synthetic-data-gen-eval

A reproducible, tool-neutral benchmark for tabular synthetic-data generation — seven open synthesizers across three public datasets, scored on fidelity, utility, privacy, and compliance, on a CPU laptop at zero licence cost.

Every figure is a pure function of benchmark/results/registry.parquet. Best-in-class result: MOSTLY AI's autoregressive model (top fidelity, near-real utility, membership-inference at chance). See the report for the full analysis.