Estate Catalog Ontology — data / model / asset / service / infrastructure (reason over the estate, don't read it)

[mdheller]

Vision

Make the whole estate ontological so agents (and humans) reason over a graph — search it, understand dependencies, answer cross-cutting questions — instead of reading every repo/config each time. Same thesis as the Model Plane ontology (#129), generalized to every estate resource.

Architecture — one core + five domains (all ontogenesis, rooted in upper-core, sourceos vocab namespaces)

Domains/estate-catalog/estate-catalog-core.ttl (cat: = https://spec.sourceos.dev/vocab/estate-catalog#)

• cat:CatalogEntry (⊑ upper:Entity) — abstract root of every cataloged thing: catalogId (URN), title, owner, status {active|deprecated|retired}, lifecycle, provenanceRef, evidenceRef.
• cat:Catalog — a named collection.
• Cross-catalog object properties = the reasoning fabric: cat:dependsOn, cat:runsOn, cat:consumesData, cat:producesData, cat:usesModel, cat:composedOf, cat:deployedTo, cat:exposes, cat:storedIn, cat:governedBy.

Five domains (each ⊑ cat:CatalogEntry):

1. Data catalog (dc:) — DataAsset (dataset|stream|table|index|corpus), Schema, DataProduct; sensitivityClass, residencyClass, retention, derivedFrom (lineage).
2. Model catalog (mc:) — Model, ModelAdapter, ModelVersion. Extends the merged model-plane domain (feat(domains): model-plane OWL/SHACL domain module (T7-7/T7-8) #129): mc:Model ↔ sourceos-spec ModelManifest by digest; mc:servedBy → mp:InferenceProvider; carries license (MIT/Apache checkable), tier, task.
3. Asset catalog (ac:) — Asset (artifact|image|package|binary|document); content-addressed digest, storageRef (zot/GCS/MinIO), signature, license. Ties to existing PackageManifest/ArtifactManifest/ReleaseManifest.
4. Application & Service catalog (sc:) — Application, Service, API, Endpoint; repo, environment, dependsOn, runsOn, consumesData, usesModel, exposes.
5. Infrastructure (inf:) — InfrastructureNode, Cluster, Environment {dev|staging|prod}, Registry, Runner, Network, StorageBucket, ComputeResource; provider (GKE|GCS|sovereign), region, capacity; hosts, stores.

Reasoning payoff (SPARQL/SHACL over the graph)

• "Which apps consume biometric/health-sensitive data and run on non-sovereign infra?" → Service —consumesData→ DataAsset(sensitivity) ⋈ Service —runsOn→ Infra(provider).
• "Which models under a non-Apache license are served by a provider that can escalate off-device?" → mc:Model(license) ⋈ mc:servedBy→ mp:InferenceProvider(escalationPermitted) — reuses the feat(domains): model-plane OWL/SHACL domain module (T7-7/T7-8) #129 residency SHACL.
• "Blast radius of retiring StorageBucket X" → traverse storedIn/dependsOn.
  Cross-catalog SHACL integrity shapes (e.g. a biometric DataAsset consumed by a Service on non-on-device infra = violation) make these invariants, not just queries.

Phasing (each on the review→remediate→merge loop)

• P1 (schema): core + five domain OWL + key SHACL integrity/cross-catalog shapes + good/bad fixtures + per-lane validator. (One PR per domain, or core+2 then +3.)
• P2 (population): ingest instances — gh-inventory (~70 repos → service catalog), model store → model catalog, GCS/zot/GKE → asset+infra. Ingestion tranche, not schema.
• P3 (query surface): SPARQL + the existing shapes/ontology-query ShapeQuery mechanism as the agent-facing "reason, don't read" API.

Notes

• Generalizes catalog/registry.ttl (og:Module) which already catalogs ontology modules — same pattern, all resource types.
• Register each module in catalog/registry.ttl; validate via shacl_gate.py + per-lane validators (same as agent-system / model-plane).
• Depends on: model-plane domain (feat(domains): model-plane OWL/SHACL domain module (T7-7/T7-8) #129, merged). Related estate context: KKO upper ontology (not yet wired), gh-inventory.

[mdheller]

Reconciliation before build (audit findings)

Do NOT fork — two existing pieces the estate catalog must bind to:

1. SocioProphet/prophet-core-catalog already exists = the canonical catalog spine (data products, sources, app integrations, lineage receipts) with per-repo topic graphs already in datasets/topic-vocabulary/graphs/*.ttl (sourceos-model-carry, workstation-contracts, exodus, mellumwork, prophet-workspace, …). This is the population/extraction + repo-ontology layer. The ontogenesis estate-catalog ontology is the shared vocabulary it binds to — not a competing catalog.
2. KBpedia/KKO is real at hellgraph/ontology/kko/kko-2.10.n3 (same ontology HellGraph uses). Estate vocabulary terms MUST subsume into KBpedia so estate ↔ HellGraph share one binding language. ontogenesis currently vendors only KKO-upper (169 terms) — full KBpedia alignment needed.

Revised architecture (dogfooding): KBpedia (hellgraph) = upper language → ontogenesis = estate binding vocabulary (cat:/dc:/mc:/ac:/sc:/inf:, rooted in KKO, extends model-plane domain #129) → repo READMEs/per-folder docs = repo self-description → prophet-core-catalog extractors populate instances typed by the vocabulary → estate graph. 'Live' test = a real cross-catalog SPARQL answer over real extracted instances, terms resolving into KBpedia.

Next: recon prophet-core-catalog's existing schemas/vocab + KBpedia/KKO structure, then author the KKO-rooted estate-catalog vocabulary aligned with both (not greenfield).

[mdheller]

Namespace correction: the epic earlier wrote cat: = https://spec.sourceos.dev/vocab/estate-catalog#. Per docs/specs/namespaces.md, that spec.sourceos.dev namespace is for sourceos-spec realizations (agent-system, model-plane). estate-catalog is a generic ontogenesis binding domain that unifies the ontogenesis-native fragments (dom:/drm:/srm:/enrm:/legal:), so the correct, consistent namespace is https://socioprophet.github.io/ontogenesis/domains/estate-catalog# — which is what PR #131 uses. Epic namespace amended to match.

[mdheller]

✅ P1 keystone merged — PR #131 (squash). The estate-catalog binding vocabulary is live on main: cat:EstateResource roof + cross-catalog reasoning properties + the new Model catalog (joined to model-plane #129), binding the existing fragments (dom:/drm:/srm:/enrm:/legal:/mp:) rather than forking them, KKO-grounded via skos:closeMatch (same KBpedia HellGraph uses). SHACL teeth (owner/status/modelDigest/license) verified both ways; global gate green.

Remaining on #130: (P1 depth) governed Alignments/kko.ttl of smap:MappingAssertion records + deeper per-catalog classes; (P2) population — wire prophet-core-catalog extractors to emit instances typed by this vocabulary (needs repo self-docs coverage — the dogfooding substrate); (P3) SPARQL/ShapeQuery agent surface = the 'live' proof (a real cross-catalog query over real instances). Reconciliation targets: prophet-core-catalog (ABox layer), agent-machine's own inference-provider.schema.json (divergence).

[mdheller]

✅ P2 population landed — prophet-core-catalog #23 (squash). The estate graph is now real and queryable: 151 sources/src.*.json → a cat:-typed, KKO-grounded RDF graph (datasets/estate-graph/estate-graph.ttl), all 151 instances SHACL-conform to the vendored estate-catalog vocabulary, and a real SPARQL query answers over it (active entries by owner: SocioProphet 148, + a real third-party finding 'NumPy Developers'). Wired into make validate with a freshness guard (stale graph fails CI). prophet-core-catalog = ABox/population; ontogenesis = TBox — the dogfooding split is live.

Still open (#130): P3 depth — cross-catalog EDGES (dependsOn/runsOn/consumesData/usesModel) from ds.agents-manifests + ds.ontologies import-graph so blast-radius/where-runs traversals go live; cat:Model typing once modelDigest available; governed Alignments/kko.ttl; vendor re-sync on vocab change.

[mdheller]

✅ P3 dependency edges + blast-radius reasoning landed — prophet-core-catalog #26. The estate graph is now a real dependency graph: 495 cat:dependsOn edges from models.used_by / services.consumers / ontologies.imports, joined to the P2 source nodes (one graph, 979 nodes / 5211 triples), all SHACL-conform. Blast-radius queries are live — real finding: upper-core.ttl is the estate's highest-blast-radius artifact (64 dependents; transitive too). 'If this breaks, what breaks' is now a graph traversal, not repo-reading.

Remaining gaps (#130): (1) cat:Model typing — model refs carry no modelDigest, so they're cat:EstateResource dct:subject "model"; typing as cat:Model awaits the model-store/model-plane digest join. (2) Richer typed edges (consumesData/runsOn/usesModel) — held as generic dependsOn to stay SHACL-safe under rdfs range-coercion. (3) The graph is a CI-regenerated snapshot, not a live service. (4) A governed Alignments/kko.ttl. These are enhancements; the reason-over-the-estate capability is now real.

[mdheller]

✅ Gap (a) closed — prophet-core-catalog #27 merged. Model refs the estate actually references are now typed cat:Model with real modelDigests (from local ollama registry manifests) joined onto the depended-on res:mdl-* nodes, so the MIT/Apache-only rule reasons over the graph: 4 non-compliant models flagged with dependent counts (license × blast radius — 2 Llama-3.2 models with 3 dependents each). Honest boundary: digest = real content address; license = model-card-declared; no weights present locally.