Benchmarked against
real human behavior.

Native validates digital twins by asking twins and real people the same questions, then comparing the answers. The human side comes from benchmark data such as prior survey results, known segment distributions, or human responses held back from twin construction. Each validation report shows the human result, the twin result, and the gap between them, so your team inspects performance directly. Native and the client agree on the benchmark, the metrics, and the success threshold before the study runs.

Human parity means a digital twin result falls within an agreed range of the human benchmark for the same question and audience. Parity does not mean a twin is identical to a person. Native measures parity with metrics such as the point gap on top-2-box scores and agreement on rank order, against a threshold agreed with the client before the study. Native’s standard parity target is within ±5 percentage points of the human result, and clients set alternative thresholds where a decision requires them. Some clients also calibrate twins toward business outcomes, such as campaign performance, alongside survey parity.

A holdout benchmark is human data kept separate from the information used to build or calibrate a twin. The twin never sees this data. The comparison therefore shows how the twin performs on new questions, rather than how well the twin repeats its inputs. Native uses holdout sets to check calibration before twins support decisions.

No. Performance varies by audience, question type, category, and available evidence, so no single headline number describes every use. Native reports validation for the audience, questions, and benchmark tested in each study. Native also sorts use cases by readiness. Some perform well on Native’s foundation data alone. Others remain directional until client-selected data or a human benchmark moves the results toward decision-grade. A result is decision-grade when the audience is clearly defined, inputs are strong, the use case is narrow, benchmarks exist, outputs are expert-reviewed, and limitations are documented.

Every simulation shows three diagnostic scores: response rate, relevance, and faithfulness. Response rate shows the share of selected twins with enough context to answer. Relevance shows whether the answers address the question asked. Faithfulness shows how closely the answers stay grounded in each twin’s source data. Where a human benchmark exists, a validation scorecard adds the benchmark comparison.

Native validates twins when the evidence behind them changes: when new source data enters a twin, when a twin is refreshed, or when a twin moves into a new use case. The revalidation cadence and thresholds are set with each client as part of the engagement’s governance terms.

Yes. Native walks prospective and current clients through its validation approach, benchmark definitions, and study design. Detailed discussion of validation results takes place under NDA.