Research Benchmarks

Misata includes an offline benchmark harness for comparing a proof-backed Misata workflow with a hand-written Faker baseline.

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.