Metaskills
RESEARCH · DIGITAL TWIN

How the System Really Works Digital Twin

A series of articles on the methodology for measuring human skills through interviews: the competency model underlying it, what is considered behavioral evidence, how we validate the assessment against human-administered evaluations, privacy and oversight decisions built into the design, and how the whole system integrates with the actual system.

Series

01 · Methodology

Competency Model

18 competencies across four areas and 108 behavioral scenarios. Why was a functional taxonomy chosen instead of a personality model, what does the description of competencies include, and why was it necessary to separate the competencies “Empathy” and “Active Listening”?

Read

02 · Methodology

How Does Behavior Become Evidence?

What the algorithm considers evidence of competence, and what it does not. The quality of the evidence, the behaviors that determine the outcome, why the confidence level is stated explicitly, and why we never calculate an average.

Read

03 · Verification in progress

Comparison with Expert Ratings

The study design, published prior to the announcement of the results: the same scenario featuring an AI-based avatar and a real actor, evaluated by independent reviewers. This is what we intend to test, and it has not yet taken place.

Read

04 · Project Framework

Privacy, Bias, and Human Error

What an evaluation may take into account, the four vectors of bias that guide our actions, the requirement that every evaluation include a quote from the transcript, and the stages in which a person participates in the process.

Read

05 · Technical Note

Technical Architecture and Integration

For those who need to approve this: the engine is separated from the interface and storage, which crosses the boundary between them; immutable profile versioning; and what the system intentionally never logs.

Read

The model in numbers

18
Competencies

Grouped into four behavioural domains, each with its own definition and boundaries.

Metaskills competency model, 2026 (internal methodology note).

108
Behavioural scenarios

Each one described across four levels of mastery.

Metaskills competency model, 2026 (internal methodology note).

6
Conversation functions

Every scenario is tagged with the goal of the conversation, not only its topic.

Metaskills competency model, 2026 (internal methodology note).

4
Confidence levels

The profile states how much evidence stands behind each competency, including when there is not enough.

Metaskills assessment methodology, 2026 (internal methodology note).

What we publish and what stays internal

We publish the method and the numbers that describe it: how competencies are structured, what counts as evidence, how confidence is decided, how the validation is designed. Anyone evaluating this product should be able to check our reasoning rather than take our word for it. What stays internal is the material that stops working the moment it is public: the full matrix of 18 competency descriptions, the 108 behavioural scenarios, the prompt structures, and the thresholds behind the confidence levels. A candidate who has read the scenario and the marking scheme is no longer being assessed on behaviour - they are being assessed on preparation, which is exactly the failure mode this model was built to avoid. We would rather draw that line out loud than leave the gap for you to notice.

At what stage we are

The competency model has been developed, and the assessment logic is in place: 18 competencies, 108 scenarios, and an evidence-based assessment with a confidence level specified for each competency. The validation process, involving independent evaluators, is underway. The project has been developed, a group of participants has been recruited, and the assessment tools are ready; all that remains is to conduct the tests. Until the test results are published, a fair description of the program Digital Twin is that it is a well-founded project with clear levels of confidence-not a validated measurement tool-and we will not refer to it as such until evidence supports that claim.

Related evidence

Study

Usability of avatar-based VR training

Forty-seven participants rated the training experience: usability, instruction clarity, avatar quality and workload.

Usability of avatar-based VR training
Pilot

Healthcare pilots in Spain

What it takes to deploy communication simulations in real healthcare - browser versus VR, clinical realism, localisation.

Healthcare pilots in Spain
Research

Research & Evidence

Everything we have published, with sources and limitations stated.

Research & Evidence

Want to see what the profile looks like in practice?

The product page shows what a learner and a team leader actually get out of this - reports, development plans and the conversation behind them.