ScaffoldAgent Dynamically Optimizes Outlines for Long-Form Deep Research

ScaffoldAgent treats outline evolution as a structured decision process with Expansion, Contraction, and Revision operations, using a utility-guided feedback mechanism that scores each operation on retrieval gain, structural coherence, and trial generation quality. This targets scaffold drift that plagues fixed or locally-refined outlines in multi-round retrieval pipelines.