Digital twins: what connects the model to the real system?

By the Vrulum team

Sources checked: 23 October 2026.

A convincing 3D model is not enough evidence that a system can predict how a particular machine or building will behave. When a project calls itself a digital twin, ask what it represents, what updates it and where its predictions have been checked.

This explainer concerns the evidence needed to assess a claimed twin. It does not assign an arrival date or readiness score to an entire technology category.

Definitions have a scope

The Digital Twin (official) definition, published on GOV.UK by the UK Defence Science and Technology Laboratory (Dstl) on 29 October 2025, ties a twin to a real-world entity, environment or process. It describes two-way information flow on a timescale appropriate to the decisions being made, and a validation envelope: the known conditions within which the twin can be trusted to mimic its real-world counterpart.

That definition is more specific than simply displaying a live dashboard. It also allows a twin to be connected, semi-connected or disconnected, provided its current state is recorded. Constant connection is therefore not a useful universal pass-or-fail rule. Ask whether a disconnected instance can still represent its counterpart adequately and what would be required before reconnecting it.

These are requirements of that published definition, not evidence that every vendor using the phrase has met them.

Follow one decision through the model

Consider a hypothetical building operator exploring a change to heating schedules. A drawing provides spatial information. A temperature dashboard provides observations. A model might explore the consequences of alternative schedules. To assess the claimed twin, follow how those elements relate to the actual building and the decision being made.

Ask which building and operating configuration are represented, which measurements enter the model, how old they are and which assumptions stand in for missing information. Then ask how predicted outcomes were compared with observed outcomes under relevant conditions.

This is an illustrative assessment scenario, not a reported demonstration of energy savings. No percentage improvement can be inferred from having the components in place.

Identify the limits of validation

The UK definition separates the twin from the physical system's safe operation: a twin can send back decisions or information, but the physical entity must maintain safe functionality even without a continuous connection to it. A demonstration of a useful forecast does not, on its own, establish authority to control safety-critical equipment.

For a practical evidence pack, request the intended use, checked operating conditions, error measures, data dates and treatment of unavailable or unreliable inputs. Ask what happens after a physical modification. A model checked for one configuration should not silently inherit approval for every later version.

Include trust and security questions

The US National Institute of Standards and Technology (NIST) announced its report IR 8356, Security and Trust Considerations for Digital Twin Technology on 14 February 2025. The announcement says the report covers twins of physical and non-physical entities, existing modelling and simulation, and cybersecurity and trust considerations. That places these issues within the technical assessment; it is not certification of a named product.

For a proposed deployment, ask who can change the model, replace incoming data or approve an output used in a real decision. Retain versions and decision records so an unexpected result can be investigated. These are editorial assessment questions, not findings that a particular system is insecure.

Read the evidence at the right level

A specification establishes a proposed framework. A supplier demonstration reports behaviour in its stated arrangement. A validation report can support a defined use under stated conditions. An operating deployment needs evidence about the actual configuration and ongoing updates.

None of those automatically establishes all the others. The useful claim is specific: this representation supports this decision, for this counterpart, inside these checked conditions. That is more informative than calling a model an exact replica of reality.

Sources and scope

Sources checked on 23 October 2026. Worked scenarios are illustrative, not product tests.