Reference

On-prem compute stack that ingests, validates, and contextualizes raw signals before publishing to the Unified Namespace.

Domain-owned schema that standardizes vocabulary, units, and business rules; enforced via JSON data contracts.

Real-time publish/subscribe backbone (e.g. MQTT) that carries CDM-tagged events in a hierarchical topic structure.

A defined agreement governing data structure, units, and timeliness for each data product and its transmission between systems.

A UMDA service that directs each task to the most appropriate language model (lightweight, domain, or general-purpose) by complexity, latency, and cost.

Governed local and central storage of harmonized CDM tables, time-series joins, and enterprise KPIs with full lineage.

Historical store of AI inferences, human inputs, and outcome metrics used for model retraining and audits.

An autonomous service that consumes CDM/UNS data, detects events or anomalies, and writes decisions back to the Feedback Data Layer.

Domain-owned dataset published with a defined schema, quality SLA, lineage, and access policy for reliable reuse across all layers.

Permission model enforced at UNS topic and UDL table levels to support zero-trust security.

A coordinated set of AI agents that can autonomously pursue manufacturing goals, collaborate, and learn from feedback.

The five-step scale of agent authority: 0 informs, 1 recommends, 2 acts with approval, 3 acts with oversight, 4 autonomous in domain. Defined in the Earning Autonomy whitepaper; risk sets each agent's ceiling.

A design where a person evaluates or approves an agent's actions. Not a temporary stage to be removed: for some decisions (batch release, safety, CAPA closure) it is the permanent, correct design. See the AI Routing & Agents guide.

Running an agent against real operations while acting on none of its output. Its proposals are logged next to what humans actually did, building the evidence base before the agent gets any authority.

Promoting an agent one autonomy level at a time against pre-written acceptance criteria, the same discipline as equipment qualification. Demotion triggers are named in the same document. The evidence record lives in the Feedback Data Layer.

How far a wrong action reaches before someone notices: one machine, one line, one site, or the whole network. A key risk factor in deciding how much authority an agent gets.

A virtual model that mirrors the state and behavior of a physical asset or process.

An end-to-end connective trace of product and process data that enables lifecycle visibility.

Placing compute at or near production equipment to enable low-latency analytics and control.

Central index documenting data sources, lineage, ownership, and quality across the architecture.

Time-series database optimized for high-frequency industrial process data.

Keeps enterprise reference data (materials, suppliers, equipment) consistent across domains.

Streams only the data that has changed in a source system to keep downstream stores in sync.

Integrates data from various sources, cleans and organizes it, and loads it into a central repository.

Composite KPI for manufacturing productivity: Availability × Performance × Quality.

A numeric measure of performance (e.g. yield, cycle time) surfaced in dashboards and reports.

Connected sensors and controllers continuously streaming shop-floor signals.

Orchestrates production operations and gathers in-process data.

Supervisory Control and Data Acquisition. Aggregates PLC signals for shop-floor visualization and control.

System for finance, planning, and supply-chain coordination.

Computerized Maintenance Management System. Schedules, tracks, and records maintenance activities and asset health.

Laboratory Information Management System for sample tracking and quality test results, integrated through CDMs.

Manages non-conformance tracking and regulatory evidence.

Manages a product's design, engineering changes, and compliance through its lifecycle.

Manages inventory, storage locations, pick-pack, and logistics.

Shop-floor consoles that publish operator input and display live events.

Processes and methods that help humans understand and trust the outputs of AI.

A security strategy that treats every user, device, and network as untrusted until continuously verified.

A normalization standard that reduces redundancy so every non-key column depends only on the primary key.

A connected, graph-based form of the Unified Data Layer that links assets, processes, materials, events, and their relationships, so people and AI can traverse context, not just query tables.

A formal model of the concepts, relationships, and rules in a domain. In UMDA it gives Common Data Models shared, machine-readable meaning and provides the schema for the knowledge graph.