LUBM Conformance Results

This page summarises pg_ripple's results on the Lehigh University Benchmark (LUBM), a canonical benchmark for OWL RL knowledge-base systems.

Overview

LUBM (Guo et al., 2005) defines 14 canonical SPARQL queries over a synthetic university-domain ontology (univ-bench.owl). The queries exercise:

rdf:type lookups with OWL RL subclass/subproperty entailment
Multi-hop property chains (memberOf + subOrganizationOf + undergraduateDegreeFrom)
Domain and range reasoning
Conjunctive patterns over asserted and inferred triples

As of v0.44.0, all 14 LUBM queries pass against the bundled univ1 synthetic fixture with 0 known failures.

pg_ripple uses a self-contained synthetic fixture (tests/lubm/fixtures/univ1.ttl) rather than the original Java UBA generator. The fixture models 1 university, 1 department, 1 research group, 4 faculty, 7 graduate students, 5 undergraduate students, 6 graduate courses, 1 undergraduate course, and 4 publications — all with explicit supertype assertions so that no OWL RL inference pass is required to match the reference counts.

A complementary Datalog validation sub-suite (tests/lubm/datalog/) separately validates that running pg_ripple.load_rules_builtin('owl-rl') and pg_ripple.infer('owl-rl') on an implicit-type-only version of the same data produces identical query results.

Query Results (univ1 fixture)

Query	Description	Inference rules exercised	Expected	Result	Status
Q1	Graduate students taking GraduateCourse0	`rdf:type` + subclass entailment	3	3	✅ PASS
Q2	Graduate students whose department is part of their undergrad university	Multi-hop join	2	2	✅ PASS
Q3	Publications by AssistantProfessor0	Direct lookup	2	2	✅ PASS
Q4	Professors in Dept0 with name/email/phone	Property star pattern	4	4	✅ PASS
Q5	Persons in Dept0 taking GraduateCourse0	`ub:Person` superclass	3	3	✅ PASS
Q6	All students	`ub:Student` superclass	12	12	✅ PASS
Q7	Students taking GC0 advised by a FullProfessor	Advisor + course conjunction	3	3	✅ PASS
Q8	Students in Dept0/University0 with email	Full join pattern	12	12	✅ PASS
Q9	Students taking courses from AssistantProfessors	3-way join	7	7	✅ PASS
Q10	Graduate students in ResearchGroup0	`ub:memberOf` lookup	3	3	✅ PASS
Q11	Sub-organizations of Department0	`ub:subOrganizationOf`	1	1	✅ PASS
Q12	Professors heading Department0	`ub:headOf`	1	1	✅ PASS
Q13	Professors acting as teaching assistants	`ub:teachingAssistantOf`	1	1	✅ PASS
Q14	All undergraduate students	`ub:UndergraduateStudent` subclass	5	5	✅ PASS

Datalog Validation Sub-suite

The Datalog sub-suite validates the OWL RL inference engine independently of the SPARQL translator.

Test	What it validates	Status
Rule compilation	`load_rules_builtin('owl-rl')` compiles ≥ 20 rules	✅ PASS
Inference iteration	`infer_with_stats()` reaches fixpoint in 1–10 iterations	✅ PASS
Inferred triple counts	Key supertype entailments produce correct row counts	✅ PASS
Goal queries	`infer_goal()` and SPARQL counts agree for Q1/Q6/Q14	✅ PASS
Materialization perf	`infer('owl-rl')` completes in < 5 s on univ1	✅ PASS
Custom rules	User-defined transitive-closure rule works correctly	✅ PASS

Running LUBM Locally

# Start pg_ripple (uses pgrx default port 28818)
cargo pgrx start pg18

# Run the LUBM suite (self-contained — no data download required)
cargo test --test lubm_suite -- --nocapture

To run the Datalog sub-suite SQL files manually:

# Assumes pg_ripple is installed and running
psql -c "SELECT pg_ripple.load_turtle(pg_read_file('tests/lubm/fixtures/univ1.ttl'), false)"
psql -f tests/lubm/datalog/rule_compilation.sql
psql -f tests/lubm/datalog/inference_iterations.sql
psql -f tests/lubm/datalog/inferred_triples.sql

Adding Known Failures

If a LUBM query fails, add a lubm:Q{N} entry to tests/conformance/known_failures.txt:

# Example — Q2 fails due to multi-hop join bug, fix in progress
lubm:Q2  multi-hop memberOf/subOrganizationOf join returns wrong count

Remove the entry once the underlying bug is fixed.

pg_ripple