OWL axioms as documentary AI-signal — domain/range and the R2-excluded set

Status note. Draft, pending ratification. The directing authority is acting ratifier (operator-ratifies; the OPDA WG→Sub-Committee→AGM chain is not a live gate at this stage). Council session-050 (COMPLETE — 5–0 unanimous, no held-as-live dissent) confirmed the hm-coverage adoption per-construct and resolved the FP/IFP carve-out; folded in below.

Context and Problem Statement

OPDA’s entailment regime (ODR-0025 / ODR-0026 / ODR-0029) runs ONLY a frozen 7-rule RDFS-Plus load-time closure and EXCLUDES rdfs:domain/rdfs:range and the rest of the “R2 set” (owl:FunctionalProperty/InverseFunctionalProperty, owl:equivalentClass/equivalentProperty, owl:Restriction/unionOf/intersectionOf/complementOf/cardinality/oneOf/hasKey) from inference. ODR-0026 §R2 (“model-but-don’t-evaluate”) authors them as documentation, never entailed.

This doctrine was stated only in fragments (ODR-0025 §R2 / ODR-0026 §R2 / ODR-0027 §R5 / ODR-0029 §R1), and the gap had a concrete cost: Council session-047 (implemented in ADR-0048) re-derived rdfs:domain/range from STANDARD OWL semantics (the “everything-becomes-a-Person” trap) and stripped documentary domain/range off the relationship-layer object properties (founds/mediates/playedBy/plays/hasAddress), pinning types in SHACL only — even though the ~750 datatype properties correctly carry both rdfs:domain/range AND the SHACL dual (ODR-0029 §R3). That is doctrinally wrong under OPDA’s own regime, and made the relationship edges invisible to the class-graph model derivation (worked around by a SHACL-derivation hack in scripts/ontology-model.mjs).

Prior art is the source of truth for the coverage. The sibling project ~/source/hm/semantic-modelling already defines this OWL coverage (ODR-0014 domain/range; ODR-0030 the permitted/excluded construct table; ODR-0054 property characteristics; ODR-0028b disjointness; ODR-0100 single typing axis). The directing authority has decided OPDA adopts hm’s coverage as its baseline rather than re-deriving it. This ADR (a) states the positive doctrine once, (b) adopts hm’s coverage per-construct, (c) corrects the relationship-layer object properties, (d) records decision-4 (annotation + binding gates).

Decision Drivers

• Doctrine consistency — the regime already excludes domain/range from inference; authoring is safe; the ~750 datatype properties already do it.
• Reuse prior art — adopt hm’s already-defined coverage; do not re-derive (cross-project consistency).
• AI/LLM signal — OWL axioms are machine-legible documentation of modeller intent.
• OWL adherence — author standards-legible axioms even though OPDA never reasons.
• Proof obligation — whatever is authored must NOT change the closure (the ADR-0035 zero-domain/range-triple test).

Considered Options

Q1 — disjunction form (decided: adopt hm):

• (a) repeated rdfs:domain “any-of” + SHACL sh:or — hm ODR-0014’s form; boolean-constructor-free; the “any-of” reading is a documented convention (valid because OPDA never evaluates domain/range). CHOSEN (hm-adopted).
• (b) owl:unionOf + sh:or — standards-correct union but a boolean class constructor; hm ODR-0030 excludes it. The session-048 panel reached (b); superseded by the directing-authority decision to adopt hm’s coverage.
• (c) SHACL-only — rejected: loses the documentary/AI signal.

Q2 — property characteristics (pending session-050): adopt hm ODR-0054 (author Functional/InverseFunctional as documentary signal) WHOLESALE, vs carve out FP/IFP per OPDA’s owl:sameAs ban (ODR-0005 §R5 / ODR-0017 §R6) + the session-035 drop-redundant-axiom rule (which session-048 found pulls toward SHACL-only / IFP-out).

Q3 — disjointness (adopt hm ODR-0028b): author scoped owl:disjointWith + SHACL sh:not/sh:xone.

Doctrine (adopt): hm ODR-0030’s permitted/excluded construct table + an excluded-construct CI meta-shape, governed per-construct.

Decision-4 (accepted): mandate + CI-gate skos:definition coverage (every class + property) and rdfs:isDefinedBy term-binding, warning-first → violation.

Decision Outcome

Positive doctrine (accepted): OPDA authors OWL/RDFS axioms as documentary, AI-legible signal of modeller intent; they are NEVER entailed. The only inference is the frozen 7-rule RDFS-Plus closure (ODR-0025 §R1). Everything else is authored for documentation + VALIDATED via SHACL, never reasoned. Per-construct discipline (from session-048, adopted): author an R2-set construct as documentary signal only where its published W3C semantics AGREE with OPDA’s enforced behaviour (“engineering-act = ontological-act”). Corollary: do not reason about OPDA from standard W3C OWL/RDFS entailment semantics.

Q1 — DECIDED (directing authority): adopt hm ODR-0014 — repeated rdfs:domain “any-of” for the disjunctions (playedBy/plays/hasParticipant/hasAddress); plain rdfs:domain/range for single-class object properties; (a) repeated-domain over (b) owl:unionOf. This supersedes the session-048 panel’s owl:unionOf verdict (the panel reasoned without hm’s ODR-0014/0030 in its brief; the directing authority chose cross-project consistency with hm — boolean-constructor-free corpus). The “any-of” reading is documented in a module header + per-property rdfs:comment; SHACL sh:or keeps 100% enforcement (ODR-0029 §R3); re-run the ADR-0035 zero-triple test.

Decision-4 — accepted (ratifier, 2026-06-17): the skos:definition + rdfs:isDefinedBy coverage gates, warning-first.

Q2 — RESOLVED (session-050, 5–0): carve out FP/IFP from the hm adoption. Do NOT author owl:InverseFunctionalProperty (out entirely) or owl:FunctionalProperty as a general documentary layer; the home is SHACL sh:maxCount 1 + dash:uniqueValueForClass, scoped within each identity-bearing sortal, never cross-sortal (a cross-sortal key is IFP in SHACL clothing — it would fire on legitimate UPRN co-reference across the 3 Property Kinds). A narrow hand-curated owl:FunctionalProperty on genuine world-fact singletons is admissible only if the modal marker is wanted. Binding reason (for permanence): (ii) a published IFP asserts the negation of ODR-0005’s 12-0 bounded-context-identity ruling — it promotes a contingent identifier (UFO Quality) to a constitutive global identity criterion (an OntoClean rigidity-clash); (i) redundancy-with-the-safe-substitute (session-035) is corroborating only. Re-open ONLY if ODR-0005 is overturned. This carve-out is the application of the adoption’s own engineering-act=ontological-act test (domain/range agrees with the SHACL act → in; IFP disagrees → out), not a deviation from it.

Q3 — RESOLVED (session-050, 5–0): adopt hm ODR-0030 table + meta-shape (re-keyed) + ODR-0028b disjointness, with three OPDA tightenings. (1) The ODR-0030 permitted/excluded construct table is adopted re-keyed to OPDA’s frozen closure (documentary-only band larger than hm’s), and the excluded-construct CI meta-shape adds a fail-on-owl:InverseFunctionalProperty rule (enforces the Q2 carve-out) while keeping owl:unionOf excluded with no carve-out (Q1’s payoff — corpus stays boolean-constructor-free). (2) owl:disjointWith + SHACL dual, under a three-part authoring bar: (i) rigid sortals with their own IC, (ii) INCOMPATIBLE ICs — not merely distinct, not complementary, (iii) a real, occurring misclassification hazard. Standing scope = Person/Organisation ALONE now. The bar excludes two classes for two distinct reasons: anti-rigid Roles (no own IC to be incompatible — Seller/Buyer = transaction-scoped SHACL, session-047 Q4; grounded in identity-continuity), AND the co-referring Property/LegalEstate/RegisteredTitle Kinds (each has its own IC, but their ICs are complementary not incompatible → co-referential via opda:identifiesSameProperty, the opposite of disjoint — authoring disjointWith between them would wrongly fire on legitimate co-reference). Relator-Kind disjointness (Transaction/Proprietorship) deferred — passes (i)/(ii) but fails (iii) (relators are lifecycle-coupled, share relata) — gated on a named consumer. (3) Pairwise not owl:AllDisjointClasses — VERIFIED both gates (scripts/fuseki-load.mjs:277, tools/opda-gen/src/opda_gen/ci/inference_closure_test.py:282) consume authored pairwise owl:disjointWith directly (GRAPH ?g1 { ?c1 owl:disjointWith ?c2 }); zero authored today (inference_closure_test.py:37), so the gate is vacuous and authoring arms it — clearing session-035 condition (i).

Supersession scope

On acceptance, partially supersedes ADR-0048 §1 (the SHACL-instead-of-OWL typing decision). ADR-0048’s relationship-emission walk + ci-object-property-coverage gate survive — the gate relaxes from “type-pinned in OWL OR SHACL” to “documentary domain/range authored AND SHACL constraint present.” Requires a parallel ODR amendment to ODR-0032 §R1 and Council session-047 Q5 (an ADR cannot supersede an ODR cross-corpus — that correction is follow-on ODR work).

Consequences

• Good — the doctrine is stated once and applies uniformly; adopts hm’s defined coverage rather than re-deriving it.
• Good — the relationship layer regains documentary domain/range and becomes visible to the model derivation natively (the ontology-model.mjs SHACL-derivation hack can then be reverted or kept belt-and-braces — implementation decision T6).
• Neutral — nothing changes at inference time: the closure must still add ZERO domain/range triples (ADR-0035 test is the proof obligation).
• Bad — authoring is a maintenance commitment; the disjointness scope/carve-out must be enforced (gate).

Confirmation

• ADR-0035 zero-domain/range-triple consistency test passes after authoring.
• make verify-ontology byte-identity; full make ci; regenerate ontology-model.json + ontology-graph-elements.json via build:data.
• Relaxed ci-object-property-coverage gate; new ci-description-coverage + ci-isDefinedBy gates (decision-4), warning-first.

Council sessions and findings

• session-048 (Council A — Queen Kendall, DA Cagle; working/session-048/): Q1 (a) dead 4–0; (b) owl:unionOf 3–1 — SUPERSEDED by the hm-coverage adoption (now (a) repeated-domain); Q2 AGAINST FP/IFP 4–0; Q3 scoped disjointWith 3–1 (DA conditional-withdraw, no held dissent). Pivotal: Hendler withdrew the “everything-becomes-a-Person” premise as applied to OPDA (authored-but-never-evaluated is sound by construction). Verified precedents: RDFS 1.1 §3.2 (multiple rdfs:domain = conjunction); OWL 2 §2.3.5 (FP/IFP → sameAs); ODR-0026 §R3 / session-035 (drop-redundant-axiom-where-safe-substitute, 8–0–0).
• session-050 (COMPLETE, 5–0 unanimous, no held dissent; Queen Kendall, DA Hendler, panel Allemang/Guarino/Cagle; working/session-050/): confirmed the hm-coverage adoption per-construct; resolved Q2 (carve out FP/IFP — reason (ii) ODR-0005 binding); refined Q3 (ODR-0030 table re-keyed + meta-shape +IFP-excluded; disjointWith Person/Organisation-alone-now + Relator-deferred + pairwise); verified both ADR-0035 gates consume authored pairwise owl:disjointWith directly. DA Hendler: Q1/Q2 HOLD (conditions met by the verdicts), Q3/Q4 WITHDRAW. Supersedes session-048 Q1 only (its Q2/Q3 survive, re-affirmed). Record: docs/ontology/odr/council/session-050-adopt-semantic-modelling-owl-coverage.md.
• Held dissents: Cagle (session-048) Q1 omit-now, Q3 cost-discipline (re-open: a consumer that reads a class-level partition off the axiom and cannot reconstruct it from sh:xone+SHACL) — both non-blocking; largely moot under the repeated-domain adoption (no owl:unionOf), to be reconciled in session-050.
• Correction (verified): currentEnergyRating (opda-property.ttl:696-702) is NOT rangeless — it carries rdfs:domain opda:Property ; rdfs:range skos:Concept + sh:in; it is the correct object-property→skos:Concept+sh:in enumeration pattern, not an anomaly. (An earlier research claim was wrong; do not “fix” it.)

Outstanding work and open tasks

Implementation rework (follow-on; each gated by make ci + the ADR-0035 zero-triple proof + byte-identity):

1. Author documentary domain/range on the relationship-layer object properties… DONE (2026-06-17, verified at file:line + independent gate re-run): emitter changes in emitters/modules/{agent,transaction,property}.py — founds (rdfs:domain opda:Relator; range any-of Role, RoleMixin), mediates (plain Proprietorship → Proprietor), playedBy/plays (any-of Role∪RoleMixin ↔ Person∪Organisation), hasParticipant (range any-of Seller, Buyer), hasAddress (restored domain any-of Property, Person, Organisation; range Address kept); concernsProperty/hasRegisteredTitle already had both. SHACL sh:or duals retained (authoritative); “never reasoned” comments preserved; the false “NO rdfs:domain / would entail” comments reversed; CI-gated “any-of” skos:editorialNote added to all three module headers. Necessary adjunct: emitters/shapes.py excludes multi-domain predicates from the single-sh:class auto-derivation (multiple sh:class on one shape = unsatisfiable conjunction) and corrects emitted sh:message strings. Verified: make verify-ontology byte-identical; make ci-ontology all gates green; the ADR-0035 zero-domain/range-triple proof holds (31 inferred types, all subclass-derived, 0 leak); 0 real owl:unionOf/FP/IFP axioms; build:data regenerated the model JSONs (relationship edges now native). Sequencing satisfied (task 8 landed first). NOT committed.
2. Q2 FP/IFP — resolve in session-050, then implement (SHACL-only carve-out vs hm-authored).
3. Scoped owl:disjointWith (Q3) — rigid Kinds only, never anti-rigid Roles, pairwise; keep SHACL dual.
4. Adopt hm ODR-0030 permitted/excluded construct table + the excluded-construct CI meta-shape… DONE (2026-06-17, commit 8681c4c): permitted/excluded OWL-construct table added as ODR-0033 §R6 (3 bands re-keyed to OPDA’s frozen closure); corpus-wide ci-excluded-construct gate (wired into make ci-ontology) generalising the object-property b3 limb to all 9 excluded families corpus-wide (positive control fires; corpus has zero → passes, arms drift protection). Decision-4 gates landed in the same commit: ci-description-coverage (skos:definition) + ci-isDefinedBy (rdfs:isDefinedBy), both wired warning-first. Initially both reported 0/303 (warning-first). BACKFILL DONE (2026-06-18, commit d937374): fanned-out per-module drafting added an ISO-704 genus–differentia skos:definition@en + rdfs:isDefinedBy → owl:Ontology to all 303 OWL terms (sourced from the PDTF v3 JSON-schema descriptions + each term’s existing rdfs:comment/scopeNote + domain). Both gates now 303/303 (100%) and have FLIPPED warning-first → hard (a future term without a definition/isDefinedBy fails CI). The gate loader gained foundation.ttl so the base owl:Ontology is recognised for the foundation classes; consumesFrom was defined + repointed to base; the baseline unit test was updated to the completed state. byte-identity re-pinned; full make ci green; no model/graph-JSON drift.
5. Reconcile ci-object-property-coverage… DONE (2026-06-17): object_property_coverage_test.py limb (b) inverted — (b1) a disjunction predicate lacking its SHACL sh:or dual fails; (b2) a disjunction predicate present while the module-header convention note is absent fails; (b3) FP/IFP authored on an object property fails (partially realises task 4’s excluded-construct check). Old “fail on not-universally-true rdfs:domain” removed. Unit tests updated. Green on re-run.
6. scripts/ontology-model.mjs:207-248 SHACL-derivation block… DONE (2026-06-17): KEEP as belt-and-braces (session-050 Q4). The object properties now carry documentary domain/range so single-class edges derive natively; the SHACL recovery is retained (not reverted) because the disjunctive “any-of” properties carry authoritative per-class typing only in SHACL sh:or (documentary multi-domain/range under-determines the precise per-class edge), and the merge is idempotent with the native edges. Comment corrected (the superseded session-047 “kept out of OWL to avoid mis-entailment” premise replaced); no code/output change.
7. hasAddress — restore documentary domain (opda-property.ttl:704-709); reconcile with the profiles.py D4 silent-loss fix across 31 forms (ADR-0048:82).
8. ODR-0032 §R1 amendment (cross-corpus) + Council session-047 Q5 amendment — an ADR cannot supersede an ODR. DONE (2026-06-17): ODR-0032 §R1 (the false “MUST NOT be authored” premise corrected), §R2 (session-050 amendment block — founds/mediates documentary domain/range + disjunction any-of + FP/IFP carve-out), §Decision-detail (the “deferred option” framing corrected + taken up), §Confirmation limb (b) inverted, depends-on: +ODR-0026; and session-047 Q5 carries a session-050 supersession banner. This unblocks task 1.
9. ODR-0008 §Q6a reasoner-independence test — re-read against the now-defined closure.
10. Positive doctrine ODR… DONE (2026-06-17, verified): created ODR-0033 (“OWL/RDFS axioms as documentary AI-signal — the consolidated doctrine”; kind: architecture, council: session-050; R1 documentary-not-entailed, R2 engineering-act=ontological-act per-construct test, R3 the no-standard-entailment corollary, R4 the FP/IFP carve-out, R5 the repeated-rdfs:domain any-of form). Stripped ALL hm/semantic-modelling references from the normative text of ODR-0025/0026/0027/0029 → relocated to each record’s §More Information as prior art (verified: zero hm tokens remain in any pre-§More-Information section; ratified rule bodies + frontmatter byte-identical to HEAD — only citations relocated/neutralised + ODR-0033 cross-refs added).
11. Proof obligations — regen the model JSONs (build:data, commit both); ADR-0035 zero-triple; byte-identity; full make ci.
12. Write the council session records (session-048/049/050) + add adoption.md track-record rows.

More Information

• Working brief: docs/HANDOVER-2026-06-17-entailment-doctrine-domain-range-as-signal-relationship-layer-rework.md.
• Prior art adopted (cite here only; strip from normative ODR text): ~/source/hm/semantic-modelling ODR-0014 (domain/range), ODR-0030 (owl-as-documentation + excluded-construct table), ODR-0054 (property characteristics), ODR-0028b (disjointness), ODR-0036 (SHACL-rules-&-OWL-inferencing).
• Verified this session via /ruflo-swarm:swarm research + adversarial verification + Council sessions 048/049 (and 050).

Amendments

• 2026-06-17 (a) — created in Draft; positive doctrine + decision-4 accepted; decisions 1/2 OPEN pending Council A.
• 2026-06-17 (b) — Council A (session-048): Q1 (b) owl:unionOf 3–1; Q2 no-FP/IFP 4–0; Q3 scoped disjointWith 3–1.
• 2026-06-17 (c) — directing-authority decision: adopt semantic-modelling’s coverage → Q1 flips to hm ODR-0014 repeated-rdfs:domain “any-of”, superseding session-048’s owl:unionOf verdict; session-050 convened to confirm consequences + the Q2 FP/IFP reconciliation.
• 2026-06-17 (d) — Council session-050 (5–0 unanimous, no held dissent): hm-coverage adoption confirmed per-construct. Q2 resolved (FP/IFP carve-out; reason (ii) ODR-0005-binding, re-open only if ODR-0005 overturned). Q3 resolved (ODR-0030 table re-keyed + meta-shape +IFP-excluded; owl:disjointWith Person/Organisation-alone-now + Relator-deferred + pairwise, gate-consumption verified at fuseki-load.mjs:277/inference_closure_test.py:282). Q4 consequences confirmed; founds/mediates authoring sequenced after the ODR-0032 amendment; ci-object-property-coverage limb (b) inverts (not merely relaxes). Session-048 superseded on Q1 only.