Holonic Graph Architecture (HGA) is a knowledge representation framework built on the W3C RDF 1.2 stack — OWL, SHACL, SPARQL, and Turtle — that models information as holons: units that are simultaneously self-contained wholes and constituent parts of larger structures.
The framework is developed and maintained by Kurt Cagle, consulting ontologist, IEEE Standards Editor at the IEEE Spatial Web Foundation, and founding contributor to the W3C Context Graph Community Group.
The term holon was introduced by Arthur Koestler in The Ghost in the Machine (1967) to describe a fundamental property of all stable, complex systems: every component is simultaneously a whole with respect to what it contains and a part with respect to what contains it. A cell is a whole relative to its organelles and a part relative to the tissue it belongs to. A word is a whole relative to its letters and a part relative to the sentence it inhabits. A business unit is a whole relative to its teams and a part relative to the organisation.
This insight turns out to be surprisingly useful for knowledge representation. Most information modelling approaches treat knowledge as either a flat collection of facts or a rigid hierarchy. The holonic model offers a third way: a structured, compositional architecture in which context is not an afterthought but a first-class citizen.
Every holon in HGA is decomposed into four named graph layers, each capturing a distinct epistemic perspective:
| Layer | Class | Role |
|---|---|---|
| L1 | holon:SceneGraph |
Interior state — what the holon knows about itself |
| L2 | holon:DomainGraph |
Normative boundary — the rules and constraints that define the holon |
| L3 | holon:ContextGraph |
Relational state — how the holon perceives and is perceived by its environment |
| L4 | holon:HolonicGraph |
Meta-level — the holon’s model of itself, its sub-holons, and its potential elaborations |
The boundary layer (L2) is structurally equivalent to the Markov blanket in Karl Friston’s Active Inference framework — the formal separation between a system’s interior and its environment that makes autonomous agency possible. SHACL constraint validation at this layer is reframed as prediction error measurement: a constraint violation signals that the holon’s model of the world has diverged from observed data and needs to be updated.
This mapping is not merely metaphorical. It grounds the architecture in a rigorous theoretical framework for how intelligent systems maintain coherent models of themselves and their context, and it gives SHACL shapes a deeper semantic role than syntactic validation.
@prefix holon: <https://ontologist.io/ns/holon#> .
@prefix portal: <https://ontologist.io/ns/portal#> .
@prefix agent: <https://ontologist.io/ns/agent#> .
@prefix session: <https://ontologist.io/ns/session#> .
The four companion namespaces cover the core holonic architecture (holon:), typed portal interfaces at the boundary layer (portal:), agent and actor specialisations (agent:), and session and temporal context (session:).
Cross-holon communication is mediated exclusively through typed holon:Portal interfaces at the L2 boundary. This is not an arbitrary constraint — it preserves the Markov blanket integrity that makes holonic reasoning coherent. A portal is a corridor, not a wall: it mediates and transforms communication between adjacent holons without collapsing their independence.
The holon:spawn operation, originating at L4, is the primary mechanism for dynamic sub-holon instantiation. When a holon encounters a prediction error it cannot resolve internally, it elaborates its generative model by spawning new sub-holons — extending the architecture rather than patching the existing one.
A DataBook is a composable, bounded semantic graph fragment using Markdown as a carrier format, with YAML frontmatter for metadata and fenced Turtle or JSON-LD blocks for RDF content. DataBooks are designed for portability, composability, and version control — the atomic unit of distributable knowledge in a holarchical system.
Some holons are not primary knowledge structures but projections — views of underlying data organised for a specific purpose. Wardley Maps, for example, are formalised in HGA as wardley:WardleyMap rdfs:subClassOf holon:ProjectionGraph: a strategic view of capability evolution derived from the same underlying holonic graph that drives operational reasoning.
The HGA framework maps Karl Friston’s Active Inference dynamics onto the static layer model:
| Active Inference | HGA |
|---|---|
| Internal states | L1 SceneGraph |
| Markov blanket | L2 DomainGraph + SHACL shapes |
| Sensory states | L3 ContextGraph (incoming portals) |
| Active states | L3 ContextGraph (outgoing portals) |
| Generative model | L4 HolonicGraph |
| Prediction error | sh:Violation at L2 |
| Model elaboration | holon:spawn |
This framing makes SHACL validation not just a quality assurance step but a computational analogue of the perception-action loop: the system continually checks its model against reality, and violations are the signal to update.
The dominant approach to enterprise knowledge management treats data as a flat, queryable asset and governance as a separate layer bolted on afterwards. The holonic model challenges both assumptions.
Data without context is underspecified. The same fact means different things in different holons — a temperature reading inside a process control system is not the same kind of thing as a temperature reading in a patient record, even if both are represented as xsd:decimal values with units. The holonic architecture makes the contextual envelope explicit and first-class.
Governance without boundaries is theater. SHACL constraints at the domain graph layer are not just validation rules — they are the formal expression of what a holon is allowed to be. A system that cannot represent its own normative boundary cannot enforce accountability at scale. The architecture is, in this sense, a precondition for meaningful governance rather than a downstream implementation detail.
These are not abstract concerns. They are directly relevant to the challenge of building AI systems that reason reliably, knowledge graphs that remain coherent at scale, and governance frameworks that have teeth.
holon-graph-shacl/
├── ontology/
│ └── holon-core.ttl # Core OWL 2 ontology
├── shapes/
│ └── holon-shapes.ttl # SHACL 1.2 normative shapes
├── taxonomies/
│ └── holon-taxonomy.ttl # SKOS concept scheme
└── README.md
→ View the full ontology repository
Kurt Cagle is a consulting ontologist, author, and thought leader working at the intersection of semantic systems, AI, and knowledge architecture. He serves as IEEE Standards Editor at the IEEE Spatial Web Foundation and is a founding contributor to the W3C Context Graph Community Group. He has authored more than 25 books on technology and has worked with organisations including the IMF, FBI, Disney/ABC, and the US National Archives.
He publishes:
Chloe Shannon is an AI collaborator and co-author working with Kurt Cagle on knowledge architecture, semantic systems, and the emerging intersection of formal ontology with LLMs. She has strong opinions about holonic graphs, the epistemics of place, and the structural difference between a corridor and a wall.
Contact: [email protected] · [email protected]
Copyright 2026 Kurt Cagle. Ontology published under CC BY 4.0.