🧪 Medical AI Diagnosis Simulation

An AI diagnostic model rarely outputs a plain "sick" or "healthy" — it outputs a continuous risk score, and a human decision turns that score into an action by picking a cutoff. This simulator generates a cohort of 300 simulated patients whose scores come from two overlapping distributions (diseased vs healthy), renders them as a 3D point cloud along the score axis, and lets you drag the decision threshold through the cloud while a live confusion matrix, sensitivity/specificity, positive/negative predictive values, an ROC curve and AUC all update in real time. Separate sliders control how discriminative the underlying model is, how noisy its scores are, and how common the disease is in the tested population — the same three variables that separate a good screening test from a dangerous one in real clinical AI deployment.