Inferal Decision Ontology

The Inferal Decision Ontology models traceable decisions without treating a decision, its record, its implementation, and its outcome as the same thing. It connects motivating goals and issues to options, qualified evaluations, deliberation, commitments, documentary records, resulting actions, and anticipated or observed consequences. The model supports architecture decisions, policy choices, governance records, human deliberation, and accountable software-agent decisions while retaining interoperable provenance.

This is an unofficial 0.1 Editor's Draft maintained in the Inferal ontology repository.

Feedback should be sent to contact@inferal.com.

This specification defines three conformance classes: decision data graph, decision publisher, and decision processor.

A decision data graph is an RDF graph that uses the public terms consistently with their definitions. A graph claiming structural conformance MUST satisfy the Decision Ontology SHACL shapes with RDFS inference over this ontology and PROV-O enabled.

A decision publisher is a person or system that emits decision data. It MUST preserve distinct identities for decisions, records, deliberations, actions, and consequences whenever more than one of those things is represented.

A decision processor consumes or traverses decision data. It MUST NOT infer implementation from selection, success from implementation, observation from anticipation, or authority from an unqualified supersession edge.

The key words MUST, MUST NOT, and SHOULD are to be interpreted as described in [[!RFC2119]] and [[!RFC8174]] when, and only when, they appear in all capitals.

RDF and OWL state open-world meaning. SHACL checks structural completeness for explicitly typed focus nodes. No operational algorithm is normative, and inverse triples need not be materialized when an OWL reasoner can derive them.

Introduction

A typical decision record mixes the question, preferred option, commitment, authoring event, implementation, and eventual outcome in one document. That is convenient prose but a poor semantic boundary: changing the document is not necessarily changing the commitment, and accepting an option does not mean it was implemented or successful.

The model draws its process distinctions from the UFO-based core ontology on decision making [[UFO-DECISION]], practical rationale terms from the Design Intent Ontology [[DIO]], and entity/activity/agent provenance from PROV-O [[!PROV-O]]. It is an Inferal interpretation, not an RDF transcription of the UFO conceptual model and not an import or extension of DIO.

Audience

Ontology authors can specialize this module for domain decisions; publishers can exchange machine-readable decision records; validators can detect broken traces; and processors can query why a commitment was made and what happened afterward without fabricating missing lifecycle stages.

Design goals

Separate commitment, documentation, implementation, and outcome.
Keep motivating goals distinct from issues or present situations.
Represent option evaluation as contextual and evidence-bearing.
Separate anticipated consequences from later observations.
Reuse PROV-O for responsibility, time, generation, and derivation.

Scope and non-goals

This ontology owns the domain-neutral lifecycle from a motivating issue and goal through deliberation, decision, record, resulting action, and consequence. A decision is identified as a particular commitment. A materially revised commitment is a new decision; its records may have their own independent revision history.

The ontology deliberately does not:

define a decision-making algorithm or require utility maximization;
claim that evaluations are objective, complete, or mutually comparable;
define universal decision statuses or organizational authority policy;
prove that an action fully implemented a decision or exclusively caused an observation;
treat absent options, evidence, actions, or consequences as known not to exist;
replace domain vocabularies for architecture, law, medicine, policy, or risk.

Competency questions

Which issue and goals motivated the deliberation that produced a decision?
Which options were considered, how were they evaluated, and under which criteria and evidence?
Which option or options did the decision select?
Which information artifact records the decision and its rationale?
Which action implemented the decision, and what consequences were observed?
Which anticipated consequence corresponds to an observed consequence?
Which later deliberation used an observed consequence as feedback?
Which decision supersedes an earlier commitment in a stated governance context?

The canonical example answers all questions except supersession. The boundary example demonstrates supersession without treating it as a global authority rule. The recipes in answer all eight questions while leaving unasserted evaluations and later lifecycle stages unbound.

Namespace and dependencies

Prefix	IRI	Role
`decision`	`https://ontology.inferal.com/modules/decision/`	Decision lifecycle concepts and relations.
`prov`	`http://www.w3.org/ns/prov#`	Entity, activity, agent, generation, use, attribution, association, time, derivation, and value.

The ontology imports PROV-O because decisions, records, options, evaluations, and consequences have entity provenance, while deliberations and resulting actions occur as activities. Publishers SHOULD use prov:wasAttributedTo for responsibility for an entity and prov:wasAssociatedWith for participation in an activity.

DIO is referenced rather than imported. Its design-specific solution, issue, evidence, and justification vocabulary is useful for mappings, but its design scope and its treatment of a design decision do not establish the generic identity boundary required here. The UFO-based model is also referenced rather than imported because its published contribution is a conceptual OntoUML analysis rather than a complete normative RDF distribution.

Conceptual model

A decision:Goal and decision:Issue motivate a decision:Deliberation. The deliberation considers decision:Option resources and contextual evaluations, then may generate a decision:Decision. A decision:DecisionRecord documents that commitment and its rationale. A decision:ResultingAction acts on the decision and may produce an decision:ObservedConsequence. Those observations can inform later deliberations.

flowchart TB
    D["Deliberation"] -->|"pursues
Goal"| G["Goal"]
    D -->|"addresses
Issue"| I["Issue"]
    D -->|"considers
Option"| O["Options"]
    D -->|"has
Evaluation"| E["Qualified
evaluations"]
    G ~~~ I
    I ~~~ O
    O ~~~ E
    D -->|"generated
Decision"| C["Decision /
commitment"]
    R["Decision
record"] -->|"records
Decision"| C
    C -->|"selected
Option"| O
    A["Resulting
action"] -->|"implements
Decision"| C
    O -->|"anticipates
Consequence"| AC["Anticipated
consequence"]
    A -->|"produced
Consequence"| OC["Observed
consequence"]
    D2["Later
deliberation"] -.->|"informedBy
Consequence"| OC

The core decision lifecycle. Edge labels are local names in the decision: namespace, and every arrow follows the RDF subject-to-object direction of the property shown.

Construct	What it is	What it does not imply
Goal	A desired state.	That the current issue already describes that state.
Deliberation	An activity considering issues, goals, and options.	That a decision was reached.
Decision	A commitment selecting or settling an option.	That it was recorded, implemented, or successful.
Decision record	An information artifact documenting decisions.	Identity with the commitment it documents.
Resulting action	An activity acting on a decision.	Complete or correct implementation.
Anticipated consequence	A projected outcome.	That the projection occurred.
Observed consequence	An outcome in an action's provenance trace.	Exclusive or scientifically proven causation.

Motivation, options, and evaluation

A conforming deliberation MUST address at least one decision:Issue, pursue at least one decision:Goal, and consider at least one decision:Option. The same resource MAY be relevant to several deliberations, but its meaning remains contextual to each activity.

An evaluation MUST assess exactly one option under exactly one criterion. Literal assessment content MAY be carried with prov:value. Evidence and evaluator provenance SHOULD be represented with decision:basedOnEvidence, prov:wasDerivedFrom, and prov:wasAttributedTo as appropriate. Structural conformance does not certify the evidence or evaluation as true.

Decision and record identity

A conforming decision MUST identify exactly one generating deliberation and at least one selected option. Every selected option MUST be among those considered by that deliberation. Every cited evaluation MUST belong to the same deliberation. Multiple selections are allowed for decisions that choose a portfolio, set, or compatible combination.

A decision record MUST document at least one decision. It MAY carry a rationale and MAY document multiple related commitments. The record's provenance concerns the information artifact; the decision's provenance concerns the commitment. Publishers MUST NOT use a single IRI for both merely because an ADR page contains the only available representation of the decision.

decision:supersedesDecision states that a newer commitment replaces an earlier one in some applicable governance context. The vocabulary does not choose that context or infer a globally current decision. Supersession cycles are structurally invalid.

Action, anticipation, observation, and feedback

An option MAY identify projected outcomes through decision:anticipatesConsequence. These resources remain anticipations even if similarly described outcomes occur later.

A resulting action MUST identify at least one decision on which it acts. An observed consequence MUST be generated by at least one resulting action. decision:correspondsToAnticipation records a judged correspondence without collapsing projection and observation or proving exclusive causation. A later deliberation MAY use an observed consequence explicitly as feedback.

Term reference

The public vocabulary is deliberately small and shallow. Its classes distinguish resources by identity and lifecycle, while its properties connect those resources into a trace. Domain ontologies can specialize the classes without changing the commitment, record, activity, and consequence boundaries established here.

The class and property definitions below are normative together with the model sections that precede them. Labels such as “decision” or “record” are not permission to collapse adjacent resources; the positive and negative boundaries in the conceptual model continue to apply to every specialization.

Classes

Term	Superclass	Normative definition
`decision:Goal`	`prov:Entity`	A desired state motivating deliberation, not the issue being addressed.
`decision:Issue`	`prov:Entity`	A question, problem, or situation addressed by deliberation.
`decision:Option`	`prov:Entity`	An alternative considered during deliberation.
`decision:Criterion`	`prov:Entity`	A contextual dimension or rule under which an option is evaluated.
`decision:Evaluation`	`prov:Entity`	A qualified assessment of one option against one criterion.
`decision:Deliberation`	`prov:Activity`	An activity considering issues, goals, options, and evaluations.
`decision:Decision`	`prov:Entity`	A commitment generated by deliberation that selects or settles options.
`decision:DecisionRecord`	`prov:Entity`	An information artifact documenting decisions and possibly rationale.
`decision:Rationale`	`prov:Entity`	An information entity explaining why a decision was made.
`decision:ResultingAction`	`prov:Activity`	An activity undertaken to act on a decision.
`decision:Consequence`	`prov:Entity`	A situation or outcome related to acting on an option or decision.
`decision:AnticipatedConsequence`	`decision:Consequence`	A projected consequence, not evidence of occurrence.
`decision:ObservedConsequence`	`decision:Consequence`	A consequence observed in a resulting action's provenance trace.

Properties

RDFS domain and range statements infer types; they are not input filters. Cardinality and cross-resource consistency are imposed by the SHACL validation regime, not by OWL.

Term	Domain → range	Normative role
`decision:addressesIssue`	Deliberation → Issue	The issue addressed by a deliberation.
`decision:pursuesGoal`	Deliberation → Goal	A desired state motivating a deliberation.
`decision:considersOption`	Deliberation → Option	An alternative considered, not necessarily selected.
`decision:hasEvaluation`	Deliberation → Evaluation	A contextual evaluation used by the deliberation.
`decision:evaluatesOption`	Evaluation → Option	The single option assessed.
`decision:appliesCriterion`	Evaluation → Criterion	The single criterion applied.
`decision:basedOnEvidence`	Evaluation → prov:Entity	Evidential input without a sufficiency or truth claim.
`decision:generatedDecision`	Deliberation → Decision	A commitment generated by the activity; inverse of `resultedFromDeliberation` in the OWL addendum.
`decision:resultedFromDeliberation`	Decision → Deliberation	The activity that generated the commitment.
`decision:selectedOption`	Decision → Option	An option selected or settled without implying implementation.
`decision:basedOnEvaluation`	Decision → Evaluation	An evaluation that informed the commitment.
`decision:recordsDecision`	DecisionRecord → Decision	A decision documented by the information artifact.
`decision:hasRationale`	DecisionRecord → Rationale	An explanation preserved by a decision record.
`decision:supersedesDecision`	Decision → Decision	An earlier commitment replaced in an applicable context.
`decision:anticipatesConsequence`	Option → AnticipatedConsequence	A projected outcome associated with an option.
`decision:implementsDecision`	ResultingAction → Decision	A commitment on which an action acts.
`decision:producedConsequence`	ResultingAction → ObservedConsequence	An observed consequence generated in the action's provenance trace.
`decision:correspondsToAnticipation`	ObservedConsequence → AnticipatedConsequence	A judged correspondence that does not assert identity.
`decision:informedByConsequence`	Deliberation → ObservedConsequence	An observed outcome used as feedback in a later deliberation.

Validation and entailment

The SHACL regime targets explicitly typed instances of every public class. It requires the minimum structural spine of deliberations, evaluations, decisions, records, actions, and observed consequences. It also rejects options or evaluations that do not belong to the generating deliberation and rejects cyclic decision supersession.

Shapes for goals, issues, criteria, rationale, and consequences validate resource kind while leaving domain content extensible. Shape conformance demonstrates structural consistency, not rationality, authority, evidence quality, implementation completeness, outcome desirability, or causal truth.

RDFS domains, ranges, subclasses, and subproperties may infer PROV and decision types. The optional OWL addendum additionally types properties and states that decision:generatedDecision is inverse to decision:resultedFromDeliberation. It defines no functional properties, equivalence classes, closed enumerations, or lifecycle cardinalities.

Query recipes

These informative SPARQL recipes operate on asserted facts. They do not turn absence into negation, validate the graph, rank options, or select an authoritative decision. Publishers SHOULD validate input separately when the structural guarantees of this specification are required.

Trace a decision lifecycle

Starting from a decision, this recipe returns its motivating goal and issue, selected option, record, action, and observed consequence. Missing records, actions, or observations remain unbound rather than being inferred from the commitment.

PREFIX decision: <https://ontology.inferal.com/modules/decision/>

SELECT ?decision ?goal ?issue ?selectedOption ?record ?action ?observedConsequence
WHERE {
  # Anchor the row in the commitment and the deliberation that generated it.
  ?decision a decision:Decision ;
      decision:resultedFromDeliberation ?deliberation ;
      decision:selectedOption ?selectedOption .
  # Goals and issues explain why the deliberation occurred.
  ?deliberation decision:pursuesGoal ?goal ;
      decision:addressesIssue ?issue .
  # A decision may exist without a record in the available graph.
  OPTIONAL { ?record decision:recordsDecision ?decision . }
  # Implementation is optional; an observation is optional within that stage.
  OPTIONAL {
    ?action decision:implementsDecision ?decision .
    OPTIONAL { ?action decision:producedConsequence ?observedConsequence . }
  }
}

Compare considered options and their evidence

This recipe emits every considered option, including an option with no asserted evaluation. Assessment, evidence, and selection are contextual facts; an unbound value means only that the input graph does not provide it.

PREFIX decision: <https://ontology.inferal.com/modules/decision/>
PREFIX prov: <http://www.w3.org/ns/prov#>

SELECT ?deliberation ?option ?evaluation ?criterion ?assessment ?evidence ?selectedBy
WHERE {
  # Begin with every considered option so options lacking assessments remain visible.
  ?deliberation a decision:Deliberation ;
      decision:considersOption ?option .
  # Join only evaluations that belong to this deliberation and assess this option.
  OPTIONAL {
    ?deliberation decision:hasEvaluation ?evaluation .
    ?evaluation decision:evaluatesOption ?option ;
        decision:appliesCriterion ?criterion .
    OPTIONAL { ?evaluation prov:value ?assessment . }
    OPTIONAL { ?evaluation decision:basedOnEvidence ?evidence . }
  }
  # Selection is independent of whether an evaluation is asserted.
  OPTIONAL {
    ?selectedBy a decision:Decision ;
        decision:resultedFromDeliberation ?deliberation ;
        decision:selectedOption ?option .
  }
}

Connect projections, observations, and feedback

This recipe keeps projected and observed consequences in separate columns. A correspondence is returned only when it is explicitly asserted, and a later deliberation appears only when it explicitly uses the observation as feedback.

PREFIX decision: <https://ontology.inferal.com/modules/decision/>

SELECT ?decision ?selectedOption ?anticipated ?action ?observed
       ?matchedAnticipation ?laterDeliberation
WHERE {
  # Selection provides the option whose projected consequences may be available.
  ?decision a decision:Decision ;
      decision:selectedOption ?selectedOption .
  OPTIONAL { ?selectedOption decision:anticipatesConsequence ?anticipated . }
  # Do not manufacture an implementation merely because the option was selected.
  OPTIONAL {
    ?action a decision:ResultingAction ;
        decision:implementsDecision ?decision .
    OPTIONAL {
      ?action decision:producedConsequence ?observed .
      OPTIONAL {
        ?observed decision:correspondsToAnticipation ?matchedAnticipation .
      }
      # Feedback exists only when a later deliberation explicitly uses the observation.
      OPTIONAL {
        ?laterDeliberation a decision:Deliberation ;
            decision:informedByConsequence ?observed .
      }
    }
  }
}

Traverse asserted supersession

The caller MUST select an input graph representing the applicable governance context. The recipe returns direct and transitive newer-to-earlier pairs within that graph. A decision with no asserted successor is not thereby globally current or authoritative.

PREFIX decision: <https://ontology.inferal.com/modules/decision/>

SELECT ?newerDecision ?earlierDecision ?direct
WHERE {
  # The + path returns one-or-more asserted supersession steps.
  ?newerDecision a decision:Decision ;
      decision:supersedesDecision+ ?earlierDecision .
  # Mark whether this pair also has a direct edge, not merely a transitive path.
  BIND(EXISTS {
    ?newerDecision decision:supersedesDecision ?earlierDecision .
  } AS ?direct)
}

Examples

@prefix decision: <https://ontology.inferal.com/modules/decision/> .
@prefix ex: <https://example.com/decision/canonical/> .
@prefix prov: <http://www.w3.org/ns/prov#> .

# The goal is the desired state; the issue is the question being decided.
ex:goal a decision:Goal .
ex:issue a decision:Issue .
# An option may project a consequence without claiming that it occurred.
ex:managed a decision:Option ;
    decision:anticipatesConsequence ex:less-operations-work .
ex:less-operations-work a decision:AnticipatedConsequence .

# The deliberation links motivation to the alternative it considered.
ex:deliberation
    a decision:Deliberation ;
    decision:pursuesGoal ex:goal ;
    decision:addressesIssue ex:issue ;
    decision:considersOption ex:managed .

# This is the commitment, not the selected option or a document about it.
ex:commitment
    a decision:Decision ;
    decision:resultedFromDeliberation ex:deliberation ;
    decision:selectedOption ex:managed .

# The ADR documents the commitment while retaining separate identity.
ex:adr
    a decision:DecisionRecord ;
    decision:recordsDecision ex:commitment .

# Implementation is a separate activity that must be asserted explicitly.
ex:deployment
    a decision:ResultingAction ;
    decision:implementsDecision ex:commitment ;
    decision:producedConsequence ex:measured-work .

# Correspondence links observation to projection without making them identical.
ex:measured-work
    a decision:ObservedConsequence ;
    decision:correspondsToAnticipation ex:less-operations-work .

@prefix decision: <https://ontology.inferal.com/modules/decision/> .
@prefix ex: <https://example.com/decision/evaluation/> .
@prefix prov: <http://www.w3.org/ns/prov#> .

ex:managed a decision:Option .
ex:cost a decision:Criterion .
ex:estimate a prov:Entity .
ex:reviewer a prov:Agent .

# The evaluation qualifies one option by one criterion. Evidence, literal
# assessment content, and responsible agent remain separate provenance facts.
ex:evaluation
    a decision:Evaluation ;
    decision:evaluatesOption ex:managed ;
    decision:appliesCriterion ex:cost ;
    decision:basedOnEvidence ex:estimate ;
    prov:value "acceptable" ;
    prov:wasAttributedTo ex:reviewer .

@prefix decision: <https://ontology.inferal.com/modules/decision/> .
@prefix ex: <https://example.com/decision/supersession/> .

ex:goal a decision:Goal .
ex:issue a decision:Issue .
ex:a a decision:Option .
ex:b a decision:Option .
# Each decision still selects an option considered by its own deliberation.
ex:first-deliberation a decision:Deliberation ;
    decision:pursuesGoal ex:goal ; decision:addressesIssue ex:issue ;
    decision:considersOption ex:a .
ex:second-deliberation a decision:Deliberation ;
    decision:pursuesGoal ex:goal ; decision:addressesIssue ex:issue ;
    decision:considersOption ex:b .
ex:first a decision:Decision ;
    decision:resultedFromDeliberation ex:first-deliberation ;
    decision:selectedOption ex:a .
# Supersession replaces the earlier commitment only in the applicable context.
# It does not establish a universal or timeless "current" decision.
ex:second a decision:Decision ;
    decision:resultedFromDeliberation ex:second-deliberation ;
    decision:selectedOption ex:b ;
    decision:supersedesDecision ex:first .

The Decision Ontology SHACL regime rejects this graph because the decision selects ex:b while its generating deliberation considered only ex:a.

@prefix decision: <https://ontology.inferal.com/modules/decision/> .
@prefix ex: <https://example.com/decision/invalid/> .

ex:goal a decision:Goal .
ex:issue a decision:Issue .
ex:a a decision:Option .
ex:b a decision:Option .
# The deliberation considered only ex:a.
ex:deliberation a decision:Deliberation ;
    decision:pursuesGoal ex:goal ;
    decision:addressesIssue ex:issue ;
    decision:considersOption ex:a .
# This selection of ex:b is the deliberate near-miss rejected by SHACL.
ex:decision a decision:Decision ;
    decision:resultedFromDeliberation ex:deliberation ;
    decision:selectedOption ex:b .

Interoperability

Domain ontologies SHOULD specialize this core rather than redefine its terms. Architecture-decision records can specialize decision:DecisionRecord; change-governance decisions can connect their domain-specific proposal and outcome structures to a generic decision trace.

DIO mappings require review rather than automatic OWL equivalence. Its design solutions overlap with options, and its design decisions overlap with decisions, but the lifecycle and identity commitments are not bidirectionally identical. No class in this ontology is equivalent to a DIO or UFO term.

Security and privacy considerations

Decision traces may expose confidential alternatives, rejected plans, evidence, responsibility, organizational disagreement, and observed outcomes. Publishers SHOULD apply access controls and data minimization to records, rationale, evidence, attribution, and consequence links. Removing sensitive provenance creates an incomplete graph; processors MUST NOT interpret that absence as evidence that no agent, evidence, or alternative existed.

Evolution and governance

Public IRIs in the Decision Ontology are stable. Future releases will deprecate and map adopted terms rather than silently move or redefine them. Domain-specific status vocabularies, authority policies, preference calculi, and DIO alignments remain extension or mapping concerns and require separate semantic review.