How validation works
Validation runs on two layers: internally, every decision path of the deterministic model is tested exhaustively — full coverage, not sampled; externally, independent experts verify the result on a statistically powered, epidemiologically representative case set.
01 · Independent experts
Recognized specialists for the indication provide their case decisions — independently, before they see what the system says.
02 · Representative case sets
Cases are stratified to the indication’s real-world epidemiology — stages, subtypes, molecular profiles in their true proportions — with sample sizes derived statistically from confidence targets, not chosen for convenience.
03 · Blind, neutral execution
Case entry and comparison are performed by an independent institute — a methodology co-developed with the Fraunhofer Institute IIS — by staff with no clinical stake and no knowledge of the expected result.
04 · Measured agreement
Concordance between expert judgment and system output is quantified with established statistics (Cohen’s and Fleiss’ kappa), against hypotheses fixed before the study starts.
05 · Reproducibility, proven
Every case is entered multiple times; identical input must yield the identical answer, every time. Determinism is a test result, not a marketing claim.
One method, every indication, every localization — and every guideline update triggers a scoped re-verification round.
We claim exactly what these studies establish: correct representation of expert knowledge, and reproducible output — no more, and no less.