Clinical

A cohort shaped by your question.

Choose whether to reproduce the reference mix or request a different balance. Then inspect the counts, the distribution errors and the effects of safeguards separately.

Build an artificial cohort

Optional stage and histotype targets
Reproducibility and release controls

Measure similarity, or also filter candidates?

Evaluate risk only: generate the requested cohort and report duplicates and similarity to the artificial training reference. No distance-based selection filter is applied. These risk results describe the returned cohort; they do not automatically remove records.

Apply the training-distance selection filter: generate a larger candidate pool, exclude candidates that match or are too close to training records, then select records to meet the requested condition quotas. Distance means similarity across modelled fields, not physical distance. The held-out evaluation set is not used to select candidates.

Filtering can take longer and change the returned records and their statistical fidelity. If too few eligible candidates remain to fill the quotas, this demo stops with a shortage error rather than silently relaxing the targets. Neither mode guarantees anonymity or differential privacy. Both modes retain the existing consistency checks and report their effects.

Your cohort and its measurements will appear here.

Stored benchmark from an earlier run

This earlier artificial experiment is independent of the cohort you generate above. Its method and complete results stay fixed when you change interactive settings.