Fairness Metric Trade-off: Equalized Odds vs Calibration

This simulator makes the algorithmic-fairness impossibility theorem tangible: two synthetic populations, each with its own base rate of true qualification, are scored by the same underlying model and rendered as 440 individually classified points in 3D. Moving each group's decision threshold recomputes the confusion matrix live and reports three standard fairness criteria — demographic parity, equalized odds, and calibration (predictive parity) — so you can watch, in real time, that closing one gap by moving thresholds tends to open another whenever the groups' base rates differ, exactly as proven by Chouldechova (2017) and Kleinberg, Mullainathan & Raghavan (2016).