Where AI agent test cases come from: synthetic vs. production data

Playground, synthesizer or real conversations? How the main sources of agent test cases compare, when to use each, and how they work in Rhesis.