Synthesis Through Simulation: Generating Coherent Enterprise Data via Scalable Agent-System Interaction
STS generates enterprise-style datasets by making an agent operate inside simulated business systems, without seeing the database schema.
The paper frames this as a way around legal and business limits on using real enterprise data for tool-calling agents. Its Generalist Populator reports 0.88 average marginal fidelity and 100% constraint satisfaction across ten simulated environments. The authors argue validity comes from the APIs and policies of the environment itself, rather than from a separate schema-matching step. They say the framework, environments, and generated datasets are open-sourced. HF Daily Papers' note
The paper frames this as a way around legal and business limits on using real enterprise data for tool-calling agents. Its Generalist Populator reports 0.88 average marginal fidelity and 100% constraint satisfaction across ten simulated environments. The authors argue validity comes from the APIs and policies of the environment itself, rather than from a separate schema-matching step. They say the framework, environments, and generated datasets are open-sourced. HF Daily Papers' note
score 5