---
status: accepted
date: 2026-06-01
updated: 2026-09-05
kind: architecture
tags: [classification, inheritance, roles, skos, facets, ufo, doctrine]
scope: []
supersedes: []
depends-on: [ODR-0005, ODR-0006, ODR-0008, ODR-0009, ODR-0011, ODR-0025, ODR-0026]
implements: [ODR-0003]
---

# Classification, Roles, Inheritance, and SKOS — opda Modelling Doctrine

> **Scope amendment — 2026-09-05.** The accepted status and original body remain historical evidence of the schema-derived model's classification and evidence changes. The SPDTF method follows the selected source ODRs under [ADR-0063's 2026-09-05 amendment](../../adr/ADR-0063-domain-led-bounded-context-working-groups.md); the [modelling decision basis](/semantic-modelling/decision-basis) identifies their scope. In particular, this record's R5 “multi-domain reads as union” convention is superseded for the SPDTF method and must not be presented as standard RDFS semantics.
>
> Source ODR-0014 retired that convention on 2026-08-28. RDFS domain/range statements retain their standard entailment meaning; repeated values are conjunctive, whether or not a runtime materialises the consequences. Under its 2026-08-30 amendment, each governed property's subject and value applicability must each use exactly one mechanism family: a universally true single RDFS commitment, Schema.org inclusion hints, or an explicit governed waiver. SHACL supplies the relevant validation constraints. This notice preserves the historical rationale without claiming that existing emitted data or validators have been migrated.

## Context and Problem Statement

opda's classification practice had drifted into imprecise vocabulary and an inconsistent mechanism. Council session-036 codified an OntoClean "load-bearing cascade" in ODR-0011 §8a but (a) used "facet" loosely to mean "a coded-value classification," conflating two distinct ideas, and (b) concluded that the evidence sub-types should stay `rdfs:subClassOf opda:Evidence` even though opda's own model calls evidence "a role a document plays." The directing authority directed opda to **adopt an established classification-over-inheritance modelling doctrine wholesale** (the prior-art source is recorded in §More Information) and to record it completely — facets, roles, inheritance, SKOS, and the boundaries between them.

That prior-art corpus settles this with a set of ratified, mutually-reinforcing decisions: multi-faceted classification (classification axes are SKOS-backed coded annotations — a body of work that *fixed 13 `rdfs:subClassOf` hierarchies that incorrectly modelled Roles as Kinds*); domain/range are documentation, never evaluated, SHACL enforces; every enumeration is a SKOS concept scheme, never an OWL subclass enumeration; roles via a `roleOf` relation, not subclassing; a role-specific property is *borne by* the role, not the Kind; and UFO Role/Phase stereotypes are SKOS concept schemes ("encode as OWL classes with subclass relationships" was rejected). This ODR is opda's adoption record and single doctrinal home for the distinctions; the specific prior-art records are cited in §More Information.

## Considered Options

* **Option A (chosen) — Adopt the classification-over-inheritance doctrine wholesale** (from established prior art, §More Information). Classification by coded set-membership by default; `rdfs:subClassOf` only for genuine Kinds; Roles never subclassing; every enumeration a `skos:ConceptScheme`.
* **Option B — Keep the session-036 dual layer (subclasses + coded facet) for evidence.** Rejected by directing authority: it retains `isA` for a Role, contradicting the classification-over-inheritance doctrine being adopted (the prior art holds "evidence is a role"; see §More Information).
* **Option C — Mint one OWL subclass per evidence kind (status quo ante).** Rejected: subclass-per-coded-value is the documented "Role-as-subclass" anti-pattern (the prior-art 13-error fix, §More Information); the kind is an `isMemberOf` axis.
* **Option D — A "facet" property that also asserts the type.** Rejected: conflates classification with attribute-bearing (R2); the type is `isMemberOf`, the attributes are the facet.
* **Option E — Mirror the prior art's seven separate ODRs (one per concern).** Rejected: opda records the doctrine once here, cross-referencing the existing per-concern ODRs rather than forking seven parallel records.

## Decision Outcome

Chosen option: "Option A — Adopt the classification-over-inheritance doctrine wholesale", because it is a ratified, decade-stable practice of the adopted prior-art corpus (§More Information), grounded in ISO 25964 faceted classification, W3C SKOS, FIBO, and OntoClean/UFO.

opda adopts the classification-over-inheritance doctrine: **classification (what type a thing is) is done by coded set-membership (`isMemberOf` a SKOS concept scheme) by default, and by inheritance (`isA` / `rdfs:subClassOf`) only for genuine Kinds that carry their own identity criterion; Roles are anti-rigid and are NEVER `rdfs:subClassOf` a Kind; a *facet* is the type-specific attributes an entity of a given type bears (it presupposes the type, it does not assign it); and every controlled vocabulary / enumeration / classification axis is a `skos:ConceptScheme` coded value, never a subclass tree** — justified because it is the ratified, decade-stable practice of the adopted prior-art corpus (§More Information), grounded in ISO 25964 faceted classification, W3C SKOS, FIBO, and OntoClean/UFO.

### Consequences

* **Amend ODR-0011 §8a** to state R1–R5 as the governing doctrine (the cascade is the `isA`-admission test); cross-reference this ODR as the doctrinal home.
* **Re-model the evidence domain (ODR-0009 + the emitter, ADR-0011/0012):** retire the three `…Evidence rdfs:subClassOf opda:Evidence` axioms; keep `opda:AttachedDocument` as the genuine Kind; make `opda:evidenceType` the `isMemberOf` classifier; re-home kind-specific attributes as role-borne facets; re-key all SHACL value-keyed; re-type the exemplars; re-point the DPV records onto `opda:evidenceType`. Reverses session-036's keep-the-subclasses disposition (R6).
* **Audit the corpus** for any remaining Role/Phase/enumeration modelled as `rdfs:subClassOf` (the prior-art Role-as-subclass audit applied to opda, §More Information): Seller/Buyer/Proprietor already correct; the evidence tree is the known case; record any others found.
* Genuine Kinds keep `isA`: `opda:Property`, `opda:LegalEstate`, `opda:RegisteredTitle`, `opda:AttachedDocument`, `opda:Organisation`/`Person`, `opda:Address`, the activity/relator/interval types — all unaffected.
* Update the session-036 record + adoption.md to note the directing-authority adoption + the R6 reversal. Status `accepted` (directing-authority ratified; greenfield — no WG gate, ODR-0003 programme retired).

## More Information

- Adopted prior art — the source the §Context/§Considered-Options/§Rules prior-art references defer to (hm `~/source/hm/semantic-modelling/docs/ontology/odr/`): hm ODR-0010 (multi-faceted classification; **the 13 Role-as-subclass fixes** cited in §R1/§R3 and the Consequences audit; its title is the "facet = classification axis" usage noted in §R2), hm ODR-0014 (domain/range as documentation — the §R5 anchor), hm ODR-0016 (SKOS for enumerations), hm ODR-0023 (enumeration modelling pattern — the §R4 anchors), hm ODR-0025 (role-view modelling), hm ODR-0026 (property distribution / role-borne properties — the "borne by the role, not the Kind" principle stated in §R2), hm ODR-0092 (UFO stereotypes as SKOS schemes — §R3/§R4 anchor).
- Consolidating doctrine: [ODR-0033](ODR-0033-owl-axioms-as-documentary-ai-signal-doctrine.md) — this record's §R5 (`rdfs:domain`/`range` are documentation; SHACL enforces) is one of the fragments it gathers into the single "author OWL/RDFS axioms as documentary AI-signal, never entailed" doctrine.
- opda records governed/amended: [ODR-0011](ODR-0011-enumeration-vocabularies.md) §8a (the cascade — the `isA`-admission test), [ODR-0009](ODR-0009-claims-evidence-provenance.md) (evidence re-model), [ODR-0025](ODR-0025-entailment-regime-and-inference-semantics.md)/[ODR-0026](ODR-0026-owl-rl-safe-ruleset-adoption-and-unevaluated-modelling-axioms.md) (domain/range unevaluated; value-keyed), [ODR-0005](ODR-0005-property-land-identity-crux.md) (identity criteria; no `owl:sameAs`), [ODR-0006](ODR-0006-agents-and-roles.md) (RoleMixin/Role precedent), [ODR-0008](ODR-0008-property-descriptive-attributes.md) §Q5a (coded value-spaces).
- Council [session-036](council/session-036-classification-over-inheritance.md) — the cascade + value-keyed enforcement (stand); its keep-the-`…Evidence`-subclasses disposition is superseded here (R6).
- Foundational: Guarino & Welty 2009 (OntoClean); Guizzardi 2005 (UFO); ISO 25964 / W3C SKOS (faceted classification); FIBO (SKOS schemes alongside OWL classes).

## Rules

### R1 — Classification: `isA` vs `isMemberOf` (the two mechanisms)

**Classification** answers *"what type/kind is this entity?"* and has exactly two legitimate mechanisms:

- **`isMemberOf` (set membership / coded SKOS value) — the DEFAULT.** The entity carries a coded value (a `skos:Concept` notation) on a classifier property; the value's vocabulary is a `skos:ConceptScheme`. One class, a value that varies. Enforced by SHACL `sh:in` (typically via `sh:targetSubjectsOf` the classifier property, entailment-free — the `opda:ownerType`/`opda:tenureKind` idiom).
- **`isA` (inheritance / `rdfs:subClassOf`) — the EXCEPTION.** Mint an OWL subclass **only** when the sub-type is a genuine **Kind** carrying **its own identity criterion** (a distinct way its instances are individuated) and/or a distinct relational structure SHACL must enforce per-type. Per ODR-0011 §8a's OntoClean cascade: anti-rigid → never a subclass; `+R∧−I` → classify (coded); `+R∧+I∧+D` → Relator; `+R∧+I∧−D` → subclass.

A type-distinction failing the `isA` test is modelled `isMemberOf` (coded). Minting a subclass per coded value is the anti-pattern (ODR-0011 §Alternatives; ODR-0024 §R4 "namespace landmines"; and the prior-art Role-as-subclass fix cited in §More Information).

### R2 — A facet is NOT a classification

A **facet** is the **set of type-specific attributes an entity of a given type bears** — it *presupposes* the entity's type and says nothing about what that type is. `opda:attestedBy` (borne by evidence playing the vouch role), `opda:documentReference`, `opda:apiEndpoint` are facets. **`opda:evidenceType` is NOT a facet — it is an `isMemberOf` classifier** (it states the kind). Do not call a classifier a "facet." A facet's attributes are *borne by* the type/role that bears them — a role-specific property is borne by the role, not the Kind (prior art, §More Information) — declared with `rdfs:domain` on the most-specific class/role that introduces them.

> **Terminology note (overload).** The library-science / ISO-25964 / Ranganathan tradition uses "facet" to mean *an orthogonal classification axis* — the opposite of this rule's sense. **opda's canonical meaning is this rule's: facet = type-borne attributes.** When an orthogonal classification axis is meant, opda says "classification axis" or "coded scheme," never "facet."

### R3 — Roles are anti-rigid → NEVER `rdfs:subClassOf` a Kind

A **Role** is anti-rigid (an instance plays it contingently and can cease to). It is modelled as `opda:Role` / `opda:RoleMixin` borne via a `roleOf` / `playedBy` relation (founded by the relevant Relator/Activity), with a coded `isMemberOf` classifier where the role sub-divides — **never** `rdfs:subClassOf` the bearer Kind (the prior art fixed exactly this error 13 times; §More Information). opda already does this correctly for `opda:Seller`/`opda:Buyer` (`opda:RoleMixin`) and `opda:Proprietor` (`opda:Role`); the doctrine generalises it.

### R4 — SKOS for every controlled vocabulary / enumeration / classification axis

Every closed value-space, enumeration, status set, kind-discriminator, and UFO-stereotype axis is a `skos:ConceptScheme` of `skos:Concept`s, referenced by a coded value — **never** an `owl:oneOf` bag and **never** an `rdfs:subClassOf` subclass enumeration (opda ODR-0011; prior art in §More Information). A scheme member binds to a genuine OWL Kind by `skos:exactMatch` (the dual layer) **only where that Kind independently earns `isA` per R1** — never `owl:sameAs` (ODR-0005 §R5).

### R5 — `rdfs:domain`/`rdfs:range` are documentation; SHACL enforces; enforcement is value-keyed

`rdfs:domain`/`range` are informative (multi-domain reads as union), never evaluated (opda ODR-0025 §R2 / ODR-0026; the consolidated documentary-axiom doctrine is ODR-0033; prior art in §More Information). SHACL is the enforcement layer, and enforcement of per-type obligations is **value-keyed** on the `isMemberOf` classifier (`sh:targetSubjectsOf` + a value-guarded `sh:or(¬P ∨ Q)` material implication, SHACL-Core), **not** keyed on a subclass `sh:targetClass` (which is entailment-relative and silently passes a value-recorded instance — session-036; Knublauch). Class-targeting is correct in **one** place: a type↔value coherence shape (the class MUST carry its matching code).

### R6 — Partial supersession of session-036 for the evidence model (directing-authority adoption)

Council session-036 concluded `opda:DocumentEvidence`/`ElectronicRecordEvidence`/`VouchEvidence` should remain `rdfs:subClassOf opda:Evidence`. **That conclusion is superseded for the inheritance question by this ODR's R1/R3**, on the directing authority's adoption of the classification-over-inheritance doctrine: opda's own model states **"evidence is a role a document plays, not every document's Kind"** — so evidence-hood is a **Role**, and by R3 a Role is never `rdfs:subClassOf`. Therefore:

- `opda:Evidence` is the role target (a `RoleMixin`); **evidence-kind is an `isMemberOf` coded classification** (`opda:evidenceType` → `opda:EvidenceMethodScheme`), not a subclass tree.
- The three `…Evidence` `rdfs:subClassOf opda:Evidence` axioms are **retired**; kind-specific attributes (`opda:attestedBy`, document/record specifics) become **facets borne by the role** (R2), enforced value-keyed on `opda:evidenceType` (R5).
- The one genuine **Kind** in the family — `opda:AttachedDocument` (its own IC: content + issuing activity, ODR-0024 §R7) — **stays an OWL class** (it earns `isA` per R1); a document *plays* the evidence role, it is not sub-classed into it.
- `VouchEvidence`'s attestation, being an Agent-founded **Relator** (`+R∧+I∧+D`), is modelled as a Relator/role-borne structure, not a subclass of Evidence.

session-036's *value-keyed enforcement* finding and the ODR-0011 §8a cascade stand and are reinforced; only its keep-the-`…Evidence`-subclasses disposition is reversed.