Model · Shadow · Twin
A digital model, a digital shadow, and a digital twin are three different things, and the difference has nothing to do with how sophisticated the software is. It is decided by which direction the data flows, and by whether the thing is allowed to act.
The distinction
The standard classification asks two questions, and only two. Does data reach the digital copy automatically? And does the digital copy change the physical thing automatically? Three of the four answers have names.
| Kind | Physical → Digital | Digital → Physical |
|---|---|---|
| Digital model | Manual | Manual |
| Digital shadow | Automatic | Manual |
| Digital twin | Automatic | Automatic |
Classification after Kritzinger, Karner, Traar, Henjes & Sihn (2018), Digital Twin in manufacturing: A categorical literature review and classification, IFAC-PapersOnLine 51(11), 1016–1022.
Side by side
A replica that does not know the real thing changed.
What it does
Represents the facility, instrument, or process. You can measure it, simulate against it, and share it. Updating it is a person’s job.
What it cannot
Notice anything. Move the real bench and the model is silently wrong until someone re-enters the data.
Who it is for
Teams planning a build-out, testing a protocol before the bench, or documenting what they have.
Levels L1–L2
More on the digital modelLive data in. No control out.
What it does
Ingests sensor, instrument, and equipment telemetry on its own. Stays current, flags drift, and forecasts from real data rather than assumptions.
What it cannot
Act. Every decision it informs still passes through a human hand — which, in a regulated lab, is frequently the point.
Who it is for
Facilities that need live truth and continuous monitoring, and labs where software is not permitted to actuate equipment.
Levels L3–L4, once connected
More on the digital shadowThe loop closes. The model changes the facility.
What it does
Everything a shadow does, and then acts on it: reschedules the robot, adjusts the setpoint, triggers the reorder, without waiting for a person.
What it cannot
Be adopted casually. Letting software actuate real equipment is a governance decision before it is an engineering one.
Who it is for
Automated and high-throughput facilities where the cost of a human in every loop exceeds the cost of the loop.
Level L5
See twin pricingThe part everyone gets wrong
A brilliant offline simulator is still a model. A live-connected twin with no forecasting at all is still a twin. Intelligence and connectivity are two different axes, and the market conflates them constantly, which is how a prediction model in a browser ends up being sold as a digital twin.
Take the same predictive web application and change nothing but its plumbing.
A researcher types parameters in and reads a forecast out.
The app never learns that the real facility changed. Someone has to tell it.
The same app ingests live instrument and sensor data and keeps its predictions current on its own.
Now it tracks reality. It still cannot do anything about what it predicts.
The same app writes its decisions back — reschedules the robot, adjusts a setpoint, places the reorder.
The loop closes. The forecast becomes an action without a human in it.
Same model. Same mathematics. Three different answers. When a vendor tells you their product is a digital twin, the useful question is not how clever it is. It is where its data comes from, and what it is allowed to change.
Our own ladder
It would be easy to call all five of our levels digital twins. Here is what they actually are.
| Level | What we build | Honestly, it is a… |
|---|---|---|
| L1Model | Geometric replica | Digital model |
| L2Inform | Instrumented asset twin | Digital model |
| L3Predict | Simulation-ready twin | Digital model A shadow once you connect it to live instrument data. |
| L4Prescribe | Optimization twin | Digital shadow A shadow when fed live data; still a model if you feed it by hand. |
| L5Living | Living twin | Digital twin A true digital twin only where the feedback actuates. Scoped to alert a human instead? Then it is a digital shadow, and we will call it one. |
The one we will not fudge
Level 5 is a true digital twin only where the feedback actuates.
If your Level 5 is scoped to raise drift alerts for a human to act on, which in a regulated lab is frequently the only permissible design, then it is a digital shadow with an excellent dashboard, and we will call it one. That decision belongs in discovery, not in the invoice.
Five levels, published USD prices, and a benchmark against NIST’s $1.70–$2.40 per square foot figure for laboratory digital twins. Nobody else in this market publishes a price at all.
Questions
The direction of automated data flow. A digital model exchanges no data automatically — a person updates it. A digital shadow receives data automatically from the physical thing, but sends nothing back. A digital twin exchanges data automatically in both directions: a change in the digital object changes the physical one. This classification comes from Kritzinger et al. (2018), and it is the standard the field uses.
Not by itself. Prediction is a separate question from connectivity. If a researcher types parameters into the app and reads a forecast, it is a digital model with a predictive layer. If the app ingests live instrument data and keeps its predictions current on its own, it is a digital shadow. If its output actuates something — reschedules the robot, adjusts a setpoint, places the reorder — then it is a digital twin. The intelligence of the model does not decide this. The plumbing does.
No. A Matterport scan or a CAD file is a digital model: it describes what a space looks like at one moment. It does not know when the real room changes, and it cannot act on it. That is a useful thing to own, and it is where most twin projects start, but calling it a twin is the most common misrepresentation in this market.
Often not. In regulated or safety-critical laboratories, letting software actuate equipment may not be permitted at all, which makes a digital shadow the correct destination rather than a failed twin. The right question is not how high up the ladder you can afford to go, but which decision you need the model to support.
Levels 1 and 2 are digital models. Levels 3 and 4 become digital shadows once connected to live data, and remain models if you feed them by hand. Level 5 is a true digital twin only where its feedback actuates. If a Level 5 is scoped to raise drift alerts for a human to act on, it is a digital shadow, and we say so.
