Holonic interaction model¶
Status: design note (conceptual). The vocabulary alignment described here is a planned follow-up; this page fixes the model before any
.ttlis written.
Defining holons is the easy half. The architecture lives in how they interact.
This page models iladub as a small set of holons connected by a governed interaction —
aligned with Kurt Cagle's holonic-graph architecture and its reference ontology,
HGA (holon:), rather than reinventing one. (Zach Welz's Context-Graph Architecture /
CGA remains useful conceptual prior art, but is no longer the alignment target — see the
anchor note below.)
What a holon is (and why interaction is the point)¶
A holon (Koestler) is something that is simultaneously a whole and a part. In the holonic-graph reading, every holon has four orthogonal aspects expressed as named graphs:
- interior — what is true inside (its assertional state);
- boundary — the rules that govern it (SHACL shapes) — a membrane, not a wall;
- projection — what it exposes to others (its outward face);
- context — provenance and history.
A holon never exposes its interior directly. Others see only its projection, and anything crossing the boundary is validated against it. Cagle's framing: the boundary is a Markov blanket — it does not prevent interaction, it governs it, and validation is "the computation of the difference between what arrived and what was expected." Interaction, not definition, is where the work happens.
iladub as interacting holons¶
iladub compiles a document by letting a raw document holon interact with semantic holons (the provided ontologies and terminologies) through a governed portal, producing a clean document holon whose interior is the grounded graph.
flowchart LR
subgraph RAW["RawDocumentHolon (source)"]
DRG["interior:<br/>Document Region Graph<br/>prose · tables · figures · provenance"]
end
subgraph SEM["SemanticHolons"]
ONT["ontologies + SKOS<br/>grounding source"]
end
PORTAL(["GroundingPortal — liminal holon<br/>concept-matching · convergence on shared IRIs"])
DRG -->|"surface concepts / mentions"| PORTAL
ONT -.->|"available concepts"| PORTAL
PORTAL -.->|"candidate groundings proposed back"| DRG
PORTAL ==>|"PromotionDecision<br/>governed membrane-crossing"| MEM
subgraph CLEAN["CleanDocumentHolon (compiled output)"]
MEM{{"boundary:<br/>contract SHACL shapes (the membrane)"}}
MEM -->|"crossed ✔ — assertion"| INT["interior:<br/>grounded graph + promotion decisions"]
MEM -.->|"held at the membrane"| CAND["CandidateConcept<br/>(proposition)"]
INT --> PROJ["projection:<br/>the FAIR graph machines consume"]
end
In one sentence:
A RawDocumentHolon and the SemanticHolons interact through a grounding Portal; concept-matching at that portal is governed by PromotionDecisions at the contract membrane; what passes is assembled into a CleanDocumentHolon whose membrane-health is its cleanliness.
The holon types¶
| Holon | Interior | Boundary (membrane) | Projection | Context |
|---|---|---|---|---|
| RawDocumentHolon | the Document Region Graph (regions, reading order) | well-formedness of the region graph | the surface concepts / mentions it offers | acquisition provenance |
| SemanticHolon (knowledge module) | the ontology / SKOS terminology | what a valid grounded instance looks like | the concepts available for grounding | who curates the vocabulary |
| CleanDocumentHolon | the grounded graph + its promotion decisions | the contract's SHACL shapes | the FAIR semantic graph | provenance-to-region; the decision log |
A SemanticHolon that exists to bridge vocabularies (e.g. SNOMED ↔ a domain concept) is specifically an alignment holon.
The interaction primitives¶
These are what a decision-holon-only model lacks — and what make the model an interaction model:
-
The membrane. A holon's boundary is its SHACL shapes. Validating the interior against the boundary yields a health — Intact / Weakened / Compromised. For the clean document holon this is the document's quality signal: "clean" is not a vibe, it is membrane health.
-
The portal. Interaction between two holons is a first-class object with its own IRI — and a portal is itself a (liminal) holon. The grounding portal carries the concept-matching logic (a transform). "The boundary between knowledge systems is where the real complexity lives, and it deserves first-class status." iladub's convergence principle — table cell, prose, and figure resolving to the same concept IRI — is multiple region-holons reconciled through the portal.
-
Projection. Holons interact through projections, never raw interiors. The clean document holon's projection is the FAIR graph downstream systems consume.
-
Topology. Containment (a holarchy: corpus ⊃ document ⊃ region) is orthogonal to peer adjacency (a holonet: documents that cross-reference each other).
-
Propagation. A change at one holon's boundary is validated, logged to context, updates the interior, and changes the projection the containing holon sees — which then runs its own loop. Re-grounding a region can ripple up to the document and the corpus, "recorded, validated, and traceable" at every level.
Where this meets iladub's existing epistemics¶
iladub already implements the membrane — it just did not name it. The mapping is direct:
| iladub concept | Holonic-interaction reading |
|---|---|
the contract (etkl:SemanticDataContract) + knowledge module |
the boundary / membrane (SHACL priors) |
| the grounded graph | the clean document holon's interior |
a iladub:CandidateConcept (proposition) |
a candidate at the membrane, not yet admitted |
| a grounded assertion | content inside the interior — it crossed the membrane |
a iladub:PromotionDecision (dec:DecisionHolon) |
the governed membrane-crossing — an auditable concept-matching decision |
| convergence on shared concept IRIs | portal reconciliation across region-holons |
| provenance to the page | the context layer |
So the assertion/proposition boundary is simply which side of the membrane a node is on, and which decision moved it there. The promotion decision is the act of crossing.
Alignment, not reinvention¶
There is already a reference-implemented ontology for holon interaction — the CGA
ontology shipped with Zach Welz's holonic library (portals with three subtypes,
membrane health, projection, holarchy vs holonet, a holon type taxonomy, PROV-O
provenance). Its author recommends alignment, not import: keep a standalone,
reasoner-free core and align to other vocabularies rather than forcing everything into
one tree.
iladub follows the same principle it applies to provenance ("don't reinvent"): iladub's
holon types and its grounding portal are defined in the iladub/dec namespaces and
aligned to the W3C Holon CG ontology, not copied from it.
Anchor (settled 2026-06-23): the alignment target is Cagle's W3C Holon CG / HGA (
holon:=http://w3id.org/holon/), not Welz's CGA (urn:holonic:ontology:), which remains useful conceptual prior art. HGA'sholon:DataHolon("information-bearing artefact … produced and consumed by the pipeline") is the natural anchor for iladub's document/semantic holons;holon:Portal(a navigational link) anchors the grounding portal, which adds the concept-matching transform. Note HGA has noholon:Boundaryclass (boundary semantics ride onholon:boundaryMode+ SHACL shapes). SeeCLAUDE.md("Holonic interaction model").
What is built¶
vocab/ontology/etkl-holons.ttl— the holon types + grounding portal + membrane health, in theetklnamespace, standalone (no HGA dependency).vocab/ontology/iladub-hga-align.ttl— the optional HGA alignment (alignment-not-import):rdfs:subClassOf holon:DataHolon/holon:Portalaxioms only.vocab/shapes/iladub-hga-shapes.ttl—iladub:HgaGroundingGovernanceShape: aholon:GroundingRecordmay reachholon:RegisteredStatusonly as the product of aniladub:PromotionDecision(iladub's invariant layered onto HGA's bare confidence gate).examples/holon-grounding-conformant.ttl+tests/holon-grounding-leak.ttl— a conforming governed-grounding traversal and a negative case; exercised bytests/test_hga_alignment.py.
Planned work (not done yet)¶
- A membrane-health check that computes and reports a compiled document's cleanliness
(
etkl:membraneHealth→ Intact / Weakened / Compromised) from validation results. - A full raw→clean traversal example spanning RawDocumentHolon → portal → CleanDocumentHolon (the current example covers the grounding-governance crossing only).
Sources¶
- Kurt Cagle, The Holonic Graph and The World Is Not a Database (The Inference Engineer / The Ontologist, 2026) — the four-layer holon, SHACL-as-priors, the boundary as a Markov blanket, portals as liminal holons.
- Zach Welz,
holoniclibrary and the Context-Graph Architecture (CGA) ontology — the reference implementation: portal subtypes, membrane health, holarchy vs holonet, alignment-not-import. - Arthur Koestler, The Ghost in the Machine (1967) — the holon and the holarchy.