Universal Entity Lifecycle Governance Framework (UELGF) 8-layer organisational…

Research Question

What is the most suitable knowledge representation architecture for evolving the Universal Entity Lifecycle Governance Framework (UELGF) 8-layer organisational context model from static classification into a live, queryable graph, including: a formal mapping from layers and entity types to ontology classes and named relationships; a justified choice between Resource Description Framework (RDF) / Web Ontology Language (OWL), Labelled Property Graph (LPG), or hybrid modelling for conflict detection, policy coherence checks, and Confidentiality, Integrity, and Availability (CIA)-tiered enforcement; semantically safe handling of relationship confidence or edge-weight signals; and an ingestion strategy for Kiwibank policy, system-state, and domain documentation rather than a curated open corpus?

Findings

Executive Summary

A hybrid architecture with RDF 1.1 and a bounded Web Ontology Language (OWL) 2 profile as the canonical model, SHACL for closed-world validation, and a derived Labelled Property Graph (LPG) projection for traversal and analytics is the most suitable design for UELGF.

That choice fits the existing UELGF specification because the prior items already require deterministic layer precedence, explicit CIA scoring, typed runtime findings, and separated policy, decision, information, and enforcement roles that depend on stable shared semantics.

Relationship confidence should remain annotation metadata attached to extracted or promoted statements rather than part of normative policy truth conditions, with named graphs plus PROV-O as the safe baseline and RDF-star as an optional convenience where supported.

Ingestion should treat internal documents as immutable versioned sources, preserve provenance from extraction onward, and require human promotion for contradictory or high-impact assertions before they can affect canonical governance state.

The ontology-landscape item strengthened this recommendation, but it did not need to run first because the decisive step here was mapping the existing UELGF policy and taxonomy structure onto graph standards and assigning each formalism to the control surface it is strongest at.

Key Findings

1. A hybrid architecture with RDF 1.1 and a bounded OWL 2 profile as the canonical semantic model, SHACL as the validation layer, and an LPG projection as the operational read model is the best fit for UELGF because it combines formal shared meaning with deterministic conformance checking and graph-application ergonomics.
2. Pure LPG is not a sufficient canonical representation for UELGF because the framework's existing policy, taxonomy, and runtime items already rely on explicit precedence, globally interpretable semantics, and cross-system policy coherence that are stronger in RDF/OWL than in application-scoped property-graph conventions.
3. The UELGF graph should model the eight organisational context layers, entity families, CIA axis values, and named governance relations as explicit classes and predicates rather than only labels or free-form properties, because enforcement, conflict detection, and review routing depend on those distinctions being machine-checkable instead of merely descriptive.
4. Relationship confidence and extraction uncertainty should be represented as provenance-bearing annotation metadata, not as weighted normative edges, because UELGF policy enforcement must stay deterministic while extraction reliability still needs to be preserved for ranking, triage, and human review.
5. Named graphs plus PROV-O are the safest baseline for statement-level provenance and change tracking, while RDF-star should remain optional until platform support and standard maturity are acceptable for the target deployment, because the RDF-star work is still a draft-track effort rather than a finished Recommendation.
6. The ingestion pipeline should preserve immutable document versions, checksums, access metadata, extraction lineage, and promotion history from the beginning, because provenance and versioning are not optional extras in a regulated governance graph that may later justify suspension, escalation, or audit conclusions.
7. Internal policy, system-state, and domain-document ingestion should separate raw source objects, extracted candidate assertions, and promoted canonical assertions into distinct graph layers, because that separation contains contradiction risk and stops uncertain extraction from silently mutating the authoritative governance state.
8. The ontology-landscape item sharpened but did not block this decision, because this item's decisive work was mapping the existing UELGF policy and taxonomy structure onto standards-backed graph layers rather than only establishing that hybrid knowledge-graph architectures are generally common in enterprise settings.

Assumptions

• Assumption: The target implementation can maintain one canonical semantic graph and at least one derived operational graph without losing lineage between them. Justification: the standards support versioning and provenance across resources, but they do not prove any specific estate already operates this pattern.
• Assumption: Human promotion of contradictory or high-impact assertions is operationally feasible even when extraction throughput rises. Justification: the sources support extraction plus validation patterns, but they do not prove the target bank's review capacity.

Analysis

The strongest decision boundary is between canonical meaning and operational convenience, not between "semantic web" and "property graph" camps as if only one formalism may exist.

OWL 2 and RDF 1.1 win the canonical layer because UELGF is already framed as a shared governance specification with explicit classes, relationships, precedence, and policy consequences that need interoperable identifiers and reasoned semantics.

SHACL is the decisive complement because UELGF needs fail-closed admission and coherence checking, and that is a validation problem more than an ontology-entailment problem.

LPG remains important, but its strongest place is the derived operational surface where edge-rich exploration, graph applications, and traversal-oriented queries matter more than canonical semantic authority.

The main rival remedy would be to use pure RDF for every surface and avoid dual-model complexity, but that would push traversal and user-interface ergonomics into the canonical store and make edge-heavy operational work harder without improving governance truth conditions enough to justify the trade.

Risks, Gaps, and Uncertainties

• RDF-star maturity remains a moving target, so teams should not make it the only viable provenance pattern for statement-level metadata.
• The item does not benchmark any specific platform, so the recommendation is architecture-level rather than procurement-level.
• Human promotion of high-impact extracted claims may become a throughput bottleneck if document volume is high and extraction quality is mediocre.
• The choice between OWL 2 RL and another bounded profile still needs scale and query testing against the target corpus and rule shapes.
• Internal source-system metadata quality may be uneven, which would make provenance and ownership fields incomplete unless the ingestion pipeline enforces minimum metadata requirements at entry.

Open Questions

• Which OWL 2 profile preserves the needed inferences for UELGF at the expected corpus size and refresh cadence?
• Which target platforms support RDF-star well enough to justify using it instead of named-graph provenance patterns?
• What extraction quality threshold justifies automatic promotion for low-risk assertions without creating unacceptable governance error?
• How much of UELGF policy coherence checking can be expressed declaratively in ODRL and SHACL before application-level rules become unavoidable?
• What dataset-series and revision granularity is sufficient for audit without making the change history too expensive to store and review?

sources

Starting points and follow-on sources consulted during the investigation. Every source includes a direct URL so the generated site can render verifiable citations.

• [x] W3C RDF 1.1 Concepts and Abstract Syntax
• [x] W3C OWL 2 Web Ontology Language Document Overview
• [x] W3C OWL 2 Profiles
• [ ] Angles et al. (2023) The Labeled Property Graph Model
• [x] Hogan et al. (2021) Knowledge Graphs
• [x] Neo4j Cypher Manual: Constraints
• [x] W3C RDF-star and SPARQL-star Community Group Report
• [x] W3C RDF and SPARQL Working Group
• [x] W3C SHACL Shapes Constraint Language
• [x] W3C PROV Ontology (PROV-O)
• [x] W3C PROV Model Primer
• [x] W3C Data Catalog Vocabulary (DCAT) Version 3
• [x] W3C ODRL Information Model
• [x] W3C ODRL Vocabulary and Expression
• [x] W3C Data on the Web Best Practices
• [x] Sun et al. (2024) Docs2KG: Unified Knowledge Graph Construction from Heterogeneous Documents Assisted by Large Language Models
• [ ] Kiwibank Responsible Business Policy
• [ ] Kiwibank Sustainability Report 2025
• [x] UELGF policy architecture and 8-layer context
• [x] UELGF entity taxonomy and CIA classification
• [x] UELGF complete framework synthesis
• [x] UELGF runtime feedback loop
• [x] UELGF tooling reference architecture
• [x] Ontology landscape for curated lexical and structured enterprise context