TRACE: Instruction Guide

TRACE is a framework designed to audit YouTube’s recommendation algorithm in a scalable and reproducible way. It uses LLM-driven personas to simulate realistic user behavior and explore recommendation dynamics. The system separates initial context from personas, enabling controlled and comparable experiments. Its modular architecture ensures scalability, transparency, and full traceability of results. This presentation serves as a user guide, explaining how to configure, run, and monitor TRACE experiments.