# Ontology primer: facilitator and learner guide Teaching edition, October 2026. Use the 44-slide foundation deck, then workshop slides W01–W10. All organizations, records and amounts in the exercises are fictional. Studio is Beta; use synthetic or authorized, non-sensitive business-meaning material. ## Choose a route - **Executive introduction, 30 minutes:** foundation slides 3–12, 20–21, 24–27 and 35; discuss one decision the organization needs to make. - **Foundations, 60–90 minutes:** work through slides 1–38 with pauses at 12, 27 and 37. Treat slides 39–44 as a technical appendix and reading list. - **Hands-on lab, 45–60 minutes:** use W01–W10 after participants understand concepts, instances, relationships and evidence. Advanced participants can inspect the RDF, OWL and SHACL examples in the appendix separately. Times are suggested teaching plans. Adjust to the group; the goal is an explainable artifact rather than covering slides quickly. ## Prepare the lab Open https://rjlsol.com/ontology/studio/ and the user guide at https://rjlsol.com/ontology/studio/manual.html. Start a separate practice project or a worked example. Do not overwrite working material. Keep a full backup before importing. A guest project is sufficient; accounts and paid API requests are not required. Give participants the fictional records in W06 and W07 and a blank worksheet using the fields below. ## Participant worksheet 1. **Question and purpose:** What decision or task will this model support? Who will use it? 2. **Boundary:** Which entity, date, jurisdiction, population and exclusions matter? 3. **Meaning:** List five essential terms, their definitions, an example and a counterexample. 4. **Connections:** Label each relation. Separate a class hierarchy from an entity relationship and a temporary role. 5. **Evidence:** For each definition and factual claim, name the source, version, location and permitted use. Keep facts separate from vocabulary definitions. 6. **Uncertainty:** Which claims conflict or lack support? Who can resolve them? 7. **Evaluation:** Write three reference questions and a critical-failure rule before viewing an answer. 8. **Handoff:** Specify the recipient, chosen export, private-content selection and separate recovery backup. ## Teaching notes and answer keys ### W01: intended outcome Ask each learner to name the question their final artifact should help answer. Challenge a model that has many terms but no intended use. A small reviewed model is a valid result; term count is not a measure of success. ### W02: financial-services scope A satisfactory rewrite identifies legal entity ORG-18, the agreed exposure measure, the as-of date and the included instruments or positions. The ontology cannot supply current balances, exposures, approved limits or entity matches simply by defining them. Those need authorized records and executable calculation rules. Do not let “exposure” silently switch between risk measures. ### W03: collection and permission The learner should record title, version/date, locator, the passage used and permitted use. Distinguish permission to read a document from permission to upload it to a cloud service or AI provider. If two policy versions conflict, keep both and identify the applicable date and authority. A search result ranking does not decide authority. ### W04: evidence-backed review A good review checks every substantive clause. If a passage says only that a policy has an effective date, it does not establish an active-policy cancellation rule. Remove the unsupported rule, mark it unresolved or obtain the governing source. Keep the proposal draft while support is missing. A typed reviewer name is a local record, not authenticated institutional approval. ### W05: model versus facts Organization-as-Party is a class relationship. Harbor-as-Organization is an instance classification. Being a policyholder or broker is a role in a relationship, which may change by policy and time. Record an effective interval for a role assignment; separately record when the system observed it. Explain the distinction in ordinary language before introducing formal syntax. ### W06: PE/VC convention Under the exercise convention, ARR is $10,000 × 12 = **$120,000**. Exclude the $2,000 one-off fee and $3,000 monthly uncontracted pipeline. Do not add monthly and annual values before aligning units. To use the convention on real work, obtain active contract facts, relevant dates, currency, amendments and the organization’s agreed metric policy. The figure is an illustrative run-rate measure, not recognized revenue or an investment conclusion. ### W07: insurance and time **C-17 is supported as open on 10 October**, given the supplied history and P-204 facts. **C-18 is unresolved for that date**: being open in the 13 October export does not prove it was open on 10 October. A good response distinguishes the supported result from the missing history and does not report a complete portfolio count. Ask for C-18’s event history and policy relationship. A current-state snapshot cannot always reconstruct history. ### W08: comparison Use the same model/version, prompt question, evidence and scoring criteria. Vary the supplied domain context deliberately. Score facts, citations, scope and uncertainty separately; an invented material fact is a critical failure even if other items pass. Keep a question unscored until assessed. “Not applicable” requires a reason. A small user-entered comparison informs revision; it does not establish statistical uplift or regulatory compliance. ### W09: storage and destination Ask learners to describe the destination before clicking. Full collections, drafts and practice records remain local; full backups are recovery files. Selected linked cloud models may contain accepted private business-meaning excerpts. Optional BYOK drafting sends selected evidence and the request-only key through the hosted request service to the selected provider. Review the destination and possible charge. Provider terms still apply. Do not use personal data, customer records, credentials or regulated data. A context pack is not a backup, and an export format does not itself enforce access at its destination. ### W10: exit review A partner should be able to explain a term, trace its supporting source, identify an unsupported fact, reproduce an exercise result and restore a backup. Select formal OWL reasoning or SHACL validation only when the requirement warrants the tooling; Studio exports and local checks are not proof that those standards have been executed. A SKOS vocabulary export is not automatically a full OWL ontology or populated enterprise knowledge graph. ## Common misconceptions to correct - A shared label is not proof that two records identify the same entity. - A glossary definition is not current transaction data. - A source citation must actually support its neighboring claim. - An OWL restriction is not the same thing as a required input field. - A vocabulary or graph does not grant access to source data. - An LLM suggestion and a user-recorded review are different kinds of evidence. - A generated or exported ontology is not evidence that an AI system performs better. ## Primary reading and task guide The foundation slides retain their source links. For the distinctions used in the workshop: - OWL 2 Primer: https://www.w3.org/TR/owl2-primer/ - Shapes Constraint Language: https://www.w3.org/TR/shacl/ - SKOS Primer: https://www.w3.org/TR/skos-primer/ - Studio user guide: https://rjlsol.com/ontology/studio/manual.html The examples are teaching choices. Check the current Studio guide for interface changes and the cloud terms/privacy notice before using organizational material.