Check synthetic data quality: fidelity, utility, privacy

Compare distributions and correlations, train-on-synthetic/test-on-real, and check no synthetic row copies a real one. Concepts: Fidelity (KS test, correlation difference); utility (TSTR); privacy (distance to closest record) Project A: Train logistic regression on synthetic, test on real test: compare ROC-AUC with training on real. Project B: Check simulated heat-wave inputs stay physically plausible (temperatures, humidity, hour patterns).