Use when preparing to fine-tune an LLM for multi-turn conversations, before generating any training data. Triggers - starting a fine-tuning project, need to define evaluation criteria, designing conversation data generation.