ScaffoldAgent: Utility-Guided Dynamic Outline Optimization for Open-Ended Deep Research

2026-06-18

AI Summary

This work addresses the limitations of static or locally optimized outlines in open-ended deep research, which struggle to adapt to structural shifts caused by continuous information accumulation and lack timely feedback on revision efficacy. To overcome these challenges, the paper introduces ScaffoldAgent, a novel framework that models outline evolution as a structured decision process encompassing expansion, contraction, and revision. It incorporates a dynamic optimization mechanism guided by downstream utility metrics—namely retrieval gain, structural coherence, and draft generation quality—to globally inform node selection, operation scheduling, and termination. By integrating structured decision-making, multi-dimensional utility evaluation, and draft-based feedback, ScaffoldAgent establishes an end-to-end system for dynamic outline optimization, significantly improving factual accuracy and overall quality of long-form research reports on the DeepResearch Bench and Gym benchmarks, outperforming existing deep research agents.

Abstract

Open-ended deep research (OEDR) requires systems to acquire knowledge through multi-round retrieval and generate coherent long-form reports. The outline plays a central role as a structural scaffold that coordinates retrieval, evidence organization, and generation. However, existing methods either fix the outline before writing or refine it with local heuristics, leading to scaffold drift under continuous information accumulation and delayed feedback for evaluating outline modifications. We propose ScaffoldAgent, a utility-guided dynamic outline optimization framework for OEDR. ScaffoldAgent models outline evolution as a structured decision process with three operations: Expansion, Contraction, and Revision, enabling controlled updates to the report scaffold. It further introduces a utility-guided feedback mechanism that estimates the downstream value of each outline operation from retrieval gain, structural coherence, and trial-generation quality. The resulting utility signal guides node selection, operation scheduling, and termination during inference. Experiments on DeepResearch Bench and DeepResearch Gym show that ScaffoldAgent consistently improves long-form report generation and factual grounding over existing deep research agents.

Problem

• open-ended deep research
• outline optimization
• scaffold drift
• long-form report generation
• dynamic outline

Innovation

• dynamic outline optimization
• utility-guided feedback
• structured decision process
• open-ended deep research
• scaffold evolution