Facial Recognition Threshold: Bias vs. Accuracy Simulator

A crowd streams through a 3D checkpoint gate; each person's face gets a similarity score drawn from real-looking impostor/genuine statistical distributions, and the current match threshold decides who is flagged as a watchlist hit. Raising the threshold cuts false positives but misses more real matches; widening the demographic error gap reproduces the documented finding that identical algorithms and thresholds can still produce unequal false-match rates across population groups. Live readouts track overall precision, the per-group false-positive rate and how many wrongful stops actually get actioned once you factor in a human-review step — the same accuracy/bias/oversight trade-off at the center of the EU AI Act's biometric restrictions and the city-level facial-recognition bans debated since 2019.