# Research Benchmarks Misata includes an offline benchmark harness for comparing a proof-backed Misata workflow with a hand-written Faker baseline. ```bash python benchmarks/bench_research_moat.py ``` The harness reports: - rows per second - foreign-key orphan count - reproducibility with a fixed seed - scenario-control support - Oracle report availability - a simple setup-effort proxy The point is not to claim that one library is always faster. The point is to make Misata's research claim measurable: it generates relational data and a proof report from one high-level description, while manual baselines need custom glue code for schema, relationships, scenario logic, and checks.