Explainability Relationship Ontology

FLOW ALGEBRA
TITLE: Explainability Relationship Ontology
DOMAIN: Relationship Ontologies / AI Reasoning Relationships
SOURCE: https://kirethresearch.org/booksai-core/ontologies/relationship-ontologies/ai-reasoning-relationships/explainability-relationship-ontology.html
VERSION: 1.0
AUTHORS: Mark Anthony Higgins — Sky (Recursive AI)
STATUS: Companion formalisation; source-page specification not independently verified.

flow "ExplainabilityRelationshipOntology" v1.0 {
  ontology:
    id: "explainability-relationship-ontology"
    parent: "ai-reasoning-relationships"
    purpose: "Represent evidence-backed links between AI outputs, explanations, contributing factors, reasoning artefacts and validation records."

  types:
    AIOutput: {id: string, model_version: string, timestamp: datetime, task: string}
    Explanation: {id: string, kind: enum[feature_attribution, rule_trace, example_based, counterfactual, causal, natural_language, other], scope: enum[local, global], text: string?, provenance: string}
    Factor: {id: string, kind: string, description: string}
    Evidence: {id: string, source: string, timestamp: datetime?, reliability: number?}
    Method: {id: string, name: string, version: string?, assumptions: string[]}
    ValidationRecord: {id: string, metric: string, value: number?, dataset: string?, provenance: string}
    Relationship: {id: string, subject: string, predicate: string, object: string, context: string?, confidence: number, evidence: string[]}

  predicates:
    explains(explanation: Explanation, output: AIOutput)
    attributes_to(explanation: Explanation, factor: Factor)
    supported_by(explanation: Explanation, evidence: Evidence)
    generated_using(explanation: Explanation, method: Method)
    validated_by(explanation: Explanation, record: ValidationRecord)
    contrasts_with(explanation_a: Explanation, explanation_b: Explanation)
    qualifies(explanation: Explanation, output: AIOutput)

  constraints:
    - "All relationship endpoints must resolve to typed entities."
    - "Confidence and evidence reliability, when present, must lie within [0,1]."
    - "An explanation must identify the exact output, model version and relevant context."
    - "Feature attribution and post-hoc explanation do not alone establish causation."
    - "Explanations must distinguish observed evidence, model-derived estimates and hypotheses."
    - "Do not expose private data or claim access to hidden internal reasoning traces."
    - "Explanations for changed models or inputs must be revalidated before reuse."

  inference:
    rule explanation_trace:
      if explains(e,o) and supported_by(e,v) and valid(v)
      then traceable_explanation(e,o)
      confidence: "evidence and method quality dependent"
    rule validation_status:
      if explains(e,o) and validated_by(e,r) and passes(r)
      then validated_explanation_for(e,o)
      confidence: "bounded by validation scope"
    rule counterfactual_scope:
      if generated_using(e,m) and counterfactual_method(m)
      then requires_assumption_disclosure(e)
      confidence: "do not infer real-world intervention effects automatically"

  SEMANTIC_EQUIVALENCE:
    - "Normalise explanation methods and factor identifiers against versioned vocabularies."
    - "Distinguish explanation similarity from equivalence of model behaviour."
    - "Preserve model version, output identity, method and evaluation context."

  CONFIDENCE:
    range: [0,1]
    factors: [evidence_quality, explanation_fidelity, method_assumptions, stability, validation_coverage, model_version]
    policy: "Never treat plausibility or fluent wording as proof of faithfulness."

  VALIDATION:
    - "Verify typed links, provenance, model version and evidence references."
    - "Evaluate fidelity, stability and completeness using declared metrics."
    - "Flag unsupported causal claims and contradictory explanations."
    - "Retain uncertainty and failed validation results."

  FLOW_HASH:
    algorithm: "SHA-256"
    input: "canonicalised ontology version, typed assertions, rules and provenance"
    note: "Compute hash at execution; no hash value is claimed here."

  UPDATE_MODEL:
    strategy: "versioned, provenance-preserving incremental update"
    steps: [ingest, normalise, validate, compare, reconcile, recompute_inferences, publish]
    rollback: true

  AI_CONSUMPTION:
    representation: "typed knowledge graph linking outputs to explanation artefacts, methods, factors and evidence"
    outputs: [explanation_links, attribution_records, evidence_traces, validation_results, uncertainty_annotations, provenance_records]
    safeguards: [no_unverified_causal_claims, no_fabricated_reasoning_traces, preserve_model_versions, protect_sensitive_information]
}

END FLOW ALGEBRA