Curated Category-G Walk — Leaf Dispositions and Modelling Rules

Context and Problem Statement

The ADR-0031 curated walk emitted or dispositioned the 188 ODR-0022 candidate Category-G descriptive leaves (156 minted as flat properties, 29 collapsed onto shared/existing properties, 3 deferred), via the ADR-0030 generator. The per-leaf modelling decisions were made directly (greenfield first cut, no WG gate) and then put to the Linked Data Council — session-028 — a full panel (Queen Allemang; Guizzardi/Kendall/Cagle; DA Hendler) for ratification.

This ODR records the rules and dispositions the council settled; the deliberation transcript and per-question verdict tally live in the session record. The council ratified the walk’s mechanical spine but returned two corrections that are normative here: the headline monetary leaves had been wrongly collapsed onto the Category-D fixtures opda:price, and the nearby-facilities bearer classes had been minted under a citation (§Q4a) that does not license them. Both are corrected below.

Considered Options

• Option A (chosen) — Adopt per-family disposition rules with two council corrections. Flat datatype properties on the nearest bearer Kind, two ratified collapse conventions, monetary leaves deferred to a MonetaryAmount-based walk, and opda:NearbyFacility re-warranted as a UFO Substance-Kind bearer.
• Option B — Collapse all amenity leaves to the bare genus / domain-less. Rejected: loses the precise bearer (a transport node has no pupils); the genus + SHACL-per-band keeps the bearer honest without the held Subkind cost.
• Option C — Mint one price property per monetary leaf. Rejected: ODR-0022 §4 forbids per-item price proliferation; the value type is shared, the bearer is not.
• Option D — Reuse the fixtures opda:price for headline amounts. Rejected: conflates Category-D and Category-G (ODR-0022 §1/G1) and erases incompatible value semantics (one-shot vs recurring vs refundable) under one bare decimal.
• Option E — Mint School/HealthCareFacility subkinds now. Rejected for the first cut: no consumer query forces the split; held-as-live (Guizzardi dissent preserved).
• Option F — Reuse opda:DocumentEvidence for attached registry documents. Rejected: the owl:equivalentClass binding entails eIDAS-Substantial evidence on every attached document.
• Option G — Keep the 7-name C-vs-G allow-list. Rejected: it silently drops every future enum-bearing Quale whose tail is not listed.

Decision Outcome

Chosen option: “Option A — Adopt per-family disposition rules with two council corrections”, because the value-space distinctions either already exist in the emitted TTL or are byte-identically recoverable, and honest coverage accounting (179/239 with the 60 uncovered reported as a follow-on, not silent omissions) is sounder than false coverage by collapsing leaves onto semantically incompatible properties.

Adopt the per-family disposition rules in §Rules for the descriptive Category-G leaves: flat datatype properties on the nearest bearer Kind by ODR-0008 §Q5a/§Q6a, two ratified collapse conventions (free-text → opda:disclosureDetail, identity → existing join predicates), monetary leaves deferred to a MonetaryAmount-based monetary walk rather than collapsed onto the fixtures price, and opda:NearbyFacility re-warranted as a UFO Substance-Kind bearer; because the value-space distinctions either already exist in the emitted TTL or are byte-identically recoverable, the walk ships as a greenfield first cut at honest 179/239 coverage — R5’s structural C-vs-G rule grew the candidate-G set 188→239 by surfacing enum-bearing attributes the old allow-list mis-binned to C, and the 60 uncovered (18 deferred monetary + 3 held opda:Room + ~39 newly-surfaced enum attributes) are a reported follow-on, not silent omissions.

Consequences

• Remediation before this leaves proposed — blocking (DA blockers): re-warrant opda:NearbyFacility + collapse subkinds + schoolType→SKOS (R4); withdraw the 18 monetary collapses → defer + fix the opda:price comment (R3), honest coverage 185 → 179/239 (R5 grows the denominator 188→239).
• Should-fix follow-ups: titleNumber→RegisteredTitle (R8); mint opda:AttachedDocument + break the Document≡DocumentEvidence conflation (R7); structural C-vs-G rule + re-run + regression test (R5); mint ~5 SKOS schemes + re-range (R6); SHACL mediaUrl/url (R11, done) + hasSubAssessment acyclicity (R11, DONE 2026-07-04 — opda:shape:RiskAssessmentAcyclicityConstraint, a sh:sparql/sh:select SPARQLConstraint walking opda:hasSubAssessment+ as a property path, since the pre-existing sh:node recursion can validate each sub-assessment’s own shape but cannot itself detect a genuine cycle).
• The Category-G monetary walk (with opda:MonetaryAmount) is the next deferred chunk and owns the ODR-0008d item-3 deferral.
• All emitter changes regenerate through ADR-0030; ci-byte-identity, ci-dup-declaration, and ci-category-g-coverage continue to gate.
• No byte-identity re-pin is forced by R5.

More Information

• ODR-0008 §Q4a/§Q5a/§Q6a/§Q3a — the descriptive-attribute binding rules this walk applies.
• ODR-0008d item-3 — the deferred opda:MonetaryAmount value-type question (R3 answers it).
• ODR-0022 §1/§3/§4/§Rules.1/.6 — Category-G curation strategy + the C↔D↔G boundary this ODR implements.
• ADR-0030 — the emission generator + gates. ADR-0031 — the walk execution plan.
• Council session-028 — the deliberation, per-question verdicts, and DA scorecard behind these rules.

Rules

R1 — Minting (the 156). A candidate-G leaf with no ratified collapse mints one flat opda: datatype/object property: range from the data-dictionary type (no enum → plain xsd: datatype; §Q5a), flat per §Q6a (no rdfs:subPropertyOf), rdfs:domain the nearest existing bearer Kind (opda:Property physical attrs; opda:LegalEstate tenure/lease/charge attrs; opda:Person/Organisation/Proprietorship agent attrs; opda:Search planning/search results; opda:Valuation comparable pricing; opda:Transaction completion undertakings; opda:RiskAssessment risk fields), and a dct:source array of its G2 schema-leaf-paths (§Q3a). The six G11 overlap leaves (builtForm, centralHeatingFuelType, currentEnergyRating, heatingType, ownershipType, propertyType) STAND on this rule (ratified, Q8).

R2 — Collapse (free-text + identity). Generic prose tails (details, description, summaryDescription, propertyFullDescription, dimensionDetails, caption, workAlreadyDone, workToBeDone) collapse to the one reusable opda:disclosureDetail (ODR-0022 §Rules.1; the question is carried by the subject + instance dct:source, never a per-leaf property). Identity leaves collapse onto the existing canonical predicate (uprn → opda:hasUPRN, address → opda:hasAddress); minting a duplicate is forbidden (ODR-0005/ODR-0015 own the canonical modelling).

R3 — Monetary leaves DEFER; they do NOT collapse onto the fixtures price. The headline/recurring monetary leaves (asking/sold/list/estimate prices, rents, deposits, ground rent, service charges, fees, deed/permit costs, potentialCost — ~18 here, ~225 across the corpus) are deferred to the Category-G monetary walk (ODR-0008d item-3). They MUST NOT be collapsed onto opda:price, which is the Category-D fixtures amount only (ODR-0022 §4); ODR-0022 §1/G1 keeps the headline price distinct. At the monetary walk: a value-structured opda:MonetaryAmount (magnitude + currency over an ISO-4217 SKOS scheme; currency-defaulting-GBP via the overlay profile, never absent on the value type), reused as the datatype across distinct per-economic-kind properties (askingPrice/groundRent/serviceCharge/deposit/fee…) each bound to its own bearer — reuse the datatype, never the bearer. The fixtures opda:price stands as a single-currency interim; its rdfs:comment MUST drop the “reuses a MonetaryAmount pattern” claim until that type exists. EXECUTED 2026-05-31 (the monetary walk, ADR-0005 §G22): opda:MonetaryAmount minted as a by-value structure (opda:amount xsd:decimal magnitude + opda:currency → the seeded-extensible ISO-4217 opda:CurrencyScheme = GBP/EUR/USD; the node shape makes both required and sh:in-restricts currency to the scheme, GBP-default via overlay). 16 per-economic-kind object properties minted, each rdfs:range opda:MonetaryAmount on its own bearer — opda:LegalEstate (annualGroundRent · annualServiceCharge · certificateOfComplianceFee · sharedOwnershipRent) · opda:Property (councilTaxAnnualCharge · annualCostOfPermit · rent · holdingDeposit · securityDeposit · potentialCost) · opda:Valuation (estimatedPrice · estimatedAmount · listPrice · soldPrice) · opda:Transaction (costsApplicableToTheDeed · feeIncludingVAT) — plus opda:amount covering the amount leaf. The opda:price comment was updated (MonetaryAmount now exists; price stays a single-currency fixtures interim, not migrated). Coverage 179 → 196/239; all 6 CI gates green; 47 SKOS schemes (+CurrencyScheme); 264 tests pass.

R4 — Nearby facilities. A referenced nearby facility is a UFO Substance Kind (a mind-independent social-physical endurant with its own identity — not a Quality of opda:Property, since the same facility neighbours many properties). opda:NearbyFacility is warranted on that UFO basis, not ODR-0008 §Q4a (§Q4a is a provenance/lifecycle/PII test and does not license it). For the first cut, opda:School / opda:HealthCareFacility collapse into the genus bearer (band-specific properties SHACL-scoped per overlay); the precise-bearer Subkind split (+ opda:TransportNode) is held-as-live (Guizzardi dissent) with re-open trigger: a consumer query needing per-band typing (the “ODR-0023 R2 axis review” arm is now spent — R2 was adjudicated session-029 and did NOT spawn the UFO sub-modules, so per-band typing rides only on a consumer query). The schoolType bands are modelled as one opda:schoolType datatype property over an opda:SchoolTypeScheme — NOT five range-less generic object properties (opda:primary/private/… are namespace landmines).

R5 — C-vs-G categoriser rule (structural, replacing the allow-list). The Category-C vs Category-G boundary is: a value-space that is a cross-cutting status flag reused across unrelated contexts → C; a substantive Property/estate Quality or Substance-Kind-label whose value-space is enumerated → G, with the enum as its SKOS range. Replace the 7-name _G_PROPERTY_QUALE_TAILS allow-list in leaf_categoriser.py with this structural signal, re-run categorise-leaves so the candidate-G count is honest (~200+, not 188), and add a regression test asserting no new enum-leaf under a Property/estate path lands in C unexamined. This corrects a report/curation-scope miscount only — the emitted terms (propertyType, hasBeenFlooded, isInsured, …) are already correct G-grade in the TTL; no byte-identity re-pin is required.

R6 — SKOS from live enums (Q-SKOS). Mint a SKOS scheme where the data actually carries an enum: opda:ConstructionTypeScheme, opda:PriceQualifierScheme, opda:TransportTypeScheme, opda:BroadbandConnectionTypeScheme, opda:OfstedRatingScheme (Ofsted-authority dct:source); re-range ownerType → opda:OwnerTypeScheme and marketingTenure → the existing tenure scheme (reuse-before-mint; do not mint a third tenure scheme). Leaves that are genuinely free strings in the data (documentTypeCode, pricingMethodology, typeOfHealthCare, religiousCharacter, supplyClassification) stay bare xsd:string. EXECUTED 2026-05-31 (the R5 follow-on walk, ADR-0005 §G23): the 40 R5-surfaced leaves minted as flat datatype properties on their bearers (opda:Property / opda:LegalEstate / opda:NearbyFacility; forTheManagedAreas/fromTheOwners as xsd:decimal, the rest xsd:string). The six enum leaves were wired to their schemes via base value shapes (sh:targetSubjectsOf + sh:in the scheme notations, no overlay needed): constructionType→ConstructionTypeScheme · priceQualifier→PriceQualifierScheme · typeOfConnection→BroadbandConnectionTypeScheme · transportType→TransportTypeScheme · ofstedRating→OfstedRatingScheme · marketingTenure→existing TenureKindScheme (reuse). ownerType re-ranged → OwnerTypeScheme with its sh:in wired the same way — closing the pre-existing Proprietor-overlay gap (the sh:targetSubjectsOf idiom needs no overlay). Coverage 196 → 236/239 (only the 3 held opda:Room leaves, §G24, remain); all 6 CI gates green; 265 tests.

R7 — Attached documents vs evidence. Registry-attached document filing-metadata (documentDate, documentTypeCode, filedUnder, retrievedOn on titlesToBeSold[].additionalDocuments[]) binds to a neutral opda:AttachedDocument; opda:DocumentEvidence (ODR-0009) is reserved for artefacts actually standing as evidence (its owl:equivalentClass alias opda:Document MUST NOT be the bearer, or every attached doc is entailed eIDAS-assured evidence). Give the document Kind an explicit IC (content + issuing activity, not documentTypeCode/documentDate).

R8 — titleNumber domain. rdfs:domain opda:RegisteredTitle — a title number identifies the HMLR register record, not the legal estate; reach the estate via the existing opda:recordsEstate join. (Allemang dissent: estate-side cited identifier — if wanted, a separate co-reference predicate, never an overload of the identifier’s domain.)

R9 — Bank-account leaves. accountName/accountNumber/sortCode are flat xsd:string datatype properties on opda:Organisation (leading-zero-significant — never xsd:integer). Promote to an opda:BankAccount Kind only on a named trigger: an account shared across organisations, account-level lifecycle/validation (IBAN/BIC/mandate), or a query retrieving accounts independent of holder.

R10 — opda:Room / opda:Building: model as a by-value structure, mint no Kinds (RESOLVED, session-030). Neither is promoted to a class: opda:Room has no data-realisable identity (roomName is a non-rigid label, not an identity principle; no positional token in source) — 8–0; opda:Building’s +O identity criterion (the ODR-0005 §3a-4 “Replacement” witness) is genuine but latent/unexercised, so minting it now is an inert empty Kind — 7–1. The length/width/roomName repeating group (roomDimensions.rooms[]) is modelled as an anonymous by-value opda:RoomDimension structure (no IC, no key — the R3 opda:MonetaryAmount pattern) attached to opda:Property via opda:hasRoomDimension (0..*, characterisation not parthood, no transitivity): opda:length/opda:width (xsd:decimal, metres-by-rdfs:comment — no UnitOfLengthScheme, reuse the opda:area precedent), opda:roomName (xsd:string, non-rigid, never a key) + a keyless SHACL shape. Emission EXECUTED 2026-05-31 via ADR-0033 (ADR-0005 §G24 closed): opda:RoomDimension + opda:length/opda:width (xsd:decimal) + opda:roomName (xsd:string) + opda:hasRoomDimension + a keyless opda:RoomDimensionShape, via the ADR-0030 generator — the curated Category-G walk is now complete at 239/239 (224 minted, 15 collapsed, 0 uncovered); all 6 CI gates green, 266 pytest pass. Held dissents (re-open on an identity fact, not a date): a thin opda:Building bearer (Kendall); a UnitOfLengthScheme (Guizzardi/Davis/Guarino); Building promotion (a fact exercising the Replacement witness); Room promotion (a stable room token + a re-identifying query).

R11 — Domain-less convention. name, displayName, mediaUrl, url stay rdfs:domain-less (genuinely cross-artefact; forcing a bearer would assert false inherence). SHACL property-shapes carry their per-context constraint; add sh:datatype xsd:anyURI + a URI sh:pattern for mediaUrl/url. Mint a common superclass only on a named query quantifying over all referenceable artefacts.

R12 — Coverage accounting. After remediation, honest coverage is 164 minted + 15 collapsed = 179/239 — R5’s structural C-vs-G rule grew the candidate-G set 188→239 by surfacing enum-bearing attributes the old 7-name allow-list mis-binned to C. The 60 uncovered are the 18 deferred monetary leaves + the 3 held opda:Room leaves + ~39 newly-surfaced enum attributes (a reported follow-on walk, not silent omissions). ci-category-g-coverage is the local-only tracker (the data dictionary is gitignored); collapsing the 18 monetary leaves onto opda:price was false coverage and is withdrawn from the register.