scripts/extract_vocab.py

#!/usr/bin/env python3
"""Extract the authoritative SSAO term set: the membership oracle for validation.

The Space Situational Awareness Ontology (Rovetto, as vendored by the NASA
mission-viz project) is the target vocabulary. Every ssao: term the model emits
must exist here, or it is a hallucination by construction.
"""
import json
import pathlib

from rdflib import Graph, RDF, RDFS, OWL, Namespace

ROOT = pathlib.Path("/Users/fabio/projects/qwen-space-ft")
KGREPO = pathlib.Path("/Users/fabio/projects/neurosymbolic-space-kg")
SSAO = "https://purl.org/space-ontology/"
SKOS = Namespace("http://www.w3.org/2004/02/skos/core#")

g = Graph()
g.parse(KGREPO / "ontology" / "SSAO_Rovetto.owl", format="turtle")


def local(u):
    s = str(u)
    return s.split("#")[-1].split("/")[-1]


def in_ns(u):
    return str(u).startswith(SSAO)


classes = {}
for s in set(g.subjects(RDF.type, OWL.Class)) | set(g.subjects(RDF.type, RDFS.Class)):
    if not in_ns(s):
        continue
    ln = local(s)
    label = g.value(s, RDFS.label)
    defn = g.value(s, SKOS.definition) or g.value(s, RDFS.comment)
    supers = [local(o) for o in g.objects(s, RDFS.subClassOf) if in_ns(o)]
    classes[ln] = {
        "label": str(label) if label else ln.replace("_", " "),
        "definition": str(defn)[:400] if defn else "",
        "subClassOf": supers,
    }

props = {}
for s in (set(g.subjects(RDF.type, OWL.ObjectProperty))
          | set(g.subjects(RDF.type, OWL.DatatypeProperty))
          | set(g.subjects(RDF.type, RDF.Property))):
    if not in_ns(s):
        continue
    ln = local(s)
    props[ln] = {
        "definition": str(g.value(s, SKOS.definition) or g.value(s, RDFS.comment) or "")[:300],
        "domain": [local(o) for o in g.objects(s, RDFS.domain) if in_ns(o)],
        "range": [local(o) for o in g.objects(s, RDFS.range) if in_ns(o)],
    }

# authoritative membership set: every ssao-namespaced term appearing anywhere
all_terms = sorted({local(n) for t in g for n in t if in_ns(n)})

# our own lifted catalogue vocabulary (the source side of the published alignment)
kgv = Graph()
kgv.parse(KGREPO / "kg" / "out" / "satcat-vocab.ttl", format="turtle")
KGNS = "https://w3id.org/tesseract/space-kg/"
kg_terms = sorted({local(s) for s in kgv.subjects(RDF.type, OWL.Class)
                   if str(s).startswith(KGNS)})

vocab = {
    "ssao_namespace": SSAO,
    "kg_namespace": KGNS,
    "classes": classes,
    "properties": props,
    "class_names": sorted(classes),
    "property_names": sorted(props),
    "all_terms": all_terms,
    "kg_class_names": kg_terms,
}
(ROOT / "data").mkdir(exist_ok=True)
(ROOT / "data" / "vocab.json").write_text(json.dumps(vocab, indent=1))
print(f"SSAO: {len(classes)} classes, {len(props)} properties, {len(all_terms)} terms total")
print(f"lifted catalogue vocabulary: {len(kg_terms)} classes")
print("sample classes:", ", ".join(sorted(classes)[:8]))