Triples: RDF from first principles

RDF describes the world as a set of triples: subject, predicate, object. Nothing else. No rows, no columns, no nesting — every fact is one small statement, and a graph is just a set of them. This post builds that idea up from the actual specification this project verifies, then parses a real example live, in your browser, against the F*-extracted engine.

Three kinds of term#

Every triple is built from terms, and RDF 1.1 recognizes exactly three disjoint kinds of term. This project's F* specification says so directly — RDF.Term.fsti is written to be read by a human who knows RDF but not F*, one concept per block:

noeq type rdf_term =
  | T_IRI     : wf_iri -> rdf_term
  | T_BNode   : bnode_id -> rdf_term
  | T_Literal : wf_literal -> rdf_term

IRI (T_IRI) — a global name for something, e.g. http://xmlns.com/foaf/0.1/knows. RDF.Term.fsti's wf_iri type refines a plain string down to "non-empty and contains a colon" — the cheap structural check every term constructor runs; full RFC 3987 grammar conformance is a separate module's job (RDF.IRI.fst).
Blank node (T_BNode) — a locally-scoped identifier for "something exists here, I'm just not naming it." Disjoint from IRIs and literals.
Literal (T_Literal) — a value: a lexical form, a datatype IRI, and — only when the datatype is rdf:langString — a language tag. RDF.Term.fsti's literal_wf predicate encodes that "iff" directly: a literal has a language tag exactly when its datatype is rdf:langString, never otherwise.

A triple is subject–predicate–object, where the subject is restricted to an IRI or a blank node (never a literal — RDF 1.1 Concepts §3.1), the predicate is always an IRI, and the object can be any of the three term kinds. A graph is a set of triples.

Parsing one, live#

Here's a small foaf graph: Alice knows Bob.

@prefix foaf: <http://xmlns.com/foaf/0.1/> .
@prefix ex:   <http://example.org/> .

ex:alice a foaf:Person ;
  foaf:name  "Alice" ;
  foaf:knows ex:bob .

ex:bob a foaf:Person ;
  foaf:name "Bob" .

That's three statements about Alice (type, name, knows-Bob) and two about Bob — five triples in total. The a keyword is Turtle (and SPARQL) shorthand for the rdf:type predicate, so ex:alice a foaf:Person is the type triple, not an extra one. Parse it with the same F*-extracted Turtle parser the W3C conformance suite runs against (Parser.Turtle.fst, 313 pass, 0 fail on the rdf-turtle suite — see the plan doc for the full RDF 1.1 scorecard). The cell below runs that parse right now, in this page, using the fn typed API (see README.md for the full cell-authoring contract):

turtle = `
  @prefix foaf: <http://xmlns.com/foaf/0.1/> .
  @prefix ex:   <http://example.org/> .

  ex:alice a foaf:Person ;
    foaf:name  "Alice" ;
    foaf:knows ex:bob .

  ex:bob a foaf:Person ;
    foaf:name "Bob" .
`

const dataset = await fn.parse(turtle);
return dataset.size;

Five triples, exactly as counted above. Each one is a {subject, predicate, object} term triple — pretty() (see README.md) renders a Dataset as an s/p/o table directly, term values shortened for display. This cell references the turtle cell above by name instead of repeating the Turtle text — post 26 covers the reactive cell-naming convention in full:

const dataset = await fn.parse(turtle);
return pretty(dataset);

The o column shows quoted literals ("Alice", "Bob") alongside shortened IRIs (Person, bob) — hover any IRI cell for the full value in a tooltip. That's T_IRI vs T_Literal from RDF.Term.fsti, made concrete.

The third kind: blank nodes#

Turtle's [] syntax introduces a blank node — "some person, unnamed":

@prefix foaf: <http://xmlns.com/foaf/0.1/> .
[] a foaf:Person ; foaf:name "Anonymous Friend" .

Turtle also allows an empty prefix, which makes small examples shorter still — the live cell below uses it:

const ds2 = await fn.parse(`
  @prefix : <http://xmlns.com/foaf/0.1/> .
  [] a :Person ; :name "Blank Friend" .
`);
const quads = [...ds2];
return {
  size: ds2.size,
  subjectTermType: quads[0].subject.termType,
  sameSubjectOnBothTriples: quads[0].subject.value === quads[1].subject.value,
};

Two triples, both with the same blank-node subject — T_BNode in the F* term algebra, BlankNode in RDF/JS. That's all three term kinds now seen live: NamedNode/T_IRI, BlankNode/T_BNode, Literal/T_Literal.

What's next#

The next post in this series asks questions of a graph instead of just reading it — SPARQL SELECT, ASK, CONSTRUCT, and property paths over a small Wikidata-shaped dataset. After that, RDFS and OWL 2 RL show how a graph can imply triples nobody asserted.

Every live cell above is pinned in tests/hub/post01_test.mjs — the exact same source, executed against the real npm/factoidal typed API instead of the in-browser fn adapter.