Tag
This paper presents a unified benchmark comparing outline-driven frameworks for long-form generation with LLMs, evaluating outlines independently and finding that performance varies with framework-granularity match, supporting the decoupling of outline and writing stages.