canary rollout or a cost/quality trade-off needs."@en . dm:Executor a owl:Class ; rdfs:subClassOf prov:SoftwareAgent ; rdfs:label "Executor"@en ; rdfs:comment "A system that can execute decisions: a hosted inference API, a local runtime, a SPARQL endpoint, a rules engine or any future backend."@en . dm:Model a owl:Class ; rdfs:label "Model"@en ; rdfs:comment "A concrete model artefact or endpoint-addressable model version used by an executor."@en ; skos:scopeNote "A model is an artefact, not an agent: it is the thing an executor uses. Local files and remote endpoints are equally first-class -- see dm:modelLocation."@en . dm:Provider a owl:Class ; rdfs:subClassOf prov:Agent ; rdfs:label "Provider"@en ; rdfs:comment "The organisation or service that supplies an executor or a model."@en ; skos:scopeNote "This ontology contains NO enumeration of providers. Providers are ordinary individuals minted in whatever namespace their describer controls."@en . dm:DecisionPrimitive a owl:Class ; rdfs:label "Decision Primitive"@en ; rdfs:comment "A named native operation of a backend that realises a kind of decision, e.g. a constrained-choice head, a regression head, a SPARQL ASK."@en ; skos:scopeNote "Declared for providers who wish to model their primitives as individuals. Providers whose primitives are already classes (as Jev's are) annotate those classes with dm:implementsDecisionType instead; both styles are queryable."@en . dm:decision a owl:ObjectProperty ; rdfs:label "decision"@en ; rdfs:domain dm:DecisionBinding ; rdfs:range dm:Decision ; rdfs:comment "The decision this binding can execute."@en . dm:executor a owl:ObjectProperty ; rdfs:label "executor"@en ; rdfs:domain dm:DecisionBinding ; rdfs:range dm:Executor ; rdfs:comment "The system that performs the execution."@en . dm:model a owl:ObjectProperty ; rdfs:label "model"@en ; rdfs:domain dm:DecisionBinding ; rdfs:range dm:Model ; rdfs:comment "The model the executor uses for this binding."@en . dm:provider a owl:ObjectProperty ; rdfs:label "provider"@en ; rdfs:range dm:Provider ; rdfs:comment "The provider responsible for the subject binding, executor or model."@en ; skos:scopeNote "No rdfs:domain: provenance of supply is legitimately stated on a binding, an executor or a model, and declaring one domain would mistype the other two."@en . dm:modelIdentifier a owl:DatatypeProperty ; rdfs:label "model identifier"@en ; rdfs:domain dm:Model ; rdfs:range xsd:string ; rdfs:comment "The identifier by which the provider names this model, e.g. a version string or repository id."@en . dm:modelLocation a owl:DatatypeProperty ; rdfs:label "model location"@en ; rdfs:domain dm:Model ; rdfs:range xsd:anyURI ; rdfs:comment "Where the model is to be found: an absolute URI for a remote endpoint or a relative reference for a file on the executing host."@en ; skos:scopeNote "Relative references such as ./models/ticket-classifier.onnx are valid xsd:anyURI values. Local execution is a first-class case, not a degenerate remote one."@en . dm:framework a owl:DatatypeProperty ; rdfs:label "framework"@en ; rdfs:domain dm:Model ; rdfs:range xsd:string ; rdfs:comment "The runtime format or framework of the model, e.g. \"ONNX\", \"GGUF\", \"PyTorch\"."@en ; skos:scopeNote "A literal, to stay open-ended. Where a controlled resource is preferred, use dcterms:conformsTo on the model in addition."@en . dm:primitive a owl:AnnotationProperty ; rdfs:label "primitive"@en ; rdfs:comment "The backend primitive this binding uses, given as the IRI of the term that names it."@en ; skos:scopeNote "An annotation property because its object is frequently a vocabulary TERM (a class such as jev:Choice) rather than an individual. This keeps the ontology inside OWL 2 DL without punning and guarantees no class-membership entailment follows from naming a primitive. SPARQL sees the triple normally."@en . dm:implementsDecisionType a owl:AnnotationProperty ; rdfs:label "implements decision type"@en ; rdfs:comment "Declares that the subject -- a backend primitive, whether modelled as an individual or as a class -- is capable of realising decisions of the named dm:Decision subclass."@en ; skos:scopeNote """A CAPABILITY claim, and deliberately distinct from rdfs:subClassOf. rdfs:subClassOf answers: what kind of decision is this instance? dm:implementsDecisionType answers: what kinds of decision can this backend execute? The first is instance-level and is what a reasoner acts on. The second is term-level and is what binding validators, capability planners and UIs act on when they offer a user the backends that can run a given decision. Asserting one never implies the other."""@en . # ===================================================================== # 4. RESULTS -- what was decided # ===================================================================== dm:DecisionResult a owl:Class ; rdfs:subClassOf prov:Entity ; rdfs:label "Decision Result"@en ; rdfs:comment "The outcome of executing a decision against one input on one occasion."@en ; skos:scopeNote "Results are modelled separately from decisions so that a decision definition stays stable while an unbounded, timestamped, attributable series of results accumulates against it. That series is the evidence base for calibration studies, drift monitoring and audit."@en . dm:OutcomeScore a owl:Class ; rdfs:label "Outcome Score"@en ; rdfs:comment "One entry of a per-alternative score or probability distribution attached to a result."@en ; skos:scopeNote "Retaining the whole distribution, not only the winner, is what makes a result re-interpretable later: thresholds can be re-tuned, second choices recovered and abstention rates computed without re-running the model."@en . dm:resultOf a owl:ObjectProperty ; rdfs:label "result of"@en ; rdfs:domain dm:DecisionResult ; rdfs:range dm:Decision ; rdfs:comment "The decision that was executed to produce this result."@en . dm:inputResource a owl:ObjectProperty ; rdfs:label "input resource"@en ; rdfs:domain dm:DecisionResult ; rdfs:comment "The resource the result concerns -- the individual that was judged."@en ; skos:scopeNote "No rdfs:range: the judged resource belongs to the user's own ontology."@en . dm:selectedValue a owl:ObjectProperty ; rdfs:label "selected value"@en ; rdfs:domain dm:DecisionResult ; rdfs:comment "A value chosen by this result, identified by IRI."@en ; skos:scopeNote "Points at the option itself (e.g. ex:Technical), never at a copy of its label. The result therefore joins directly to everything the domain ontology already says about that term -- broader concepts, owners, SLAs, translations -- with no string matching anywhere. Repeatable, for multi-label results."@en . dm:numericValue a owl:DatatypeProperty ; rdfs:label "numeric value"@en ; rdfs:domain dm:DecisionResult ; rdfs:comment "The numeric value carried by this result, typically xsd:decimal, xsd:double or xsd:integer."@en ; skos:scopeNote "For an ordinal result this is a position on the scale's index line and may fall between two named scale values; the scale itself is reachable through dm:resultOf/dm:scale."@en . dm:confidence a owl:DatatypeProperty ; rdfs:label "confidence"@en ; rdfs:domain dm:DecisionResult ; rdfs:range xsd:decimal ; rdfs:comment "The executor's certainty in this result, on a 0 to 1 scale where 1 is maximal certainty."@en ; skos:scopeNote """Confidence is NOT the probability of the selected value. It summarises the shape of the distribution -- how peaked it is -- and is used to decide whether to act at all. Probability mass belongs on dm:probability. Keeping the two apart is what supports a three-way policy: act automatically, seek confirmation, or route to a human. Collapsing them is the most common and most expensive modelling error in this area."""@en . dm:probability a owl:DatatypeProperty ; rdfs:label "probability"@en ; rdfs:range xsd:decimal ; rdfs:comment "The probability mass assigned to the outcome described by the subject, from 0 to 1."@en ; skos:scopeNote "No rdfs:domain: used on a dm:OutcomeScore for the mass on one alternative, and directly on a dm:DecisionResult of a binary decision for the probability that the proposition is true."@en . dm:outcomeScore a owl:ObjectProperty ; rdfs:label "outcome score"@en ; rdfs:domain dm:DecisionResult ; rdfs:range dm:OutcomeScore ; rdfs:comment "An entry of the per-alternative distribution behind this result."@en . dm:scoredValue a owl:ObjectProperty ; rdfs:label "scored value"@en ; rdfs:domain dm:OutcomeScore ; rdfs:comment "The alternative -- an option or a scale value -- that this entry scores."@en ; skos:scopeNote "No rdfs:range: the alternative is a term from the user's own ontology."@en . dm:generatedBy a owl:ObjectProperty ; rdfs:label "generated by"@en ; rdfs:domain dm:DecisionResult ; rdfs:comment "The executor or model that produced this result."@en ; skos:scopeNote """No rdfs:range, because the value is legitimately either a dm:Executor (a prov:SoftwareAgent) or a dm:Model (an artefact), and forcing one would mistype the other. Full provenance reuses PROV-O directly rather than duplicating it: dm:DecisionResult is a prov:Entity, so prov:generatedAtTime records when, prov:wasAttributedTo records the responsible agent, and prov:wasGeneratedBy records the activity when a request-level activity is modelled."""@en . # ===================================================================== # 5. SHACL -- optional validation, no reasoning required # Shapes are named (no blank nodes) so they can be cited and reused. # ===================================================================== dm:DecisionResultShape a sh:NodeShape ; rdfs:label "Decision result shape"@en ; sh:targetClass dm:DecisionResult ; sh:property dm:ResultOfConstraint, dm:ConfidenceConstraint . dm:ResultOfConstraint a sh:PropertyShape ; sh:path dm:resultOf ; sh:nodeKind sh:IRI ; sh:minCount 1 ; sh:maxCount 1 ; sh:message "A decision result must cite exactly one decision, by IRI."@en . dm:ConfidenceConstraint a sh:PropertyShape ; sh:path dm:confidence ; sh:datatype xsd:decimal ; sh:minInclusive 0.0 ; sh:maxInclusive 1.0 ; sh:maxCount 1 ; sh:message "Confidence must be a single decimal between 0 and 1."@en . dm:OutcomeScoreShape a sh:NodeShape ; rdfs:label "Outcome score shape"@en ; sh:targetClass dm:OutcomeScore ; sh:property dm:ScoredValueConstraint, dm:ProbabilityConstraint . dm:ScoredValueConstraint a sh:PropertyShape ; sh:path dm:scoredValue ; sh:nodeKind sh:IRI ; sh:minCount 1 ; sh:maxCount 1 ; sh:message "An outcome score must score exactly one alternative, by IRI."@en . dm:ProbabilityConstraint a sh:PropertyShape ; sh:path dm:probability ; sh:datatype xsd:decimal ; sh:minInclusive 0.0 ; sh:maxInclusive 1.0 ; sh:minCount 1 ; sh:maxCount 1 ; sh:message "An outcome score must carry exactly one probability between 0 and 1."@en . dm:ScaleValueShape a sh:NodeShape ; rdfs:label "Scale value shape"@en ; sh:targetClass dm:ScaleValue ; sh:property dm:PositionConstraint . dm:PositionConstraint a sh:PropertyShape ; sh:path dm:position ; sh:datatype xsd:integer ; sh:minCount 1 ; sh:maxCount 1 ; sh:message "Every scale value must carry exactly one integer position."@en . dm:DecisionBindingShape a sh:NodeShape ; rdfs:label "Decision binding shape"@en ; sh:targetClass dm:DecisionBinding ; sh:property dm:BoundDecisionConstraint . dm:BoundDecisionConstraint a sh:PropertyShape ; sh:path dm:decision ; sh:nodeKind sh:IRI ; sh:minCount 1 ; sh:maxCount 1 ; sh:message "A binding must bind exactly one decision, by IRI."@en .