Synthetic Tabular Generators Fail to Preserve Behavioral Fraud Patterns: A Benchmark on Temporal, Velocity, and Multi-Account Signals

Develop a benchmark suite specifically for synthetic tabular data generators focused on fraud detection. The suite should test for temporal and velocity signal retention, not just static statistical distribution matching.

Suggested repo: fraud-bench

"Stop training synthetic data that hides the fraud."

Estimated effort: 35h