The CBPA sample.

Clean reference ontology — what the demo ends up at after applying the methodology. Twenty-four domain classes extending gist 14 by IRI; full SKOS annotations; OWL property restrictions throughout; Account aligned to gist:Agreement, Payment to gist:Event, PaymentInstruction to gist:Task.

The deliberately flawed demo version. Four primary anti-patterns — deep hierarchy explosion via MT103_Payment / MT202_Payment / PACS008_Payment subclasses, plan/occurrence collapse on Payment, parallel property invention (hasType, containsAccount), and orphan domain classes (Branch, Customer) — plus mis-typed Specifications and minor hygiene flaws. Drop this on the linter for 52 findings across all seven rules.

SHACL shapes for the clean ontology. Enforces the structural constraints implied by the competency questions, including disjointness on Payment / PaymentInstruction to keep occurrence-time properties (payer, payee, viaChannel, settlementStatus) and planning-time properties (instructsAmount, requestedChannel, declaredPurpose, authorisedBy) on the right classes.

Synthetic instance data. Four banks (including two correspondent banks), three branches, six Korean and international customers across five risk ratings, eight accounts spanning KRW/USD/JPY/EUR, eight payments routed via SWIFT/RTGS/CHIPS, nine payment instructions, three KYC reviews, and two suspicious activity reports. Designed so every CQ has both a yes-case and a no-case. Validates clean against the SHACL shapes.

Twelve worked competency questions for CBPA. Each has a use case (AML alert triage, compliance dashboard, settlement reconciliation, KYC follow-up), the ontology elements it exercises, and a SPARQL skeleton against the clean ontology. CQ7 is the marquee plan/occurrence query — "which PaymentInstructions were authorised but never resulted in a Cleared Payment".