Bounded-Context Representation

Context and Problem Statement

The UK property transaction is modelled as Domain-Driven Design bounded contexts — six primary industry contexts (Estate Agency, Conveyancing, Mortgage Lending, Surveying, Property Data Services, Property Technology), five upstream contexts (HMLR, Local Authority, MHCLG Material Information, Identity & Verification, Trust & Verifiable Claims) and two spanning concerns (Transaction lifecycle; Participants & roles). PDTF is their Published Language (/modelling/bounded-contexts). Two representation questions were open: (1) how to denote a term’s bounded-context membership in the ontology — a separate namespace per context, SKOS, or both; and (2) how to disambiguate a term whose label carries different meanings in different contexts (“how do I say this property as defined in Context A vs in Context B?”).

The answer is constrained by ratified foundations. The single opda: namespace was fixed 9-0 at Session 001 Q7 (ODR-0004) — per-form/per-overlay namespaces were evaluated and rejected as “scatters one concept across many namespaces.” owl:sameAs is forbidden for identity (ODR-0005 Rule 5). Spanning leaves that recur across overlays collapse to one opda: term, with per-context variation pushed onto SHACL overlay profiles (ODR-0008 declare-once-reconcile-overlays; ODR-0010). The decisive empirical fact, surfaced during deliberation: a single entity belongs to many bounded contexts (opda:Address is used across ~10 contexts; one Participant plays role-typed parts in several). Because an IRI inhabits exactly one namespace string, per-context namespaces would force either duplicated classes or owl:sameAs to re-link them — both prohibited.

This ODR was deliberated as a two-round Linked Data Council (session-019; Queen: Kendall (FIBO); Devil’s Advocate: Davis; panel: Evans & Vernon, Gandon, Hendler, Baker, Cagle, Guizzardi). Both rounds returned 7–0: round 1 rejected per-context namespaces and “both”; round 2 confirmed that homonyms are disambiguated by local name, not namespace. It is kind: pattern per ODR-0001 A9 — UFO meta-category (bounded context = a perspectival community of practice, a SKOS-classifiable non-rigid facet), identity criterion over named hard cases (charge, valuation), and an artefact realisation (the SKOS scheme + opda:definedInContext triple shape + the derivation rule).

Considered Options

• Option A (chosen) — Single namespace + SKOS-context-scheme + profile-derived membership + local-name homonym disambiguation. The chosen approach, ratified by Council Session 019 (two rounds, both 7–0).
• Option B — A separate namespace per bounded context. Rejected 7–0: an entity belongs to many contexts, so per-context namespaces force duplicated terms or owl:sameAs (forbidden, ODR-0005); re-confirms the S001 Q7 / ODR-0004 namespace decision.
• Option C — Both (namespaces and SKOS). Rejected 7–0: inherits Option B’s defect for nil benefit; the namespace half is the part that breaks.
• Option D — Hand-authored SKOS membership tags. Rejected as the primary mechanism: re-encodes the overlay→term facts the profiles already carry and drifts the moment an overlay moves context.
• Option E — Per-context namespace “to disambiguate” a homonym. Rejected: an IRI is opaque (RFC 3986) and disambiguates on its whole string — a distinct local name gives a fully distinct identifier already; the namespace adds zero disambiguating power and re-imports the multi-membership break for the non-homonym majority.
• Option F — Build per-context skos:scopeNote / SKOS-XL polysemy machinery now. Rejected as YAGNI: ~0 genuine homonyms are attested; the pattern is ratified but gated (Rule 8).

The diagram maps each rejected option to its fatal flaw.

Decision Outcome

Chosen option: “Single namespace + SKOS-context-scheme + profile-derived membership”, because a bounded context is a perspective on shared terms (not a minting authority), context-of-use is many-to-many (multi-membership is the rule), and per-context namespaces would require the forbidden owl:sameAs to re-link the shared majority of terms.

Adopt a single-namespace + SKOS-context-scheme + profile-derived-membership representation: keep the one opda: minting namespace; model the bounded contexts as a skos:ConceptScheme of skos:Concepts in opda:; derive each term’s context membership from the existing SHACL overlay profiles (single source of truth) rather than hand-authoring it; and disambiguate a term that genuinely means different things in different contexts by minting distinct local names in the one namespace (governed by the Identity-Criterion test below), never by a distinct namespace — chosen because a bounded context is a perspective on shared terms, not a minting authority, so context-of-use is a many-to-many classification (multi-membership is the rule) while a namespace can only encode a single definition-home, and per-context namespaces would require the forbidden owl:sameAs to re-link the shared majority of terms.

Consequences

• Build now. Emit opda:PDTFBoundedContextScheme and the 13 context concepts as SKOS reference data in opda:, reusing opda:hasSteward (Rule 2). Mint opda:definedInContext and opda:servesContext as owl:AnnotationProperty (membership, not logical typing).
• Derive, gated. Author the profile→context SHACL-AF derivation rule (implements: ODR-0017); it ships dormant, switched on only when Rule 8’s gate clears. Re-point opda:overlaysContext (currently a profile-layer IRI) onto the new …Context concepts so derivation has a real target.
• Homonyms. Apply the Rule 4 decision table whenever a same-label/two-meaning candidate appears; build no polysemy scaffolding below the gate. opda:Valuation stays one class until a consumer proves distinct ICs.
• Hand-off. The exact skos:Concept shapes, labels, definitions and steward assignments for the 13 contexts go to ODR-0011’s SKOS steward (concept-shape discipline); the derivation rule follows ODR-0017.
• odr-review lint extension. Flag any opda: term whose IRI encodes a context prefix; flag any rdfs:subClassOf/rdf:type whose object is a context concept; flag any owl:sameAs between homonym siblings.
• Namespace block. Generator output for this ODR’s opda: declarations may carry dct:status "draft" as a publication-grade marker (the namespace string is ratified — greenfield, no WG; inherited ODR-0004 block lifted); the record itself is accepted because the namespace question it answers is already settled by ODR-0004.

More Information

• Methodology: ODR-0001 §What an ODR records (per-kind discipline; A9 kind: pattern).
• Council deliberation provenance: session-019 — Bounded-Context Representation (two rounds; Queen Kendall; DA Davis; panel Evans & Vernon, Gandon, Hendler, Baker, Cagle, Guizzardi; both rounds 7–0).
• Implementation-planning follow-on: session-021 — Bounded-Context Implementation Plan (2026-05-30; Queen Kendall; DA Davis) proposed splitting Rule 8 to un-gate a generated opda:definedInContext home. This was reversed the same day by session-022 — Form↔SHACL Profile Convention (Queen Baker; DA Davis; 6–0), which found definedInContext reinvents rdfs:isDefinedBy + dct:source + dct:subject and retired it (Rules 5 + 8, above). S022 governs; the home is the three standard predicates, community-ownership authored-or-absent under the gate. Amendments council-ratified — greenfield, no WG.
• Foundations: ODR-0004 (single opda: namespace — S001 Q7 9-0; the home this pattern refuses to multiply); ODR-0005 Rule 5 (no owl:sameAs — the reason per-context namespaces fail); ODR-0008 (declare-once-reconcile-overlays — constraint divergence is not term divergence); ODR-0010 (overlay profiles — where context constraints live and the source of derived membership); ODR-0011 (SKOS scheme + steward discipline); ODR-0006 (roles/phases stay one Kind — the boundary that keeps homonym-splitting from swallowing role-modelling); ODR-0002 (vocabulary catalogue — the new scheme is catalogued there); ODR-0017 (the derivation rule’s pattern).
• Industry context map: /modelling/bounded-contexts (the 6 + 5 + 2 contexts and the overlay→context table).
• W3C / spec: SKOS Reference (Miles & Bechhofer 2009) — skos:Concept, skos:ConceptScheme, mapping relations; DCMI Terms — dct:subject, dct:source; RFC 3986 (IRI opacity — meaning is fixed by triples, not by the #// split).
• DDD: Evans 2003 Domain-Driven Design (Bounded Context; Published Language; Context Map; Anti-Corruption Layer); FIBO modular methodology (namespace-by-module-of-definition; homonym co-residence — MarketValue/LiquidationValue).
• Foundational ontology: Guizzardi 2005 Ontological Foundations Ch. 4 (Kinds, Roles, anti-rigidity); OntoClean (Guarino & Welty) — identity-criterion test for counting Kinds.
• Related ODRs: programme anchor ODR-0003.

The graph shows ODR-0019’s declared dependency and implementing relationships.

Rules

These rules are normative for all bounded-context representation in the OPDA ontology.

1. Single namespace — no per-context namespaces

All OPDA-minted terms live in the one opda: (https://w3id.org/opda/#) namespace, per ODR-0004. A bounded context MUST NOT be encoded as an IRI namespace/prefix (estate-agency:, conveyancing:, …). Rationale, load-bearing: an entity belongs to many contexts; a namespace encodes a single home; per-context namespaces therefore force duplication or owl:sameAs (forbidden, ODR-0005 Rule 5). Modularity by ontological concern (the seven TTL module files sharing opda:) is unaffected — that is namespace-by-module-of-definition, which is sound; this rule forbids only namespace-by-context-of-use.

2. Bounded contexts as a SKOS scheme

Each bounded context is a skos:Concept (NOT an owl:Class — context membership is anti-rigid and perspectival; subclassing or typing would weld a perspective into rigid identity), organised by one skos:ConceptScheme in opda:. The scheme reuses the existing opda:hasSteward annotation property to carry steward plurality without namespace plurality.

opda:PDTFBoundedContextScheme a skos:ConceptScheme ;
    skos:prefLabel "PDTF Bounded Contexts"@en ;
    opda:hasSteward "OPDA Architecture WG"@en .

opda:ConveyancingContext a skos:Concept ;
    skos:inScheme opda:PDTFBoundedContextScheme ;
    skos:prefLabel "Conveyancing"@en ;
    skos:definition "DDD bounded context for legal transfer of title."@en ;
    opda:hasSteward "Law Society / SRA"@en .
# … 12 more: EstateAgency, MortgageLending, Surveying, PropertyDataServices,
#   PropertyTechnology (6 industry); the 5 upstream; the 2 spanning.

3. Term→context membership is DERIVED, not hand-authored

Which terms serve which context is derived from the SHACL overlay profiles (the overlay a payload arrives through is its context: piq→Surveying, fme1→Mortgage Lending, ta6→Conveyancing). A generator-emitted SHACL-AF rule (ODR-0017) reads opda:ValidationContext → opda:requires → <term> and materialises <term> opda:servesContext <context> — a generated view, never hand-edited. The profiles are the single source of truth (ODR-0008 declare-once). Note: today opda:overlaysContext targets a profile-layer IRI (profiles/foundation), not an industry context; the derivation rule MUST map the profile-layer target onto the new …Context concepts of Rule 2. Hand-authored dct:subject membership is permitted only for a term that no profile references (a pure domain class with no overlay edge).

4. Homonym disambiguation — the Identity-Criterion decision rule

A shared label is not a shared Universal. Apply Guizzardi’s OntoClean question — “what makes it the same one over time?” — and branch. Identity lives in the local name, never the namespace.

Concept-layer firewall (Baker): two SKOS concept-glosses do NOT compel two OWL classes. When the datatype is shared, stay at the concept layer — case (iii). Climb to two OWL classes only when a reasoner or SHACL must treat them as different Kinds — case (i). Provenance short-circuit (Cagle): the overlay disambiguates at validation time, so a consumer never meets a free-floating homonym on a single path; the profile binds the context-correct term before (i)–(iii) is consulted at runtime.

UFO meta-category of the disambiguated terms (per Guizzardi’s OntoClean analysis — the literal categories that ground the IC test): a valuation is a Relator/Mode (a valuation result individuated by its generating activity and standard) → three Substance Kinds (opda:MortgageValuation / opda:RedBookValuation / opda:AVMValuation, all ⊑ opda:Valuation) only when their lifecycles prove distinct identity criteria to a real consumer (today PDTF emits one opda:Valuation, which is correct restraint); charge spans a Relator (a legal encumbrance over the estate, opda:LegalCharge), a Quality/Mode (a monetary amount, opda:Fee), and a second Relator (the statutory opda:LocalLandCharge) → three Kinds. A bounded context itself is none of these — it is an anti-rigid, perspectival Role-like community of practice, which is exactly why it is a SKOS facet (Rule 2) and never a namespace, a subclass, or an identity-bearing term. Roles and phases are not new Kinds (Rule 8).

5. Term home & provenance — rdfs:isDefinedBy + dct:source (+ gated dct:subject); opda:definedInContext RETIRED

Amended by Council Session 022 (2026-05-30; 6–0; council-ratified — greenfield, no WG) — supersedes the original opda:definedInContext design retained below for history. S022 found opda:definedInContext reinvents three published standards and retired it, decomposing “a term’s home” into three orthogonal axes, each a standard predicate:

• (a) module-of-origin (ontological concern) → rdfs:isDefinedBy → the owning module owl:Ontology IRI (RDFS §2.6). Always emitted; currently 0× in the corpus. NB OPDA’s modules partition by concern (property/agent/claim/…), NOT by the six contexts — so this answers “which module defines it,” not “which community owns it.”
• (b) provenance / source authority → dct:source (+ prov:wasAttributedTo) — already emitted per term (verified: it points at ODR sections + legislation/EUR-Lex/OIDC, never a context).
• (c) community-ownership → dct:subject → a Rule 2 Context concept (DCMI aboutness) — authored-or-absent, NEVER derived (not from dct:source, not from servesContext degree — that would be the OntoClean level-confusion), and gated (Rule 8) to the non-derivable residue (genuine homonym / steward-disputed ownership), empty today.

No bespoke opda:definedInContext predicate is minted; rdfs:isDefinedBy cannot serve the community axis (its object is a defining document, not a community). Council provenance: [session-022], reversing [session-021]‘s “author a generated home for every term” decision (S021 Queen Kendall retracted that chairing). opda:servesContext (usage) remains a derived dormant rule per ODR-0020 Rule 5.

Original S021/S020 design (RETIRED, retained for history): opda:definedInContext (an owl:AnnotationProperty; object = a Rule 2 Context concept) records the context a term’s definition originates in. Per-context definition is rdfs:comment + dct:source; per-context provenance of a payload is reinforced by opda:profileURI on the ValidationContext. “The term as defined in Context A” is answered by triples, never by parsing a prefix.

opda:LegalCharge a owl:Class ; rdfs:subClassOf opda:RegisteredCharge ;
    rdfs:comment "Mortgage instrument securing a debt against the estate."@en ;
    opda:definedInContext opda:MortgageLendingContext ;
    dct:source <https://www.legislation.gov.uk/ukpga/2002/9> .   # LRA 2002

opda:LocalLandCharge a owl:Class ; rdfs:subClassOf opda:RegisteredCharge ;
    rdfs:comment "Encumbrance registered in the Local Land Charges Register."@en ;
    opda:definedInContext opda:ConveyancingContext ;
    dct:source <https://www.legislation.gov.uk/ukpga/1975/76> .  # LLCA 1975

# "the charge AS DEFINED IN Conveyancing" — triples, not a namespace lookup
SELECT ?term ?def WHERE {
  ?term opda:definedInContext opda:ConveyancingContext ; rdfs:comment ?def .
}

6. Cross-term linking — SKOS mapping, never owl:sameAs

Disambiguated sibling terms relate by SKOS mapping — skos:closeMatch (lexical cousins, not identical), skos:relatedMatch (associative), skos:broadMatch (hierarchical) — or a neutral rdfs:subClassOf supertype where an IC is shared. owl:sameAs is forbidden (ODR-0005 Rule 5); instance co-reference uses opda:identifiesSameProperty.

7. Naming convention

Disambiguation is carried by the local name, qualified by sense (opda:registeredCharge vs opda:serviceCharge; opda:buildingRegulationsCompletion vs opda:completionAndMoving — both already in PDTF). Context concepts use the …Context local-name suffix (opda:ConveyancingContext). The namespace never varies to disambiguate.

8. YAGNI activation gate

Restored to its pre-S021 form and refined by Council Session 022 (2026-05-30; 6–0; council-ratified — greenfield, no WG). S021’s “un-gate the bare opda:definedInContext home annotation” carve-out is WITHDRAWN: S022 found definedInContext reinvents three published standards and retired it (see Rule 5). There is no always-emitted home predicate to un-gate — recording a term’s module-of-origin via rdfs:isDefinedBy is ordinary module hygiene (not gated, not polysemy machinery), and community-ownership via dct:subject is authored-or-absent under the gate below.

Roles and phases stay ONE Kind (ODR-0006) — a valuer or a withdrawn valuation is a role/phase, not a new Kind. Context-scoped-definition machinery (per-context skos:scopeNote registries, SKOS-XL label resources, a sense register) is ratified as a dormant pattern but MUST NOT be built until both: (a) ≥3 attested same-label / contradictory-definition collisions exist that distinct local names cannot already separate; and (b) a named consumer needs a community-scoped home that the standard predicates of Rule 5 (rdfs:isDefinedBy + dct:source + a gated dct:subject) cannot already answer. Below that line the standing instruction is: mint two plainly-named terms, write two rdfs:comments, stop. The corpus today attests zero genuine domain homonyms (duplicate-prefLabel: 0; the one same-label/two-definition dictionary hit is free-text boilerplate); PDTF already disambiguates at local-name grain.

Anti-patterns

The flowchart routes a candidate “same term, two contexts” through the decision rule.

• Never put the bounded-context perspective in the IRI namespace (Rule 1).
• Never rdfs:subClassOf a context concept, nor type a term as a context (Rule 2 — context is a SKOS facet, not a Kind).
• Never owl:sameAs between homonym siblings or across contexts (Rule 6).
• Never hand-maintain term→context membership that a profile already implies (Rule 3).
• Never build polysemy machinery below the activation gate (Rule 8).