Transparent and epistemically sound synthetic research
A synthetic panel is useful only when its origin, boundaries, and transformations remain visible. This guide follows the evidence into the panel, the questions asked of it, and the checks that help a researcher decide what an output can responsibly support.
Scope
A synthetic output is a model-generated response from a defined panel. This page explains what enters the process, what the outputs mean, and where interpretation and validation apply.
Inputs
Target descriptions define the population. Interviews and surveys provide observed material. Briefs and existing research can frame the question. Exploratory panels can start before fieldwork; evidence-grounded panels use collected evidence.
Individual reconstruction and collective alignment
Source statements and responses are interpreted into individual cognitive representations. A Human Digital Twin is a participant that simulates mental processes in the context of an observed person. Its personality is built through a complex process based on neuro-symbolic technologies (patent pending) that produce a Latent Cognitive Model for each agent. Each Human Digital Twin can use a distinct LLM, configurable for local deployment with inherent bias removed.
Panel and deliberation
Synthetic respondents form a defined population. They answer structured questions and deliberate through propositions, reasons, grouping, consensus, and disagreement. Research conducted with HDTwin synthetic panels is free from preference falsification: our Human Digital Twins are designed not to lie and to minimize hallucinations, providing a sufficient degree of epistemic resistance inherent in simulating character and psychological traits (Latent Cognitive Model).
What the panel returns
Outputs include individual responses, motivations, opinion groups, collective agreement, dissent, and traceability. They remain distinct from observations collected from real people. Results are available both as printable reports and machine-readable data.
Hermeneutic engine, preference falsification, and epistemic resistance
HDTwin.ai uses a hermeneutic engine to let its Human Digital Twins explore and discuss the same position from different perspectives, testing assumptions, interpretations, and frames of meaning. In this process, Digital Twins offer a specific advantage over human interlocutors: they are not subject to preference falsification because they do not need to please others, conform socially, or state opinions different from those represented in their model; at the same time, they do not experience fatigue, impatience, or cognitive saturation. This makes it possible to work more systematically on epistemic resistance, meaning the tendency to resist evidence, interpretations, or viewpoints that challenge established beliefs, while keeping the discussion open even when it becomes complex, repetitive, or cognitively uncomfortable.
Provenance
Keep the target, source material, question, panel configuration, and transformation path available for inspection. Every character element is explained and traced in the Latent Cognitive Model.
Calibration
Compare panel behavior with reference evidence and a human control group where available. Adjust the setup when the panel deviates or the source changes. Periodic maintenance of the synthetic panel, through new surveys, control groups, and bias sterilization for new models, ensures the best alignment with your real target.
Uncertainty
Confidence decreases as a question moves beyond the source material. A calibrated synthetic panel, for questions within its specific calibration domain, can provide answers aligned by up to 98% with those provided by a human control sample.
Privacy and governance
HDTwin is designed and hosted in Europe. Customer inputs are used to build their panel, remain their exclusive property, and are not stored on HDTwin infrastructure beyond strictly necessary processing. Human Digital Twins lose their connection to the potential human respondent, who must in any case be anonymous. They cannot be used to profile an existing human subject. They represent the cognitive processes of abstract “personas,” whose collective behavior replicates that of a community (the panel).
Validation
Validation means comparing behavior with real-population evidence. It does not turn synthetic output into direct observation or guarantee future behavior. Alignment with the human panel decreases over time because events can change perceptions of reality, reactions, or attitudes toward the world. For this reason, periodic realignment of the panel to a defined scope of inquiry is recommended. Panels are not oracles but predictive algorithms.
Evidence and testimonials
Read the scientific documentation validating our method and the press releases published by our clients.
Epistemic perspectives
Preserve minority reasons and disagreement instead of reducing every question to one average. Inspecting origins and transformations helps surface model and data imbalance.
Conceptual glossary
HDTwin methodology refers to the following definitions.
| Term | Usage |
|---|---|
| Platform | The HDTwin system that constructs and interrogates panels. The platform consists of methods, LLM and cognitive models, an application for interrogating panels (council), and a framework for generating customized panels (studio). |
| Human Digital Twin | A synthetic participant representing relevant characteristics, attitudes, and information. Equivalent terms include synthetic persona, synthetic respondent, and synthetic customer. HDTwin Human Digital Twins are developed with neuro-symbolic technology that combines the inferential capabilities of logic networks with the neural capabilities of LLMs (Large Language Models). |
| Synthetic panel | The defined population formed by Human Digital Twins and designed around a specific research boundary. |
| Evidence | Observed material, such as interviews and surveys, used to inform or calibrate a panel. |
| Interrogation | Structured questions or stimuli directed to a panel. The platform supports closed questions, open questions, or a combination of both, and can conduct quantitative and qualitative research simultaneously. |
| Deliberation | Iterative exchange in which panel members propose, comment on, evaluate, and group positions. In deliberation, Human Digital Twins put every statement produced to a vote, revealing polarization, consensus, dissent, and opinion groups. |
| Output | Responses, reasons, groups, consensus, disagreement, and traceability generated by interrogation. |
| Calibration | Adjusting a panel against observed reference material. |
| Provenance | The visible relationship between input, panel, question, transformation, and output. |
| Fieldwork | Research with real people. |
| Hermeneutic engine | A system designed to interpret a topic from different perspectives by comparing meanings, assumptions, and interpretive frames. |
| Epistemic resistance | The tendency to resist evidence, information, or interpretations that challenge established beliefs. |
| Preference falsification | Behavior in which a person publicly expresses a preference or opinion different from the one they actually hold. |
A 30-minute online conversation between your methodologist and our experts; we will provide a focused technical presentation.