RDF, RDF 1.2, and RDF-star

Why SutraDB uses RDF-star — the most expressive superset of the W3C data model.

The Three Levels of RDF

The RDF data model has evolved through three stages. Each is a strict superset of the one before:

RDF-star   ⊃   RDF 1.2   ⊃   Legacy RDF 1.1

Legacy RDF 1.1

The original W3C standard. Data is organized as triples: (subject, predicate, object). Simple and powerful, but with one major limitation — if you want to make a statement about a statement, you need reification, which is verbose and awkward:

-- The fact:
:Alice :knows :Bob .

-- To say "Alice knows Bob with 90% confidence" in legacy RDF:
:statement1 rdf:type rdf:Statement .
:statement1 rdf:subject :Alice .
:statement1 rdf:predicate :knows .
:statement1 rdf:object :Bob .
:statement1 :confidence 0.9 .
-- That's 5 triples to annotate one fact!

RDF 1.2 (W3C Working Draft)

The W3C's in-progress update to RDF. Introduces triple terms — a triple can be referenced as a value, but only in the object position:

-- The fact:
:Alice :knows :Bob .

-- In RDF 1.2, annotate via rdf:reifies:
:statement1 rdf:reifies << :Alice :knows :Bob >> .
:statement1 :confidence 0.9 .
-- Better (3 triples), but the triple term can only be an object
RDF 1.2 restriction Triple terms can only appear in object position. You cannot write << :Alice :knows :Bob >> :confidence 0.9 directly — you need the intermediate rdf:reifies node.

RDF-star (What SutraDB Uses)

RDF-star removes the restriction. Triple terms can appear in any position — subject, predicate, or object. This is the most natural way to annotate triples:

-- The fact:
:Alice :knows :Bob .

-- In RDF-star, annotate directly:
<< :Alice :knows :Bob >> :confidence 0.9 .
-- Just 2 triples. The triple IS the subject.

This is especially powerful for vector embeddings on edges:

-- An embedding on a relationship (RDF-star)
<< :paper42 :discusses :TransformerArchitecture >>
    :hasEmbedding "0.23 -0.11 0.87 ..."^^sutra:f32vec .
<< :paper42 :discusses :TransformerArchitecture >>
    :confidence 0.91 .

Comparison Table

FeatureLegacy RDFRDF 1.2RDF-star
Basic triples (S, P, O) Yes Yes Yes
Triple terms as objects No Yes Yes
Triple terms as subjects No No Yes
Direct edge annotation Reification (verbose) Via rdf:reifies Native
Triples to annotate one fact 5 3 2

Why RDF-star?

SutraDB's primary use case is storing vector embeddings alongside graph structure. The << s p o >> :hasEmbedding ... pattern is the natural representation for "this edge has an embedding." RDF 1.2's object-only restriction would require an extra indirection node for every annotated edge.

Full compatibility Since RDF-star is a superset of RDF 1.2, which is a superset of legacy RDF, any existing RDF data works in SutraDB without modification. You only need RDF-star syntax when you want to annotate triples directly.

Practical impact

How SutraDB Stores RDF-star

Quoted triples are handled efficiently using content-addressed hashing:

  1. Each IRI and literal is interned to a 64-bit integer ID at write time
  2. A quoted triple << S P O >> is hashed as xxHash3(S_id, P_id, O_id) → u64
  3. This hash becomes the triple term's ID, usable in any position in another triple
  4. All indexes (SPO/POS/OSP) operate on u64 IDs — no string comparisons on the hot path

Collision probability for xxHash3 on u64 is negligible at any realistic graph size.

Oxigraph's Different Choice

Oxigraph, the closest existing Rust triplestore, migrated from RDF-star to RDF 1.2 in v0.5. Their rationale was alignment with the W3C working draft. SutraDB took the opposite position: RDF-star's direct edge annotation is essential for the vector embedding use case, and since it's a superset, RDF 1.2 compatibility comes for free.