Why the Triple Is the Wrong Grain

Mike's Notes

This is part of a series of thoughtful opinions about using Ontology in software information systems.

Pipi has an existing BORO Engine (bor) (inspired by Chris Partridge's book) that imports triples to extract entities and relationships, then uses them to build reference relational databases for back-end industry workspaces.

Part two: This is the view of James Lee Stakelum. Discovered via the Ontolog Forum.

Resources

References

  • Reference

Repository

  • Home > Ajabbi Research > Library > Authors > James Lee Stakelum
  • Home > Ajabbi Research > Library > Authors > John F Sowa
  • Home > Ajabbi Research > Library > Subscriptions > Ontolog Forum
  • Home > Ajabbi Research > Research > Program > Ontology
  • Home > Handbook > 

Last Updated

15/08/2026

Why the Triple Is the Wrong Grain

By: James Lee Stakelum
Medium: 13/07/2026

Epistemic architect and creator of the Symbol Grounding Framework, AI operating systems, machine governance languages, and zero‑integration machine‑to‑machine communication.

...

RDF/OWL was built for a metadata document web and lacks the native structure required for automated machine-to-machine AI communication. The Symbol Grounding Framework (SGF) solves this by replacing the fragile “triple” with the “Synapse” — a holistic atomic unit that locks in event context, provenance, and epistemic status natively. By enforcing a closed grammar of 33 primitives, this architecture guarantees deterministic alignment and prevents data from shattering when meaning crosses system boundaries.

...

The Contract

By the end of this essay, you will have seen twelve things the triple cannot do, eleven conservation laws it violates, and one architecture that satisfies all of them. If even one failure is wrong, discard the argument. But if all twelve hold, the triple must be replaced.

Why Triples Were Chosen

In 1999, the W3C standardized RDF. The web was a document web. Triples were small, simple, and easy to parse. They worked for metadata, for catalog descriptions, for linking resources within a single domain.

Twenty‑five years later, machines act on meaning. They drive cars, diagnose diseases, execute trades, coordinate logistics. The triple has not changed.

The Gearbox Analogy

You cannot solve a gearing problem with a bigger engine. You can only solve it with a gearbox. The industry has been building bigger engines — larger models, longer contexts, more parameters. The triple is still the same single gear. Every new relation needs a new predicate, every new domain needs a new ontology, every new system needs new bilateral mappings. The architecture described here is the gearbox: a closed grammar of 33 primitives — 15 roles, 8 link types, 1 generic verb, 4 identity relations — that never grows. The vocabulary can grow infinitely.

First Principles: What Any Meaning‑Transport Architecture Must Guarantee

When meaning crosses a system boundary, the receiving system must be able to:

  1. Identify what the claim refers to, without ambiguity.
  2. Understand who did what to whom, where, when, how, why.
  3. Know where the claim came from and how it was produced.
  4. Know the epistemic status of the claim — fact, inference, guess, or definition.
  5. Preserve the interpretive context — testimony, verdict, hypothetical.
  6. Separate known facts from unknowns.
  7. Refuse claims that violate local policy.
  8. Guarantee that definitions eventually terminate — no infinite regress.
  9. Represent events as first‑class objects.
  10. Distinguish chronological order from narrative order; truth of saying from truth of what was said.
  11. Support deterministic alignment — verifiable proof, not similarity score.

These eleven requirements are conservation laws for meaning transport. Violate any one, and meaning is lost structurally.

The Thesis

The triple violates eight of these eleven laws. Events are shattered, provenance is optional, epistemic status is absent, context is invisible, conflict cannot be represented, order is flattened, alignment is probabilistic.

Each failure forces a specific architectural component into existence. By the twelfth failure, the entire Symbol Grounding Framework (SGF) emerges — not by invention, but by necessity.

The running example: “Beethoven composed the Ninth Symphony in Vienna in 1824.” A simple claim. The triple cannot represent it correctly.

...

Part I: Structural Failures (Failures 1–3)

Failure 1: The triple cannot represent an event as a whole.

RDF/OWL representation (five triples):

:Beethoven rdf:type :Composer .
:NinthSymphony rdf:type :Symphony .
:Beethoven :composed :NinthSymphony .
:NinthSymphony :composedIn :Vienna .
:NinthSymphony :composedInYear "1824" .

Five triples. The verb :composed is an edge label — nothing anchors the event. Location and time are attached to the symphony, not to the composing. The event is shattered. If a second event says :Beethoven :composed :FifthSymphony, the two events share the same agent node but have no structural boundary.

The triple cannot represent an event as a whole. Therefore the architecture must preserve the event as a single object, with the verb at the center and every participant attached through a named role.

SGF provides the Synapse — a hub‑and‑spoke structure. The verb is the hub (the VerbHub). Participants attach through exactly 15 fixed semantic roles.

Synapse: s1
VerbHub: compose
HAS_AGENT: Beethoven
HAS_PATIENT: Ninth_Symphony
HAS_LOCATION: Vienna
HAS_TIME: 1824

One object. The VerbHub is not an edge label. The event is whole.

...

Failure 2: The triple cannot carry its own provenance.

:Beethoven :composed :NinthSymphony .

This triple is structurally identical whether it came from a scholarly biography or a random comment. Named graphs are optional — provenance can be stripped in transit.

The triple cannot carry its own origin. Therefore every claim must carry its own provenance architecturally, not as an external wrapper.

Every Synapse carries a Derivation Tag — part of its header, not optional. The tag records how the claim entered the graph (EXPRESSED, INFERRED, ASSERTED, PROVISIONAL, GHOST) and includes the source document, byte offset, and content hash.

The Artefact Store Invariant: Every Synapse points to its source artefact. A Synapse without provenance is structurally invalid.

...

Failure 3: The triple cannot carry its own epistemic status.

“Beethoven was born in Bonn” (well‑documented). “Beethoven was born in 1770” (inferred from baptismal records). Both are identical triples. The system cannot distinguish a fact from an inference.

The triple cannot carry its authority level. Therefore every claim must carry its epistemic status natively.

SGF provides a 7‑tier epistemic hierarchy: CORE_DEFINITION, CONSTITUTIVE, SOURCED, CLAIMED, INFERRED, PROVISIONAL, GHOST. Two birth‑related Synapses, two different statuses — the system treats them differently.

...

Part II: Grounding Failures (Failures 4–6)

Failure 4: The triple cannot carry its own grounding.

The URI :Beethoven points to a document, not a concept. Is it the composer, the dog, the crater? The URI does not disambiguate. The ontology is external — if the receiver lacks it, the triple is a dangling reference.

The triple cannot carry its own disambiguation. Therefore every identifier must carry its own sense‑level distinction and trace to a shared foundation.

SGF provides the Canonical ID: en.beethoven.ludwig_van_beethoven.person.core. The microgloss ludwig_van_beethoven distinguishes this sense from others. Every Canonical ID has a finite IS‑A path to the Prime Registry — approximately 65 irreducible semantic primes (DO, HAPPEN, SOMEONE, SOMETHING, GOOD, BAD, etc.). Paths terminate at primes. No infinite regress.

...

Failure 5: The triple cannot represent events in the TBox.

RDF/OWL defines “scalpel” as a subclass of cutting_tool with parts blade and handle. It does not capture what a scalpel does — cut living tissue with precision.

The triple cannot define purpose. Therefore definitions must include events using the same structure as assertions.

SGF defines concepts along three axes: IS‑A (what it is), HAS‑PART (what it has), and DESCRIPTIVE_SYNAPSE (what it does). The VerbHub in a definition is identical to the VerbHub in a claim. The Four Theodore Roosevelts prove that events in the TBox are necessary for entity resolution: four men with the same name and overlapping IS‑A paths are distinguished only by events (San Juan Hill, Utah Beach, investment banking, 19th‑century business). Properties alone fail. Events in definitions are required.

...

Failure 6: The triple cannot distinguish truth of saying from truth of what was said.

“The defendant said ‘I am innocent’” — does the system believe the defendant is innocent, or only that he said it? The triple cannot distinguish.

The triple conflates utterance and content. Therefore the act of saying and the content said must be separate Synapses with separate epistemic statuses.

SGF nests the inner claim within the outer claim. The outer is tagged SOURCED (someone said it). The inner is tagged UNVERIFIED (not yet determined as true). The system can assert the utterance without asserting the content.

...

Part III: Relational Failures (Failures 7–9)

Failure 7: The triple cannot preserve conflicting claims.

Source A: Bonn. Source B: Koblenz. Both asserted as triples. The system cannot represent the conflict without averaging.

The Horowitz Paradox: The average of two conflicting truths is often a third, hallucinated lie that neither source supports. The triple cannot avoid this because it has no mechanism to preserve both claims with provenance.

The triple cannot represent contradiction. Therefore the architecture requires explicit typed links between claims, including CONTRADICTS.

SGF provides 8 link types — PRECEDES, CAUSES, ENABLES, SUPPORTS, CONTRADICTS, ELABORATES, SUPERSEDES, DEPENDS_ON. Conflict is recorded explicitly: Link: s_bonn CONTRADICTS s_koblenz. Query‑time Trust Lenses decide which claims to surface.

...

Failure 8: The triple cannot distinguish chronological order from narrative order.

“The king died. He had eaten poison the night before.” First poison, then death — but the story tells death first. The triple cannot represent both chains.

The triple flattens order. Therefore two distinct link types must exist: one for chronological or causal order (PRECEDES), one for story order (NARRATIVE_NEXT).

SGF stores both chains. The system can query the timeline using PRECEDES, or reconstruct the narrative using NARRATIVE_NEXT.

...

Failure 9: The triple cannot represent a generic relational noun without a new predicate.

“Alice has citizenship in France.” Every relational noun (citizenship, employment, ownership) forces a new predicate in RDF/OWL. The grammar explodes.

The triple cannot absorb domain relations without predicate explosion. Therefore a single generic verb must absorb all relational nouns, with the relation type stored as a concept in an attribute role.

SGF provides RELATES_TO. The relation type is a concept, not a role. Concepts are open. Roles are closed.

...

Part IV: Communication Failures (Failures 10–12)

Failure 10: The triple cannot carry its own act type.

“Beethoven composed the Ninth” — is this an INFORM, a QUERY, or a COMMAND? The triple cannot distinguish.

The triple cannot declare communicative intent. Therefore every message must declare its act type from a fixed, closed set.

SGF provides 13 act types: INFORM, ADVISE, REQUEST, QUERY, COMMAND, PROMISE, PROPOSE, ACCEPT, REFUSE, CANCEL, CONFIRM, ACK, ERROR. The act type is declared, not inferred.

...

Failure 11: The triple cannot carry its own admission context.

A claim from one system may be meaningless to another without shared lexicon, provenance, or integrity checks. The triple has no envelope.

The triple cannot carry the context needed for admission. Therefore an envelope must carry identity, payload, lexical hydration material for unfamiliar terms, integrity checks, and freshness data.

SGF provides the Stranger Rule: any non‑core term arrives with enough material for the receiver to hydrate it — a mini‑lexicon entry with an IS‑A link to the shared Core Lexicon. Two systems that have never met can exchange meaning on first contact.

...

Failure 12: The triple cannot support deterministic alignment.

Two ontologies: :Composer vs :MusicalCreator. RDF/OWL alignment is probabilistic — similarity scores, no proof.

The triple cannot support verifiable alignment. Therefore deterministic comparison requires a closed grammar, Canonical IDs, a Prime Registry, and architectural provenance.

SGF’s alignment engine (SOAM) decomposes both concepts into their Synapses, normalizes through the layered lexicon, and performs recursive bisatisfiability over slot fillers. Every comparison produces either a ProofTrace (verifiable record of every comparison) or a GapReport (exactly which slot failed and how to fix it). Deterministic. Verifiable.

...

But What About OWL?

OWL adds formal semantics, class expressions, and inference to RDF. But it operates on the same atomic substrate: the triple. Every OWL axiom is expressed as triples. The fundamental unit is unchanged.

OWL does not:

  • Make provenance architectural.
  • Make epistemic status native.
  • Distinguish saying from said.
  • Guarantee termination (no prime registry).
  • Support deterministic alignment.
  • Prevent the Horowitz Paradox — OWL has no mechanism to store conflicting claims with provenance.
OWL is a powerful ontology language. It is not a fix for the triple’s structural limitations.

What about SHACL? SHACL adds closed‑world validation. But SHACL’s CWA conflicts with OWL’s OWA. The Oudshoorn proof (2026) demonstrates that reconciling them is ExpTime‑complete — the system may never terminate. This architecture uses a consistent CWA. No reconciliation needed.

...

Why Workarounds Fail

Reification shatters events into 7+ triples. Named graphs add a label but do not change the atom. N‑ary relation patterns require inventing new classes and predicates for every event type. Each workaround adds complexity without correcting the grain. The Synapse embeds all of these features natively.

Silence is structurally superior to confident nonsense. If >2% of terms cannot be resolved, the pipeline halts.

The 11 Conservation Laws

Each law: RDF/OWL fails because ___. The architecture succeeds because ___.

  • Identity Conservation: RDF fails (opaque URIs). Architecture succeeds (Canonical IDs).
  • Role Conservation: RDF fails (open predicates). Architecture succeeds (15 closed roles).
  • Provenance Conservation: RDF fails (named graphs optional). Architecture succeeds (Derivation Tags architectural).
  • Frame Conservation: RDF fails (no native frames). Architecture succeeds (7 frame types).
  • Uncertainty Conservation: RDF fails (OWA conflates known and unknown). Architecture succeeds (epistemic status hierarchy).
  • Consequence Conservation: RDF fails (OWL/SHACL conflict). Architecture succeeds (consistent CWA).
  • Trust Conservation: RDF fails (no native trust). Architecture succeeds (epistemic status hierarchy).
  • Grain Conservation: RDF fails (triple is grain). Architecture succeeds (Synapse is grain).
  • Termination Conservation: RDF fails (no prime registry). Architecture succeeds (Prime Registry).
  • Event Conservation: RDF fails (reification needed). Architecture succeeds (VerbHub native).
  • Governance Conservation: RDF fails (SHACL bolts on). Architecture succeeds (Omega native).
The Separation Rule: Provenance is not trust. Trust is not truth. Context is not source. RDF conflates all four. This architecture keeps them orthogonal.

...

The Necessity Chain

The architecture was discovered through adversarial elimination: start with minimal components, test against failures, add components only when a failure cannot be repaired. Repeat until no new failures appear. The Necessity Chain documents 42 failures. The 33 primitives — 15 roles, 8 link types, 7 frame types, 3 composition mechanisms — are the set for which no counterexample could be found.

Remove any primitive, and the system breaks predictably:

  • Remove HAS_AGENT: cannot distinguish “Alice bit Bob” from “Bob bit Alice.”
  • Remove Derivation Tag: cannot distinguish sourced claim from provisional.
  • Remove Prime Registry: cannot guarantee termination.
  • Remove CONTRADICTS: cannot represent conflicting claims.
  • Remove RELATES_TO: every relational noun forces a new role — predicate explosion returns.

Every component is necessary. None are decorative.

...

Conclusion: The Third Protocol

TCP/IP moved bytes. HTTP moved documents. This architecture moves meaning under discipline.

The triple fails 12 specific capabilities. OWL, SHACL, RDF-star, and every workaround operate on the same broken grain. The Oudshoorn proof shows the Semantic Web stack is internally inconsistent.

The Probabilistic Sandwich: The LLM proposes; the gates dispose. The probabilistic layer never touches the file system. The deterministic graph handles storage, reasoning, and governance. The LLM is encased in a deterministic reliability harness.

The choice is binary. You cannot keep the triple and add OWL, SHACL, or any other patch. The triple is the root. Replace the grain.

Between the Babel Tax and the Stranger Rule. Between probabilistic alignment and deterministic verification. Between a stack that violates 8 of the 11 conservation laws and one that satisfies all 11.

This architecture is the Symbol Grounding Framework (SGF) — an open, freely usable architecture for grounded, transportable, verifiable machine meaning. The Synapse is its atomic unit. The grammar is closed. The vocabulary is infinite.

The specification is published. Build something.

You can learn more about the SGF project here:

Contact: JamesLeeStakelum@Proton.me

By ,
Tags , ,

No comments:

Post a Comment

Subscribe to: Post Comments (Atom)