AI Ontology Bottom-Up: A Live Virtuoso Meshup

CREATE TABLE aio_source_system (
  system_id   INTEGER NOT NULL PRIMARY KEY,
  system_name VARCHAR(64),
  system_type VARCHAR(32)
);

CREATE TABLE aio_business_entity (
  entity_id   INTEGER NOT NULL PRIMARY KEY,
  entity_name VARCHAR(64),
  entity_type VARCHAR(32)
);

CREATE TABLE aio_business_metric (
  metric_id   INTEGER NOT NULL PRIMARY KEY,
  metric_name VARCHAR(64)
);

As first created: exactly the shape the article describes, and exactly missing the validity period it flags as absent.

CREATE TABLE aio_mined_claim (
  claim_id           INTEGER NOT NULL PRIMARY KEY,
  entity_id          INTEGER NOT NULL,
  metric_id          INTEGER NOT NULL,
  source_system_id   INTEGER NOT NULL,
  formula_expression VARCHAR(255),
  confidence         FLOAT,
  observed_at        DATETIME
);

CREATE TABLE aio_reviewer_feedback (
  feedback_id    INTEGER NOT NULL PRIMARY KEY,
  claim_id       INTEGER NOT NULL,
  reviewer       VARCHAR(64),
  new_confidence FLOAT,
  reviewed_at    DATETIME
);

Applied AFTER initial load, deliberately, to demonstrate the article's own recommended fix as an addition. VERIFIED: both statements returned "Done".

ALTER TABLE aio_mined_claim ADD valid_from DATE;
ALTER TABLE aio_mined_claim ADD valid_to   DATE;

VERIFIED loaded. Row 1 (CRM) is deliberately superseded so the current-vs-historical distinction is visible in the data, not just asserted in prose.

INSERT INTO aio_source_system (system_id, system_name, system_type) VALUES
  (1, 'Cloud Warehouse', 'warehouse'),
  (2, 'CRM', 'crm'),
  (3, 'BI Tool', 'bi_tool'),
  (4, 'Metadata Catalog', 'catalog');

INSERT INTO aio_business_entity (entity_id, entity_name, entity_type) VALUES
  (1, 'Opportunity', 'entity'),
  (2, 'Deal', 'entity'),
  (3, 'Pipeline', 'entity'),
  (4, 'Customer', 'entity');

INSERT INTO aio_business_metric (metric_id, metric_name) VALUES
  (1, 'win_rate'),
  (2, 'pipeline_value');

INSERT INTO aio_mined_claim (claim_id, entity_id, metric_id, source_system_id, formula_expression, confidence, observed_at, valid_from, valid_to) VALUES
  (1, 1, 1, 2, 'closed_won / all closed opportunities', 0.82, '2026-06-01', '2026-06-01', '2026-06-30'),
  (2, 1, 1, 1, 'closed_won / closed opportunities excluding unqualified', 0.76, '2026-06-03', '2026-06-03', NULL),
  (3, 1, 1, 3, 'closed_won / new-business pipeline only', 0.65, '2026-06-05', '2026-06-05', NULL),
  (4, 3, 2, 4, 'sum(open opportunity amount)', 0.90, '2026-06-02', '2026-06-02', NULL);

INSERT INTO aio_reviewer_feedback (feedback_id, claim_id, reviewer, new_confidence, reviewed_at) VALUES
  (1, 1, 'data_steward_1', 0.55, '2026-07-01');

VERIFIED loaded into named graph http://demo.openlinksw.com/schemas/aiontology at 85 triples. systemName/entityName/metricName each carry owl:equivalentProperty skos:prefLabel (authority-control reuse); fromSystem carries owl:equivalentProperty prov:wasDerivedFrom (provenance reuse, instead of a bespoke predicate). aio:MinedClaimShape is the axioms/constraints layer Gap 2 says was missing.

@prefix : <http://demo.openlinksw.com/schemas/aiontology/> .
@prefix owl: <http://www.w3.org/2002/07/owl#> .
@prefix skos: <http://www.w3.org/2004/02/skos/core#> .
@prefix prov: <http://www.w3.org/ns/prov#> .
@prefix sh: <http://www.w3.org/ns/shacl#> .
@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .

:SourceSystem a owl:Class ; rdfs:label "Source System"@en .
:BusinessEntity a owl:Class ; rdfs:label "Business Entity"@en .
:BusinessMetric a owl:Class ; rdfs:label "Business Metric"@en .
:MinedClaim a owl:Class ; rdfs:label "Mined Claim"@en .
:ReviewerFeedback a owl:Class ; rdfs:label "Reviewer Feedback"@en .

:systemName a owl:DatatypeProperty ; owl:equivalentProperty skos:prefLabel .
:entityName a owl:DatatypeProperty ; owl:equivalentProperty skos:prefLabel .
:metricName a owl:DatatypeProperty ; owl:equivalentProperty skos:prefLabel .
:fromSystem a owl:ObjectProperty ; owl:equivalentProperty prov:wasDerivedFrom .
:validFrom a owl:DatatypeProperty ; rdfs:range xsd:date .
:validTo   a owl:DatatypeProperty ; rdfs:range xsd:date .
:confidence a owl:DatatypeProperty ; rdfs:range xsd:float .

:MinedClaimShape a sh:NodeShape ;
    sh:targetClass :MinedClaim ;
    sh:property [ sh:path :confidence ; sh:minInclusive 0.0 ; sh:maxInclusive 1.0 ] ;
    sh:property [ sh:path :validFrom ; sh:lessThan :validTo ] .

VERIFIED valid Turtle (116 triples via rdflib) before upload. Five rr:TriplesMap instances — SourceSystemMap, BusinessEntityMap, BusinessMetricMap, MinedClaimMap (3 rr:parentTriplesMap joins: aboutEntity/aboutMetric/fromSystem), ReviewerFeedbackMap (1 join: feedbackOn) — mapping the same five DB.kidehen tables to the same aio: ontology as the hand-authored Quad Map below. Each map subject carries an explicit a rr:TriplesMap type, required by Virtuoso's R2RML processor. The five rr:TriplesMap subjects are asserted for real in this document's companion RDF (Turtle/JSON-LD) — not just quoted here as text.

@prefix rr: <http://www.w3.org/ns/r2rml#> .
@prefix aio: <http://demo.openlinksw.com/schemas/aiontology/> .
@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .

<#SourceSystemMap> a rr:TriplesMap ;
    rr:logicalTable [ rr:tableName "DB.kidehen.aio_source_system" ] ;
    rr:subjectMap [
        rr:template "http://demo.openlinksw.com/aiontology/source_system/{system_id}#this" ;
        rr:class aio:SourceSystem ;
    ] ;
    rr:predicateObjectMap [ rr:predicate aio:systemName ; rr:objectMap [ rr:column "system_name" ] ] ;
    rr:predicateObjectMap [ rr:predicate aio:systemType ; rr:objectMap [ rr:column "system_type" ] ] .

<#BusinessEntityMap> a rr:TriplesMap ;
    rr:logicalTable [ rr:tableName "DB.kidehen.aio_business_entity" ] ;
    rr:subjectMap [
        rr:template "http://demo.openlinksw.com/aiontology/business_entity/{entity_id}#this" ;
        rr:class aio:BusinessEntity ;
    ] ;
    rr:predicateObjectMap [ rr:predicate aio:entityName ; rr:objectMap [ rr:column "entity_name" ] ] ;
    rr:predicateObjectMap [ rr:predicate aio:entityType ; rr:objectMap [ rr:column "entity_type" ] ] .

<#BusinessMetricMap> a rr:TriplesMap ;
    rr:logicalTable [ rr:tableName "DB.kidehen.aio_business_metric" ] ;
    rr:subjectMap [
        rr:template "http://demo.openlinksw.com/aiontology/business_metric/{metric_id}#this" ;
        rr:class aio:BusinessMetric ;
    ] ;
    rr:predicateObjectMap [ rr:predicate aio:metricName ; rr:objectMap [ rr:column "metric_name" ] ] .

<#MinedClaimMap> a rr:TriplesMap ;
    rr:logicalTable [ rr:tableName "DB.kidehen.aio_mined_claim" ] ;
    rr:subjectMap [
        rr:template "http://demo.openlinksw.com/aiontology/mined_claim/{claim_id}#this" ;
        rr:class aio:MinedClaim ;
    ] ;
    rr:predicateObjectMap [ rr:predicate aio:formulaExpression ; rr:objectMap [ rr:column "formula_expression" ] ] ;
    rr:predicateObjectMap [ rr:predicate aio:confidence ; rr:objectMap [ rr:column "confidence" ; rr:datatype xsd:float ] ] ;
    rr:predicateObjectMap [ rr:predicate aio:observedAt ; rr:objectMap [ rr:column "observed_at" ; rr:datatype xsd:dateTime ] ] ;
    rr:predicateObjectMap [ rr:predicate aio:validFrom ; rr:objectMap [ rr:column "valid_from" ; rr:datatype xsd:date ] ] ;
    rr:predicateObjectMap [ rr:predicate aio:validTo ; rr:objectMap [ rr:column "valid_to" ; rr:datatype xsd:date ] ] ;
    rr:predicateObjectMap [
        rr:predicate aio:aboutEntity ;
        rr:objectMap [ rr:parentTriplesMap <#BusinessEntityMap> ;
                       rr:joinCondition [ rr:child "entity_id" ; rr:parent "entity_id" ] ] ;
    ] ;
    rr:predicateObjectMap [
        rr:predicate aio:aboutMetric ;
        rr:objectMap [ rr:parentTriplesMap <#BusinessMetricMap> ;
                       rr:joinCondition [ rr:child "metric_id" ; rr:parent "metric_id" ] ] ;
    ] ;
    rr:predicateObjectMap [
        rr:predicate aio:fromSystem ;
        rr:objectMap [ rr:parentTriplesMap <#SourceSystemMap> ;
                       rr:joinCondition [ rr:child "source_system_id" ; rr:parent "system_id" ] ] ;
    ] .

<#ReviewerFeedbackMap> a rr:TriplesMap ;
    rr:logicalTable [ rr:tableName "DB.kidehen.aio_reviewer_feedback" ] ;
    rr:subjectMap [
        rr:template "http://demo.openlinksw.com/aiontology/reviewer_feedback/{feedback_id}#this" ;
        rr:class aio:ReviewerFeedback ;
    ] ;
    rr:predicateObjectMap [ rr:predicate aio:reviewer ; rr:objectMap [ rr:column "reviewer" ] ] ;
    rr:predicateObjectMap [ rr:predicate aio:newConfidence ; rr:objectMap [ rr:column "new_confidence" ; rr:datatype xsd:float ] ] ;
    rr:predicateObjectMap [ rr:predicate aio:reviewedAt ; rr:objectMap [ rr:column "reviewed_at" ; rr:datatype xsd:dateTime ] ] ;
    rr:predicateObjectMap [
        rr:predicate aio:feedbackOn ;
        rr:objectMap [ rr:parentTriplesMap <#MinedClaimMap> ;
                       rr:joinCondition [ rr:child "claim_id" ; rr:parent "claim_id" ] ] ;
    ] .

VERIFIED: HTTP 201. Uploads the R2RML Turtle document so Virtuoso's own procedure can subsequently fetch it via the internal dav: scheme (see the generation call below) rather than an anonymous public HTTP GET.

curl -X PUT -u "kidehen:<password>" -H "Content-Type: text/turtle" \
  --data-binary @aiontology-r2rml.ttl \
  http://demo.openlinksw.com/DAV/home/kidehen/r2rml/aiontology-r2rml.ttl

VERIFIED: executed successfully, producing 20 auto-named sys:qm-* quad maps plus one urn:qm:* container, all added to virtrdf:DefaultQuadStorage, targeting the SAME graph the hand-authored mapping used (http://demo.openlinksw.com/aiontology#). Signature: R2RML_GENERATE_LINKED_VIEW(in source varchar, in destination_graph varchar, in graph_type int default 0, in clear_source_graph int default 1). source accepts file:, dav:, http:, https: schemes — dav: is used here (an internal-access scheme, not a public HTTP fetch, so no anonymous DAV read grant was needed). graph_type 0 = virtual graph (used here, matching the hand-authored view's own zero-copy semantics); 1 = physical. Doc reference (verified resolving via curl, HTTP 200): docs.openlinksw.com/virtuoso/r2rmlgenlviewisql.

DB.DBA.R2RML_GENERATE_LINKED_VIEW(
  'dav:/DAV/home/kidehen/r2rml/aiontology-r2rml.ttl',
  'http://demo.openlinksw.com/aiontology#',
  0
);

VERIFIED: re-running the Step 5 verification query and the SPARQL win-rate query against the R2RML-generated view returned byte-identical results to the hand-authored mapping's own version — 4 SourceSystem, 4 BusinessEntity, 4 MinedClaim, 2 BusinessMetric, 1 ReviewerFeedback; the same 3-row CRM/Cloud Warehouse/BI Tool contradiction with temporal validity attached. Equivalence is a demonstrated property of this deployment, not an assumption carried over from the hand-authored mapping's prior verification.

VERIFIED EXECUTED: "26 RDF metadata manipulation operations done". This is a valid alternative to the R2RML-driven deployment above: the same mapping, declared directly as Virtuoso's native ALTER QUAD STORAGE syntax rather than generated from a portable R2RML document. Entity IRI templates are confirmed live via SPARQL, though NOT yet HTTP-dereferenceable — no URL rewrite rule was configured for this demo.

SPARQL CREATE IRI CLASS <aio_source_system_iri> "http://demo.openlinksw.com/aiontology/source_system/%d#this" (in system_id integer not null) ;
SPARQL CREATE IRI CLASS <aio_business_entity_iri> "http://demo.openlinksw.com/aiontology/business_entity/%d#this" (in entity_id integer not null) ;
SPARQL CREATE IRI CLASS <aio_business_metric_iri> "http://demo.openlinksw.com/aiontology/business_metric/%d#this" (in metric_id integer not null) ;
SPARQL CREATE IRI CLASS <aio_mined_claim_iri> "http://demo.openlinksw.com/aiontology/mined_claim/%d#this" (in claim_id integer not null) ;
SPARQL CREATE IRI CLASS <aio_reviewer_feedback_iri> "http://demo.openlinksw.com/aiontology/reviewer_feedback/%d#this" (in feedback_id integer not null) ;

-- All FROM tables MUST be qualifier.schema.table -- a bare table name fails
-- with a misleading "Alias <table> is not defined" error.
SPARQL
ALTER QUAD STORAGE virtrdf:DefaultQuadStorage
  FROM DB.kidehen.aio_source_system AS s
  FROM DB.kidehen.aio_business_entity AS e
  FROM DB.kidehen.aio_business_metric AS m
  FROM DB.kidehen.aio_mined_claim AS c
  FROM DB.kidehen.aio_reviewer_feedback AS f
{
  GRAPH <http://demo.openlinksw.com/aiontology#> {
    <aio_source_system_iri> (s.system_id) rdf:type <http://demo.openlinksw.com/schemas/aiontology/SourceSystem> as virtrdf:aio_map_sys_type ;
      <http://demo.openlinksw.com/schemas/aiontology/systemName> s.system_name as virtrdf:aio_map_sys_name ;
      <http://demo.openlinksw.com/schemas/aiontology/systemType> s.system_type as virtrdf:aio_map_sys_stype .
    <aio_mined_claim_iri> (c.claim_id) rdf:type <http://demo.openlinksw.com/schemas/aiontology/MinedClaim> as virtrdf:aio_map_claim_type ;
      <http://demo.openlinksw.com/schemas/aiontology/formulaExpression> c.formula_expression as virtrdf:aio_map_claim_formula ;
      <http://demo.openlinksw.com/schemas/aiontology/confidence> c.confidence as virtrdf:aio_map_claim_conf ;
      <http://demo.openlinksw.com/schemas/aiontology/validFrom> c.valid_from as virtrdf:aio_map_claim_vf ;
      <http://demo.openlinksw.com/schemas/aiontology/validTo> c.valid_to as virtrdf:aio_map_claim_vt ;
      <http://demo.openlinksw.com/schemas/aiontology/fromSystem> <aio_source_system_iri> (s.system_id)
        where ( ^{c.}^.source_system_id = ^{s.}^.system_id ) as virtrdf:aio_map_claim_sys .
    <aio_reviewer_feedback_iri> (f.feedback_id) rdf:type <http://demo.openlinksw.com/schemas/aiontology/ReviewerFeedback> as virtrdf:aio_map_fb_type ;
      <http://demo.openlinksw.com/schemas/aiontology/feedbackOn> <aio_mined_claim_iri> (c.claim_id)
        where ( ^{f.}^.claim_id = ^{c.}^.claim_id ) as virtrdf:aio_map_fb_claim .
    -- (aboutEntity/aboutMetric/BusinessEntity/BusinessMetric mappings omitted here for
    -- length; the full 26-predicate statement is in the consolidated batch script below
    -- and in the companion TTL's :aioQuadMapDeployment entity.)
  }
} ;

VERIFIED real output, all five mapped classes present. Re-run against the R2RML-generated view and confirmed byte-identical (both deployment paths target the same graph, http://demo.openlinksw.com/aiontology#):

SPARQL SELECT ?type (SAMPLE(?entity) AS ?sampleEntity) (COUNT(?entity) AS ?entityCount)
FROM <http://demo.openlinksw.com/aiontology#>
WHERE { ?entity a ?type }
GROUP BY ?type ORDER BY DESC(?entityCount)

DDL + sample data + IRI classes + Quad Map + grants, in the order a fresh database needs them. The CREATE TABLE/ALTER TABLE/INSERT/GRANT portion is vanilla SQL and runs unmodified on effectively any SQL engine; the CREATE IRI CLASS/ALTER QUAD STORAGE portion is Virtuoso's own Linked Data Views syntax, VERIFIED on 08.03.3335, and the part most likely to need adjustment on a different version. The GRANT statements let this instance's public demo and vdb logins read the tables, so the live SQL/SPASQL links elsewhere in this document work for a visitor who authenticates as one of those instead of the owning account. Suggested filename aiontology.sql; suggested run command isql demo.openlinksw.com:21118 kidehen <password> < aiontology.sql (or paste into Virtuoso Conductor's Interactive SQL page). The ontology TBox above is loaded separately as Turtle, not part of this SQL batch. This script deploys the now-alternative hand-authored Quad Map path; see the R2RML batch script below for the separate, now-primary R2RML deployment script, kept deliberately distinct rather than folded into this one.

-- aiontology.sql -- run top to bottom against a fresh DB.kidehen schema
CREATE TABLE aio_source_system (system_id INTEGER NOT NULL PRIMARY KEY, system_name VARCHAR(64), system_type VARCHAR(32));
CREATE TABLE aio_business_entity (entity_id INTEGER NOT NULL PRIMARY KEY, entity_name VARCHAR(64), entity_type VARCHAR(32));
CREATE TABLE aio_business_metric (metric_id INTEGER NOT NULL PRIMARY KEY, metric_name VARCHAR(64));
CREATE TABLE aio_mined_claim (claim_id INTEGER NOT NULL PRIMARY KEY, entity_id INTEGER NOT NULL, metric_id INTEGER NOT NULL,
  source_system_id INTEGER NOT NULL, formula_expression VARCHAR(255), confidence FLOAT, observed_at DATETIME);
CREATE TABLE aio_reviewer_feedback (feedback_id INTEGER NOT NULL PRIMARY KEY, claim_id INTEGER NOT NULL, reviewer VARCHAR(64),
  new_confidence FLOAT, reviewed_at DATETIME);

ALTER TABLE aio_mined_claim ADD valid_from DATE;
ALTER TABLE aio_mined_claim ADD valid_to   DATE;

INSERT INTO aio_source_system VALUES (1,'Cloud Warehouse','warehouse'), (2,'CRM','crm'), (3,'BI Tool','bi_tool'), (4,'Metadata Catalog','catalog');
INSERT INTO aio_business_entity VALUES (1,'Opportunity','entity'), (2,'Deal','entity'), (3,'Pipeline','entity'), (4,'Customer','entity');
INSERT INTO aio_business_metric VALUES (1,'win_rate'), (2,'pipeline_value');
INSERT INTO aio_mined_claim VALUES
  (1,1,1,2,'closed_won / all closed opportunities',0.82,'2026-06-01','2026-06-01','2026-06-30'),
  (2,1,1,1,'closed_won / closed opportunities excluding unqualified',0.76,'2026-06-03','2026-06-03',NULL),
  (3,1,1,3,'closed_won / new-business pipeline only',0.65,'2026-06-05','2026-06-05',NULL),
  (4,3,2,4,'sum(open opportunity amount)',0.90,'2026-06-02','2026-06-02',NULL);
INSERT INTO aio_reviewer_feedback VALUES (1,1,'data_steward_1',0.55,'2026-07-01');

GRANT SELECT ON aio_source_system TO "demo";
GRANT SELECT ON aio_source_system TO "vdb";
GRANT SELECT ON aio_business_entity TO "demo";
GRANT SELECT ON aio_business_entity TO "vdb";
GRANT SELECT ON aio_business_metric TO "demo";
GRANT SELECT ON aio_business_metric TO "vdb";
GRANT SELECT ON aio_mined_claim TO "demo";
GRANT SELECT ON aio_mined_claim TO "vdb";
GRANT SELECT ON aio_reviewer_feedback TO "demo";
GRANT SELECT ON aio_reviewer_feedback TO "vdb";

SPARQL CREATE IRI CLASS <aio_source_system_iri> "http://demo.openlinksw.com/aiontology/source_system/%d#this" (in system_id integer not null) ;
SPARQL CREATE IRI CLASS <aio_business_entity_iri> "http://demo.openlinksw.com/aiontology/business_entity/%d#this" (in entity_id integer not null) ;
SPARQL CREATE IRI CLASS <aio_business_metric_iri> "http://demo.openlinksw.com/aiontology/business_metric/%d#this" (in metric_id integer not null) ;
SPARQL CREATE IRI CLASS <aio_mined_claim_iri> "http://demo.openlinksw.com/aiontology/mined_claim/%d#this" (in claim_id integer not null) ;
SPARQL CREATE IRI CLASS <aio_reviewer_feedback_iri> "http://demo.openlinksw.com/aiontology/reviewer_feedback/%d#this" (in feedback_id integer not null) ;

-- Full ALTER QUAD STORAGE statement (26 mapping predicates, one per target
-- graph as required) is in the companion TTL's :aioQuadMapDeployment entity.

A second, separate consolidated script for the R2RML deployment path — it does not replace aiontology.sql above, which remains the correct script for the hand-authored ALTER QUAD STORAGE path. Runs the same DDL/data/grants as that script's opening section (vanilla SQL, any engine), then uploads the R2RML mapping document via curl PUT, then calls DB.DBA.R2RML_GENERATE_LINKED_VIEW, with the RDF_VIEW_DROP_STMT_BY_GRAPH safety check commented in as the recommended first step on any redeploy against a non-empty target graph. Suggested filename aiontology-r2rml-deploy.sh; the SQL portions are intended to be piped through isql (isql demo.openlinksw.com:21118 kidehen <password>), the curl portion runs as shown from any shell.

#!/bin/sh
# aiontology-r2rml-deploy.sh -- R2RML deployment path, end to end
# Run the DDL/data/grants portion via isql first (identical to aiontology.sql's
# opening section above), then:

# 1. (Redeploy only) If the target graph already has ANY quad-map registered
#    against it, drop every one first:
#    SELECT RDF_VIEW_DROP_STMT_BY_GRAPH('http://demo.openlinksw.com/aiontology#');
#    -- run whatever it returns, then verify the graph is empty (COUNT(*) = 0).

# 2. Upload the R2RML mapping document (aiontology-r2rml.ttl) to a
#    Virtuoso-fetchable WebDAV path:
curl -X PUT -u "kidehen:<password>" -H "Content-Type: text/turtle" \
  --data-binary @aiontology-r2rml.ttl \
  http://demo.openlinksw.com/DAV/home/kidehen/r2rml/aiontology-r2rml.ttl

# 3. Ask Virtuoso to generate the Quad Map from it (run via isql):
#    DB.DBA.R2RML_GENERATE_LINKED_VIEW(
#      'dav:/DAV/home/kidehen/r2rml/aiontology-r2rml.ttl',
#      'http://demo.openlinksw.com/aiontology#',
#      0
#    );

# 4. Verify (see the Step 5 query and the SPARQL win-rate query below) --
#    results are byte-identical to the hand-authored Quad Map path.