Proposed European Standard · RFC · v1.3
DWS Ontology Standard
A decision-centric ontology for EU-sovereign enterprise AI — unifying Data, Logic, Action, and Security in one model.
Updated 29 April 2026 · Lifetime Oy · Espoo, Finland
Why an ontology, and why now
Off-the-shelf LLMs fail in regulated industries because they have no model of the enterprise — no semantic objects, no committed actions, no policies that propagate at runtime. The decision-centric ontology pattern that solves this has been proven by US incumbents over the last twenty years, but always on US-hyperscaler infrastructure. Europe needs the same architecture, sovereign by design, with EU regulations baked in as first-class entities — not encoded by every customer from scratch.
The DWS Ontology Standard is our attempt at that. We are publishing it as an open RFC because terminology, schemas, and policy envelopes are stronger as shared standards than as proprietary moats. The value lives in the runtime and the regulatory content, not the schema shape.
What v1.3 adds — etymology and philosophical background
Before discussing schemas, it helps to ground the term itself. The word ontology comes from two Greek roots: ὄν (ón) — "that which is", the present participle of einai ("to be") — and λόγος (lógos) — "doctrine, account, rational explanation". Literally, ontology is the doctrine of being: the study of what exists and in what manner.
Ontology is one of the central branches of metaphysics. Already in classical Greece, philosophers such as Aristotle examined what it means for something to be, what kinds of entities exist (objects, properties, relations, events), and what the basic structure of being is. The term ontology itself was consolidated only in the 17th–18th centuries, particularly within scholastic and early-modern philosophy, when metaphysics was systematised into distinct sub-fields.
Today the term is applied across philosophy (the study of the fundamental nature of existence), computer science and artificial intelligence (a formal model of concepts and the relations between them — classes, properties, relations), and the philosophy of science (the question of which entities a given theory commits to). The original meaning — structuring what is — is preserved while the application surface shifts. The DWS Ontology Standard builds on this foundation: EU regulations and industrial concepts are structured into a machine-readable form that humans and agents query through the same interface.
The four pillars
1. Data — the nouns
Semantic objects, properties, and links representing the enterprise in its own language. Spatial, temporal, causal, and entity ontologies; ERPs, MES, IoT, and unstructured sources unified — not flattened into golden tables.
2. Logic — the reasoning
SQL functions, ML models, optimizers, simulators, and LLM tools exposed through one logic-binding registry. Agents and humans see the same surface; new capability = new registry entry, not new integration.
3. Action — the verbs
Decisions are staged, reviewed, committed, and written back to ERP, WMS, edge devices, and email. Every action carries a rollback plan and a policy envelope. Agents propose; humans (or trusted automations) commit.
4. Security — the envelope
Role + purpose + markings evaluated at runtime on every data, logic, action, and tool invocation. Covers EU AI Act Articles 12 and 14 (logging + human oversight) automatically, not by review.
What makes DWS different
- EU regulations as first-class entities. Fit for 55, ETS, CBAM, CSRD, EU AI Act, NIS2, and GDPR are pre-modeled. Customers inherit compliance scaffolding instead of reinventing it.
- Sovereign by design. Data residency in Finland and Germany. No CLOUD Act exposure. Gaia-X tier-3 path. Run by EU-domiciled Lifetime Oy.
- CCP — concept_context_policy. Per-customer, per-industry, per-locale terminology preference and forbidden-term enforcement. KELA forbids "Physical AI"; Finnish construction prefers "embodied carbon ledger". The ontology adapts; the data does not need to be rewritten.
- Active pruning + collision protocol. Contradictory artifacts trigger forced reconciliation with human gates. Knowledge ages out instead of accumulating into a graveyard.
- Built for agents. Every product exposes CLI + MCP + Skill. The ontology is queryable by humans and agents through the same interface.
v1.2 roadmap — closing the operational gap
Knowledge-centric ontologies (v1.0–1.1) capture learning. To run operations, we must close
seven gaps. Each is dated and owned in
ONTOLOGY.md §13.
| Gap | Target |
|---|---|
| Actions / write-back layer | Q3 2026 |
| Per-decision lineage | Q3 2026 |
| Purpose- and marking-based policies | Q3 2026 (before AI Act enforcement) |
| Scenarios / Global Branching | Q4 2026 |
| Logic Binding registry | Q4 2026 |
| Multi-tier agent memory | Q1 2027 |
| Embedded edge ontology (Lifetime Fleet) | Q2 2027 |
Open European standard, RFC
We propose the DWS Ontology Standard as a candidate European standard, in the spirit of Gaia-X: schemas open, runtime competitive. The full specification — pillar definitions, schema templates for Action, Scenario, DecisionRecord, LogicAsset, PolicyEnvelope, and CCP — is maintained in ONTOLOGY.md and revised in the open.
Comments, critique, and adoption are welcome. Reach Risto at [email protected]. We intend to submit the spec to CEN-CENELEC JTC 21 (AI) and EFRAG (regulatory entity layer) once two pilot customers have adopted it.