A digital replica is not automatically a digital twin. Classify any system across coupling, identity, scale, engine, and autonomy—then test the claim.
The twin is the coupled system, not the digital model alone. In medicine, the return path may be a clinician-authorized intervention rather than an automatic actuator. This atlas uses a strict data-flow taxonomy and makes broader usage explicit.
An entity-specific digital representation, kept synchronized by observations and used in a governed feedback loop to understand, predict or influence its real-world counterpart.
Evidence baseThe coupling ladder
Select each stage. The critical threshold is the move from an automatically updated shadow to a governed, bidirectional twin.
The physical and digital counterparts exchange data automatically in both directions. Predictions or decisions can alter the physical system, directly or through a governed human-in-the-loop action.
High fidelity, 3D visualization, AI and real-time dashboards do not by themselves create a twin. Ask whether a specific counterpart is synchronized and whether a governed return path exists.
Six independent lenses
“Twin” is only the coupling class. A complete description adds five other dimensions so different systems can be compared without collapsing unlike claims.
This is the fastest way to test whether a system is truly a twin. The direction and automation of data exchange matter more than visual realism.
Offline scenario exploration; no operational counterpart.
Humans transfer measurements, parameters or decisions.
The physical system updates the digital representation.
Observation and governed influence form a closed loop.
[coupling] + [identity] + [scale] + [engine] + [purpose]
“A digital shadow of a specific patient’s heart, using a hybrid model for risk prediction.”
Claim checker
Six diagnostic signals. The result separates the formal coupling class from personalization and advanced maturity.
A dashboard that only stores or displays measurements is not yet a model.
An asset, patient, organ, process or other identifiable entity.
Recurring sensor, clinical, imaging, omics or operational data.
Direct control or a tracked, human-authorized intervention.
Not only a population average or a generic reference model.
Learning is optional for a twin, but signals higher maturity.
Exact comparison
Learning, AI and 3D are optional capabilities. Automated synchronization and a governed feedback path are the defining structural features.
| Defining property | Simulation | Digital model | Digital shadow | Digital twin |
|---|---|---|---|---|
| Physical counterpart | Optional | Yes | Yes | Yes |
| Physical → digital | None / offline | Manual | Automatic | Automatic |
| Digital → physical | None | Manual | Manual | Automatic or governed |
| Entity-specific state | Optional | Optional | Usually | Required |
| Operational feedback loop | No | No | No | Yes |
| Learning / adaptation | Optional | Optional | Optional | Optional |
* “Automatic or governed” includes a traceable human-in-the-loop return path where direct actuation would be unsafe or inappropriate, as in clinical care.
Living systems
In biology, the physical counterpart changes, adapts and is only partially observable. A defensible twin therefore needs identity, longitudinal updating, uncertainty and a declared context of use.
Protein · pathway · drug–target system
Structure, kinetics, binding, perturbation assays
Mechanism, affinity, dose or pathway response
A specific cell state or cell population
Single-cell omics, imaging, lineage, perturbations
State transition, phenotype or intervention response
Organoid, tissue, heart, liver, gut or brain
Imaging, biomarkers, flow, electrophysiology, function
Local dynamics, toxicity, efficacy or functional change
One person across interacting physiological systems
Clinical history, wearables, imaging, omics, exposures
Trajectory, risk, treatment choice or monitoring plan
Cohort, care pathway or public-health system
RWD, registries, environment, utilization and policy
Scenario impact, resource allocation or trial design
Combine longitudinal measurements with provenance, timing and missingness. A one-time baseline creates personalization, not a live twin.
Use mechanistic, data-driven, agent-based or hybrid engines. Quantify identifiability and uncertainty—not only fit.
State the user, decision, population, operating range and harm of error before choosing validation evidence.
Record the intervention and outcome, monitor drift and update only within approved safety and accountability boundaries.
They may predict cellular behaviour brilliantly, but usually represent a cell type or distribution rather than one continuously observed living counterpart. Call them computational or foundation models until identity, updating and feedback are demonstrated.
Worked examples
The same application can be a model, shadow or twin depending on how it operates today.
Sensor data updates equipment state; the model predicts degradation and a governed maintenance or control action changes operation.
Live telemetry alone creates a shadow. It becomes a twin when simulations drive verified changes to HVAC, lighting or occupancy policy.
A calibrated cardiac model is not automatically a twin. Repeat observations plus a governed treatment-feedback loop move it toward one.
A virtual cell becomes a twin only when it tracks a particular biological counterpart over time and informs a closed experimental loop.
A twin can still be wrong
Coupling tells you what the system is. Credibility tells you whether it is fit for a specific decision.
More detail is not always better; enough fidelity is context-dependent.
Check implementation, real-world performance, uncertainty and applicability.
“Real time” means fast enough for the use case, with latency made explicit.
Define ownership, overrides, monitoring, failure modes and accountability.
Common confusions
Digital-twin language varies across fields. These distinctions make claims easier to compare.
Evidence base
The framework synthesizes cross-industry terminology with biomedical requirements. It is a practical taxonomy, not a new formal standard.
Interpretive note. Some standards use “digital twin” as a broader umbrella. This atlas keeps coupling class and maturity separate so a digital shadow is not mistaken for an operational closed-loop twin.