RDFLib Reasoning Engine

Agentic AI should be able to take advantage of basic reasoning in an efficient manner.

RDFS and OWL 2 RL reasoning define the semantics for two entailment regimes. These regimes can be computed using forward-chaining rules, though practical approaches may also benefit from special handling for transitive closures. The rdflib-reasoning-engine package provides an efficient general purpose forward chaining rule engine (i.e., RETE) that works with RDFLib.

Tutorials and demos

If you want to see the current RDFS inference surface in action, start with:

• ../notebooks/demo-rdfs-inference.ipynb: compact end-to-end tutorial for RDFLib-backed RDFS materialization plus a proof view of the rule applications behind one inferred triple
• ../notebooks/demo-rdfs-retraction.ipynb: ordinary RDFLib graph.remove(...) usage with dependency-aware retraction of inferred triples
• ../notebooks/demo-contradiction-rules.ipynb: dual-channel OWL 2 RL contradiction diagnostics (ContradictionRecord), recorder policies, contradiction-goal proof reconstruction, and rendering
• ../notebooks/demo-proof-reconstructor.ipynb: proof-focused companion notebook covering proof reconstruction, Mermaid rendering, markdown rendering, and raw proof inspection

The RDFLib integration is intentionally ordinary at the call site: adding triples materializes supported conclusions, and removing asserted triples retracts dependent conclusions through the same Graph API. Contradiction diagnostics are likewise engine-managed when the active ruleset includes contradiction-producing rules; if no recorder is supplied, the default policy records the contradiction and raises.

Feature Matrix

Status values:

• Implemented: available in the package today
• In progress: scaffolding or partial implementation exists, but the feature is not yet wired end-to-end
• Not started: identified feature target with no concrete implementation yet

RDFS entailment rules

This matrix tracks the intermediate RDFS rule target using the cached RDF 1.1 Semantics specification at docs/specs/rdf11-semantics/optimized.html. The informative RDFS entailment rules are identified there as rdfs1 through rdfs13.

OWL 2 RL contradiction diagnostics (false-family, non-mutating)

These rules do not materialize new entailment triples into logical closure. They emit ContradictionConsequent actions and ContradictionRecord diagnostics (dual-channel model per DR-027). Coverage is limited by the current rule IR and builtins; see rdflib_reasoning.engine.rulesets.owl2_rl_contradictions. Each row tracks exactly one OWL 2 RL rule identifier from docs/specs/owl2-reasoning-profiles/INDEX.md.

OWL 2 RL entailment rules (triple materialization)

The authoritative local reference for this matrix is docs/specs/owl2-reasoning-profiles/INDEX.md, which indexes Section 4.3 of the cached OWL 2 Profiles specification. Status here refers to forward-chaining materialization of OWL 2 RL entailment rules. Rule identifiers that have false consequents in the specification are listed under OWL 2 RL contradiction diagnostics above. Each row tracks exactly one OWL 2 RL rule identifier.

Engine capabilities

This matrix tracks functional engine features independent of standards coverage.

Current integration baseline

The supported RDFLib integration path is Store events -> BatchDispatcher -> RETEStore -> RETEEngine.

The rdflib-reasoning-engine package is part of the rdflib-reasoning metapackage.