Skip to content

Composing ontologies

A single sound ontology is useful; ontologies that compose are where the network value lives. Use this page when a question, claim, or workflow depends on two or more independently meaningful ontologies, owners, identity layers, or visibility domains.

Knowledge composes upward, like code libraries

Section titled “Knowledge composes upward, like code libraries”
  • Canonical wrefs are import statements — rooted in durable identity and rename-safe. Current committed fields remain exact-version-bound; a relationship that must follow future target versions requires the proposed explicit identity-binding contract rather than documentation fiction.
  • Namespaces make imports partial: a consumer can take your identity layer (*/identity/**) and a different provider’s domain layer — one library for most things, a different one for date handling.
  • Release Sets are version tags; the ontology Shapes are the package manifest: contracts are the API docs; golden cases and competency evaluations are the test suite and the CI badge a prospective consumer inspects before depending on you.
  • Naming contracts and predicate declarations are what make an ontology composable-onto by strangers — the public interface, in both senses (the writer-convergence rule they enforce is in platform consequences).

What a composable ontology exports: charter and non-goals · owning authority · namespaces and identity rules · load-bearing external-identifier semantics · semantic contracts · predicate declarations with direction · provenance guarantees · release identity · the evaluation record · its permitted composition boundary. All of it queryable in */ontology/**.

MechanismMeaningPrimary contract concern
Shared canonical anchorMembers assert about the same canonical identityAuthority, identity stability, reference direction
Shared grounding dependencyMembers independently depend on the same source layerProvenance, rights, cadence, correlated failure
BridgeRival or differently scoped anchors related by attributed mapping assertionsEvidence, contestability, direction, temporal validity
Subscription/derivation dependencyOne ontology consumes another’s release or write streamUpstream identity and rebuildability
Cross-ontology derived assertionA claim whose evidence spans ontologiesMethod, temporal alignment, anti-inferences, the observational line

Bridges deserve their own contracts. The two recurring situations: a private record → a public canonical anchor (RefersToOrganization about Arc(CRM account, public Organization) — for a firm, the single highest-leverage join it will ever build), and one provider’s anchor → a rival’s (SameFacility about Bond{theirs, ours} — the adapter pattern between competing identity layers, and itself curatable knowledge). For any bridge, settle: owning repo and writer · directed or symmetric · evidence class and resolution method · review policy · temporal validity · behavior when either endpoint changes or is deprecated.

Contestability is a feature. One owning repo per canonical Thing within a provider’s ontology — but multiple providers may publish competing canonical layers, consumers choose or combine, and cross-provider Same*/Distinct* assertions preserve disagreement rather than declaring a universal winner. Disagreement between transparent, evidence-carrying layers is knowledge.

Cross-ontology assertions carry the same safeguards as everything else: the precise identities and periods joined, the releases and evidence used, method and version, observation window, anti-inferences — and the observational line holds: a public composition may state a supported co-occurrence; conclusions belong to an owner.

The return path: composition exposes member defects — ambiguous identity, missing contracts, unreachable evidence. The member ontology is fixed through its own process (Stages 0–10) and re-released; the composition re-evaluates. Never patch around a broken member contract with hidden local conventions.

The mirror model (world-model and self-model) is composition’s most valuable instance: an organization joins the public world-model to its private self-model through bridges, and its decision beliefs cite evidence in both. That pedigree-spanning property is queryable — “show me active decisions whose evidence never reaches the self-model” is a hygiene report — and the latency of the join (how fast an upstream change becomes a re-priced, re-staffed decision) is an engineering property an organization can own. Public layers commoditize by design; the interface does not.

The normative core comes first and is not commercial: shared layers are auditable; interpretations are attributed; propositions are contestable; the observational line holds; private beliefs stay private. Given that, composition has natural economics:

  • Namespace design is packaging. If rights were to scope by path, the naming contract would also be the price sheet — free base layers, premium enrichment layers, never-sold private beliefs. The namespace discipline that makes this possible is worth adopting for its query, token, and subscription value regardless.
  • Layer-4 beliefs are the sellable class of knowledge. A curator who composes a free base into predictions and assessments holds something others would pay for — and maintained trust is what they would actually be buying: releases, passing evaluations, correction speed, a measurable track record.
  • Provenance is the receipt. “Why do we believe this?” doubles as “what am I paying for?” The self-documenting ontology is, commercially, the storefront.