Students conduct a rigorous stress test of their mitigated models to visualize the trade-off between predictive accuracy and fairness. They write a technical report justifying the 'cost' of fairness for specific use cases (e.g., medical diagnosis vs. ad targeting).