In models that support it (like Claude), how does "pre-filling" the assistant's response (e.g., starting with {) help with structured output?

In an API inference call, what is the functional difference between the System, User, and Assistant roles?

Instead of one giant 2,000-word prompt, why might you split a task into smaller, sequential prompts?

Why are delimiters like ###, """, or <xml> tags important in long prompts? How do they help prevent the model from getting confused between instructions and data?