Glossary

The glossary fixes a shared vocabulary for the publication. Definitions are deliberately narrower than marketing usage: each term must be useful when specifying an experiment, reviewing evidence, or assigning authority.

AI Scientist

A reasoning and experimentation layer that remains outside the control loop.

Related: AI Scientist

Challenger Model

A candidate compared with a champion under the same experimental conditions.

Related: Algorithm Arena

Champion Model

The currently selected candidate under stated evidence and constraints.

Related: Algorithm Arena

Digital Thread

The connected information trace that follows an asset across its lifecycle.

Related: Architecture

Digital Triplet

A research direction combining a physical machine, a digital twin, and an AI Scientist.

Related: Architecture, AI Scientist

Digital Twin

A computational representation of a physical asset and its relevant behavior.

Related: Architecture, Twin Capsule

Experiment Fabric

The system that treats experiments, their configurations, and their evidence as first-class objects.

Related: Experiment Fabric

Fault Injection

A controlled modification of a twin or input to study detection behavior.

Related: Fault Lab

Feature Factory

The transformation of raw industrial signals into engineered features.

Related: Feature Factory

Fleet Intelligence

Cross-asset learning and validation that preserves engineering context.

Related: Fleet Intelligence

Gray-box Model

A hybrid model that combines physical structure with learned components.

Related: Algorithm Arena

Model Provenance

The reproducibility record for an asset, data, features, model, code, and configuration.

Related: Experiment Fabric

Physics-Informed ML

Machine learning constrained or informed by engineering relationships.

Related: Feature Factory, Algorithm Arena

Residual Intelligence

Reasoning from the difference between physical measurement and digital-twin prediction.

Related: Feature Factory

Synthetic Fault

A fault scenario generated by a model, rather than observed in an operating plant.

Related: Fault Lab

Twin Capsule

A bounded record of an asset, signals, features, failure modes, and provenance.

Related: Twin Capsule

Twin Lab

An isolated environment for investigation on a digital representation rather than a physical machine.

Related: Architecture, Experiment Fabric

Twin Registry

A catalogue that identifies and locates Twin Capsules.

Related: Twin Capsule

Uncertainty Quantification

Characterizing uncertainty that affects an inference or recommendation.

Related: Experiment Fabric, Research

Validation

Assessing whether a result is fit for its intended engineering decision.

Related: Experiment Fabric

Verification

Checking that an implementation conforms to its specified design.

Related: Experiment Fabric

The distinctions matter. Verification asks whether an implementation conforms to its design; validation asks whether the result is fit for the intended engineering decision. A Champion Model is selected only under stated evidence and constraints, while a Challenger Model remains comparable under the same conditions. Neither title grants operational authority.