Skip to main content

Arca Suite OM 2.1 Native: Ontologies Flow Natively from OpenMetadata 2.1 into the Knowledge Graph

Arca Suite now runs against OpenMetadata 2.1. An ontology published in the Ontology Studio — sealed, approved by an independent reviewer, activated — exports its release RDF natively into the ArcaQ knowledge graph: the graph now materializes the sealed release itself (209 triples measured on the live graph), not a reinterpretation of it. Around that pipe: a daily ontology quality watch, and a governance kernel that got stricter. Everything below was proven against the running platform.

Key Takeaways

  • OpenMetadata 2.1 native integration — the platform runs against OM 2.1. The publish flow enforces independent approval server-side: the person who seals a release cannot be the one who approves it. The connector then fetches the sealed release RDF through the native export API and writes it into the knowledge graph as a complete snapshot — 209 triples, counted on the live graph.
  • Version-aware integration — capability probes keep 2.0 deployments on the proven path while the 2.1 native-RDF path activates only when the server supports it. OM 2.1.0 is a release candidate (rc1): we run it, we track it with stable-watch, and we say so.
  • Ontology quality watch — six competency questions are re-evaluated daily against the live graph, seeded on the «nine tentacles» doctrine: the map is already drawn, engineer only the delta. Drift surfaces automatically on the dashboard, persisted in Postgres — it survives a pod replacement.
  • Kernel hardening — the governance kernel got stricter: a governed merge gate (plan + certificate + entailment preservation), a fail-closed SPARQL guard, a global No-LLM kill switch, and export-control invariants. The full gap list is internal; the summary is: all of it is proven live.

What changed

Until now, feeding an OpenMetadata ontology into the knowledge graph meant materializing it ourselves — reinterpreting the catalog's tables into triples. With OM 2.1, OpenMetadata serves the ontology release as RDF directly. Arca Suite now uses that native export:

  • Publish, seal, approve, activate. An ontology authored in the Ontology Studio is captured as a sealed release; approval is routed through a task assigned to an independent reviewer — the server refuses a self-approval — and only then can the release be activated.
  • Native RDF into the graph. The connector reads the active sealed release through the export endpoint and writes it to the knowledge graph as a complete snapshot. On the live run, the graph held exactly the 209 triples of the release — classes, properties, axioms, domains and ranges — verifiable by SPARQL.
  • Fail loud, not silent. Two real contract gaps between our client and the rc1 server were found live (a wrong export route, a content-negotiation mismatch), fixed, tested and re-proven. When the native path cannot run, the connector says why — it never feeds the graph a document it cannot vouch for.
  • Version-aware. A capability probe detects what the OM server supports: 2.0 deployments keep the proven materialization path; the 2.1 native path activates only when available.

The ontology watch: six questions, every day

The «nine tentacles» doctrine says an ontology is a living map: most of it is already drawn, so you engineer only the delta — and you need instruments that tell you where the map and the territory diverge. Six competency questions now run daily against the live graph:

  • Does the class hierarchy hold as designed? (convergence)
  • Are personal-data properties marked, and attached to the personal-data class? (description fields)
  • Is every axiom traced back to the published OM release — zero orphans? (delta)
  • Do all concepts and properties carry descriptions? (quality feedback loop)
  • Do classes carry language-tagged FR/EN labels?

Five currently answer clean. The sixth reports an expected gap: the rc1 export condenses display names into untagged literals, so language-tagged labels are missing — surfaced as a review candidate for human decision, not hidden and not auto-fixed. That is the watch doing its job: drift appears on the dashboard by itself.

Kernel hardening

This release also closes the audit-gap wave on the governance kernel. In compact form — the full gap list stays internal:

  • Governed merge gate — merging generated artifacts requires a plan, a certificate, and proof that existing entailments are preserved.
  • Fail-closed SPARQL guard — a query that cannot be validated does not run.
  • Global No-LLM kill switch — one switch removes LLM mediation from the platform, globally.
  • Export-control invariants — exports obey the platform's governance invariants by construction, not by convention.

Each item was proven live against the running platform before this release was named.

Version marker

  • arcaq-api — efc07299b line (kernel 2026-11-11).
  • arca-decision-room — 0.3.7 (unchanged).
  • arcax — 0.6.7 (unchanged).
  • OpenMetadata — 2.1.0-rc1, a release candidate; stable-watch pending.
  • Arca Suite — OM 2.1 Native (2026-11-11), this release. Previous release: Vault GO (2026-11-10).

What we do not claim

  • OM 2.1.0 is a release candidate: we run it in production against scenario data and track it with stable-watch — this is not a GA-support announcement, and not a certification of any kind.
  • The published ontology (arcaq-crm-demo) is scenario data, exercised on our own infrastructure — not a customer deployment.
  • «Proven live» refers to Arca Suite's own gates on our platform — it is not third-party certification, and we don't use that word.
  • The kernel-hardening list above is a summary; the complete gap list is internal engineering material.

Availability & what's next

Same release mechanic as before: end-to-end green, documentation, and the four sites updated in the same cycle — arcaq.com, getarcasuite.com, arcax.ai, talentys.eu — with the version marker moving each time.

Explore the platform: What is ArcaQ, Ontology, Ontology Governance, and Knowledge Graphs.