--- type: CKG Bundle title: Open Knowledge Format (OKF) × Compressed Knowledge Graphs description: What OKF is, how it works, and why a CKG is the OKF-conformant superset — documented as a compressed knowledge graph that is itself a valid OKF bundle. resource: https://cloud.google.com/blog/products/data-analytics/how-the-open-knowledge-format-can-improve-data-sharing tags: [okf, ckg, knowledge-layer, open-standard, llm-context, semantic-layer] timestamp: 2026-06-18T00:00:00Z ckg: id: okf-ckg version: 1.0.0 confidence: high source: "Google Cloud OKF announcement (McVeety/Hormati, 2026-06-12) + github.com/GoogleCloudPlatform/knowledge-catalog/okf" --- # OKF × CKG > **This file is itself a valid Open Knowledge Format bundle.** The frontmatter above is OKF; the `ckg:` block + the typed edges below are the CKG superset. Google's Knowledge Catalog (`kcmd push`) would ingest this unchanged. ## META - domain: Open Knowledge Format ↔ Compressed Knowledge Graphs - nodes: 22 · edges: 26 - sources: Google Cloud OKF blog (2026-06-12); GoogleCloudPlatform/knowledge-catalog repo (okf/ spec + GA4 sample bundle) - integrity: 0% hallucination by construction — every node declared + sourced; relationships explicit, not inferred - license: CC BY 4.0 · Graphify.md --- ## THE ARTICLE (compressed) On **June 12, 2026**, Google Cloud published the **Open Knowledge Format (OKF)** — "an open specification that formalizes the LLM-wiki pattern into a portable, interoperable format." In plain terms: Google standardized the *container* that knowledge-for-LLMs travels in — a directory of markdown files with YAML frontmatter. *No compression scheme, no runtime, no SDK.* It's the USB port, not the device. That is a tailwind, not a threat. OKF legitimizes the category at the highest level and stops **exactly** where the value begins. The same week, AI-strategy advisor Darlene Newman put the gap in one line: *"A spreadsheet in a context window is a document pretending to be infrastructure."* A document is portable; it is not a **semantic layer**. The semantic layer is the structure that lets two people, two systems, or two agents *reach the same conclusion for the same reason* — vocabulary, taxonomy, relationships, rules, provenance. A **Compressed Knowledge Graph (CKG)** is the markdown document that *is* infrastructure: **OKF-portable on the outside, semantic-layer-grade inside.** Every CKG is a valid OKF bundle; not every OKF bundle is a CKG. OKF gives you portability; the CKG layer adds what OKF deliberately omits — **compression** (~269 vs ~2,982 tokens/query, ~11× fewer), **deterministic traversal** (typed edges, not generation → 0% knowledge-graph hallucination by construction), **calibrated confidence + provenance**, and **L2 human attestation**. The move: don't compete with OKF — **conform to it, then go further.** Adopt the lingua franca; sell the structure that makes meaning reusable. --- ## NODES (ConceptID, ConceptLabel, Dependencies, TaxonomyID) ```csv ConceptID,ConceptLabel,Dependencies,TaxonomyID 1,LLM context problem,,FOUND 2,LLM-wiki pattern,1,FOUND 3,Open Knowledge Format (OKF),2,OKF 4,YAML frontmatter,3,OKF 5,Markdown body,3,OKF 6,type field (required),4,OKF 7,Common fields: title/description/resource/tags/timestamp,4,OKF 8,Arbitrary extra keys (extensible),4,OKF 9,Cross-links [text](/path.md),5,OKF 10,Bundle = directory hierarchy,3,OKF 11,Auto-generated index.md,10,OKF 12,Graph-shaped relationships,9|10,OKF 13,Knowledge Catalog ingestion (kcmd / Dataplex),3,OKF 14,Format not platform,3,OKF 15,Compressed Knowledge Graph (CKG),3,CKG 16,Typed nodes & edges (DAG),15,CKG 17,Compression (~11x fewer tokens),16,CKG 18,Deterministic traversal,16,CKG 19,0% hallucination by construction,18,CKG 20,Confidence + provenance,16,CKG 21,L2 human attestation,20,CKG 22,CKG = OKF-conformant superset,15|3,CKG ``` ## KEY EDGES (typed) ``` OKF --formalizes--> LLM-wiki pattern OKF --requires--> type field (exactly one required key) OKF --serializes_as--> markdown + YAML frontmatter OKF --omits--> compression / determinism / attestation Knowledge Catalog --ingests--> any OKF bundle (kcmd push) CKG --is_a--> OKF bundle (superset) CKG --adds--> compression, determinism, confidence, provenance, L2 Typed edges --enable--> deterministic traversal Deterministic traversal --yields--> 0% knowledge-graph hallucination (by construction) "ckg:" extension --rides_on--> OKF "arbitrary extra keys" (forward-compatible) ``` --- ## OKF EXAMPLES **1 — A plain OKF document (Google's spec):** ```markdown --- type: Concept title: GLP-1 Prior Authorization description: Payer gate before a GLP-1 is dispensed. tags: [glp1, payer, access] timestamp: 2026-06-18T00:00:00Z --- # Prior Authorization Most plans require step therapy and a documented BMI/comorbidity first. ``` **2 — The same file as a CKG (still valid OKF):** ```markdown --- type: CKG Concept # the one OKF-required field title: GLP-1 Prior Authorization tags: [glp1, payer, access] timestamp: 2026-06-18T00:00:00Z ckg: # extension — plain OKF readers ignore it id: 42 confidence: 0.94 source: ClinicalTrials.gov + 2026 plan policy --- # Prior Authorization — typed edges, not prose depends_on: Step Therapy(38) | BMI>=30 or comorbidity(31) enables: Dispense GLP-1(45) gated_by: Formulary Tier(40) ``` **3 — Bundle structure (a directory of files, graph-shaped via links):** ``` glp1-ckg/ index.md # auto-generated entry point concepts/prior-authorization.md concepts/step-therapy.md # [[linked]] from prior-authorization.md references/formulary-tier.md ``` --- ## OKF SPEC (compressed) - **Container:** a directory of `.md` files, each = one concept/resource. - **Frontmatter (YAML):** `type` is the **only required** field; common fields `title`, `description`, `resource`, `tags`, `timestamp`; **arbitrary extra keys allowed** (this is where `ckg:` lives — forward-compatible). - **Body:** markdown — prose, schemas, example queries; readable by humans and LLMs. - **Links:** standard markdown `[text](/path/to/concept.md)` → graph-shaped, richer than tree parent/child. - **Bundles:** directory hierarchies + auto-generated `index.md` for progressive navigation. - **Principle:** *"minimally opinionated, freely extensible"* — interoperability surface, not a content model. *"Format, not platform."* ## WHY CKG (the superset = what OKF leaves out) | OKF gives you | CKG adds on top | |---|---| | portable markdown, any tool | **compression** — ~11× fewer tokens/query | | human + agent readable | **typed edges, DAG** — deterministic traversal | | frontmatter for filtering | **confidence + provenance** per node | | interoperability surface | **L2 human attestation** (signed, versioned) | | a document | **infrastructure** — same answer, same reason | --- *Built by Graphify.md — own the knowledge layer, rent the model. Numbers: Macro-F1 0.471 vs RAG 0.123 (Track 1, 44 domains); ~11× fewer tokens; ~42× RDS; 0% hallucination by construction. Clone & re-run: github.com/Yarmoluk/ckg-benchmark · graphifymd.com*