# Agent KB Pack

This file explains how another agent should call the ontology research KB without reading the whole repository first.

## Metrics

- source_count: 803
- core_sources: 150
- candidate_sources: 332
- low_priority_sources: 321
- selected_sources: 382
- chunks: 2157
- claims: 1307
- entities: 1804
- relations: 16721

## Entry Points

- `research/AGENT_GUIDE.md`: human-readable retrieval protocol.
- `research/sources/triaged_sources.jsonl`: all source records with buckets, scores, and quality signals.
- `research/chunks/chunks.jsonl`: retrieval chunks.
- `research/extractions/claims.jsonl`: atomic claims.
- `research/graph/entities.jsonl` and `research/graph/relations.jsonl`: object-relation graph.
- `research/indexes/source_index.json` and `research/indexes/concept_index.json`: fast lookup indexes.
- `research/synthesis/evidence_matrix.json`: article routes and source IDs.
- `research/synthesis/nature_submission_pack.md`: Nature-facing writing pack.

## Query Recipes

### Palantir ontology architecture

Retrieve official Palantir architecture evidence about Ontology, AIP, OMCP, actions, permissions, evals, and observability.

```powershell
python scripts\search_kb.py "Palantir AIP ontology action type MCP evals observability" --concept palantir_ontology --limit 8
```

### Ontology foundations

Retrieve foundational ontology definitions and design principles.

```powershell
python scripts\search_kb.py "explicit specification conceptualization formal ontology knowledge sharing" --concept ontology --limit 8
```

### Scientific ontology infrastructure

Retrieve scientific ontology infrastructure evidence for Nature-style argumentation.

```powershell
python scripts\search_kb.py "Gene Ontology OBO Foundry ROBOT ODK HPO Monarch DeepGO-SE" --concept scientific_ontology_infrastructure --limit 10
```

### Ontology evaluation and lifecycle governance

Retrieve Phase 16 sources about ontology conceptual quality, pitfall scanning, evolution, reporting, reuse, and lifecycle governance.

```powershell
python scripts\search_kb.py "OntoClean OOPS ontology evolution MIRO NeOn FAIRsharing" --concept ontology_quality --limit 10
```

### Semantic data infrastructure

Retrieve Phase 14 sources about typed schema, ontology mappings, ML dataset metadata, OBDA/VKG, and machine-actionable metadata.

```powershell
python scripts\search_kb.py "LinkML SSSOM Croissant Wikidata OBDA FAIR Data Point virtual knowledge graph" --concept semantic_schema --limit 10
```

### Mappings and dataspaces

Retrieve Phase 15 sources about data-to-RDF mappings, KG construction pipelines, and federated dataspace semantics.

```powershell
python scripts\search_kb.py "R2RML RML Morph-KGC International Data Spaces Gaia-X Dataspace Protocol" --concept mapping_language --limit 10
```

### Policy, privacy, and industrial semantics

Retrieve Phase 16 sources about federated query, privacy/access-control vocabularies, decentralized linked data, and industrial information models.

```powershell
python scripts\search_kb.py "SPARQL federated query DPV XACML Solid OPC UA SAREF eClass IEC CDD" --concept industrial_semantic_standards --limit 10
```

### LLM + KG + RAG

Retrieve evidence about LLM/KG integration, RAG, GraphRAG, ontology-grounded RAG, and KGQA.

```powershell
python scripts\search_kb.py "LLM knowledge graph RAG GraphRAG ontology grounded retrieval" --concept llm_kg --limit 10
```

### 2026 frontier evidence

Retrieve recent 2026 sources and preprints; use as frontier direction, not settled consensus.

```powershell
python scripts\search_kb.py "phase11" --limit 10
```

### Governance and sociotechnical critique

Retrieve governance, public-sector, privacy, accountability, and sociotechnical critique evidence.

```powershell
python scripts\search_kb.py "classification infrastructure public sector governance NHS FDP privacy accountability" --concept ai_governance --limit 10
```

### Commercial material boundary

Retrieve commercial implementation comparators; do not use these as academic proof.

```powershell
python scripts\search_kb.py "semantic layer data product enterprise AI context engine" --limit 8
```

## Evidence Boundaries

- Cite core/candidate sources first.
- Use commercial materials only for implementation, adoption, and market context.
- Use Palantir official docs for architecture claims, not independent validation.
- Mark Phase 11 arXiv items as preprint/frontier evidence.