Model card
Every number on this page is read from reports/metrics.json. Evaluated on held-out synthetic patients generated by a different, richer simulator than the twin.
Cohort realism
Is the synthetic cohort believable?
Intended use
- A proof-of-concept clinician console for Type 2 Diabetes, built for the Happiest Health Digital Twin Challenge 2026.
- Decision support research only: forecasts, alerts and What-If simulations on synthetic patients.
- Shows how a doctor could interact with a virtual patient before advising a real one.
Limitations
- All patients are synthetic. Results have not been validated on real people and must not be used for treatment.
- The truth simulator is richer than the twin, but both are models; real physiology, sensors and behaviour are messier.
- Subtype inference is weak (see the confusion matrix) and is shown as a hint, never a diagnosis.
- Food values are approximate (IFCT 2017 / International GI Tables where available, estimates otherwise).