Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep Researcher with Sequential Plan Reflection and Candidates Crossover (Deep Researcher Reflect Evolve)

Saurav Prateek

TL;DR

The paper addresses the challenge of producing high-quality, PhD-level research reports by moving beyond parallel sub-task scaling to a sequential refinement paradigm. It introduces the Deep Researcher with Sequential Plan Refinement via Reflection and a Candidate Crossover mechanism, supported by a Global Research Context and a one-shot report writer. Evaluated on the DeepResearch Bench using RACE and FACT criteria, the approach achieves 46.21 overall and demonstrates superior coherence and fact density compared to several leading DRAs, highlighting the advantages of sequential scaling for integrated reasoning. The work suggests that maintaining a centralized memory and dynamic plan adaptation can significantly improve both the accuracy and efficiency of automated doctoral-level research generation.

Abstract

This paper introduces a novel Deep Researcher architecture designed to generate detailed research reports on complex PhD level topics by addressing the inherent limitations of the Parallel Scaling paradigm. Our system utilizes two key innovations: Sequential Research Plan Refinement via Reflection and a Candidates Crossover algorithm. The sequential refinement process is demonstrated as an efficient method that allows the agent to maintain a centralized Global Research Context, enabling it to look back at current progress, reason about the research plan, and intelligently make changes at runtime. This dynamic adaptation contrasts with parallel approaches, which often suffer from siloed knowledge. The Candidates Crossover algorithm further enhances search efficiency by deploying multiple LLM candidates with varied parameters to explore a larger search space, with their findings synthesized to curate a comprehensive final research response. The process concludes with One Shot Report Generation, ensuring the final document is informed by a unified narrative and high fact density. Powered by the Gemini 2.5 Pro model, our Deep Researcher was evaluated on the DeepResearch Bench, a globally recognized benchmark of 100 doctoral level research tasks. Our architecture achieved an overall score of 46.21, demonstrating superior performance by surpassing leading deep research agents such as Claude Researcher, Nvidia AIQ Research Assistant, Perplexity Research, Kimi Researcher and Grok Deeper Search present on the DeepResearch Bench actively running leaderboard. This performance marginally exceeds our previous work, Static DRA, and reinforces the finding that sequential scaling consistently outperforms the parallel self consistency paradigm.

Deep Researcher with Sequential Plan Reflection and Candidates Crossover (Deep Researcher Reflect Evolve)

TL;DR

The paper addresses the challenge of producing high-quality, PhD-level research reports by moving beyond parallel sub-task scaling to a sequential refinement paradigm. It introduces the Deep Researcher with Sequential Plan Refinement via Reflection and a Candidate Crossover mechanism, supported by a Global Research Context and a one-shot report writer. Evaluated on the DeepResearch Bench using RACE and FACT criteria, the approach achieves 46.21 overall and demonstrates superior coherence and fact density compared to several leading DRAs, highlighting the advantages of sequential scaling for integrated reasoning. The work suggests that maintaining a centralized memory and dynamic plan adaptation can significantly improve both the accuracy and efficiency of automated doctoral-level research generation.

Abstract

This paper introduces a novel Deep Researcher architecture designed to generate detailed research reports on complex PhD level topics by addressing the inherent limitations of the Parallel Scaling paradigm. Our system utilizes two key innovations: Sequential Research Plan Refinement via Reflection and a Candidates Crossover algorithm. The sequential refinement process is demonstrated as an efficient method that allows the agent to maintain a centralized Global Research Context, enabling it to look back at current progress, reason about the research plan, and intelligently make changes at runtime. This dynamic adaptation contrasts with parallel approaches, which often suffer from siloed knowledge. The Candidates Crossover algorithm further enhances search efficiency by deploying multiple LLM candidates with varied parameters to explore a larger search space, with their findings synthesized to curate a comprehensive final research response. The process concludes with One Shot Report Generation, ensuring the final document is informed by a unified narrative and high fact density. Powered by the Gemini 2.5 Pro model, our Deep Researcher was evaluated on the DeepResearch Bench, a globally recognized benchmark of 100 doctoral level research tasks. Our architecture achieved an overall score of 46.21, demonstrating superior performance by surpassing leading deep research agents such as Claude Researcher, Nvidia AIQ Research Assistant, Perplexity Research, Kimi Researcher and Grok Deeper Search present on the DeepResearch Bench actively running leaderboard. This performance marginally exceeds our previous work, Static DRA, and reinforces the finding that sequential scaling consistently outperforms the parallel self consistency paradigm.
Paper Structure (10 sections, 6 figures, 2 tables)

This paper contains 10 sections, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Sequential Refinement approach vs Parallel Scaling
  3. Deep Researcher Design
  4. High Level Design
  5. Candidate Crossover algorithm
  6. Agent’s Memory: Global Research Context
  7. Sequential Research Plan Refinement via Reflection
  8. One Shot Report Generation
  9. Evaluation
  10. Conclusion

Figures (6)

  • Figure 1: Comparison of Deep Researchers on 4 key dimensions of RACE framework
  • Figure 2: Deep Researcher - High Level Design
  • Figure 3: Deep Researcher - Candidate Crossover algorithm
  • Figure 4: Allocation of tasks among fields and Distribution of task counts by language
  • Figure 5: Comparison of Language tasks on 4 key dimensions of RACE framework
  • ...and 1 more figures