Inference-Time Scaling of Verification: Self-Evolving Deep Research Agents via Test-Time Rubric-Guided Verification

TL;DR

This work tackles the unreliability of Deep Research Agents by introducing DeepVerifier, a rubric-guided, inference-time verification framework that leverages an automatically constructed DRA Failure Taxonomy to decompose verification into tractable sub-tasks. By integrating a decomposition, verification, and judge module, it enables self-evolving feedback loops at test time, achieving notable improvements in GAIA and XBench-DeepResearch without retraining. The authors further provide DeepVerifier-4K, a supervised fine-tuning dataset enabling open-model reflection and verification through Fine-Tuning, and demonstrate performance gains across multiple backbones and benchmarks. The approach offers a scalable, plug-and-play mechanism to enhance reliability and advance open-source capabilities in automated knowledge discovery systems.

Abstract

Recent advances in Deep Research Agents (DRAs) are transforming automated knowledge discovery and problem-solving. While the majority of existing efforts focus on enhancing policy capabilities via post-training, we propose an alternative paradigm: self-evolving the agent's ability by iteratively verifying the policy model's outputs, guided by meticulously crafted rubrics. This approach gives rise to the inference-time scaling of verification, wherein an agent self-improves by evaluating its generated answers to produce iterative feedback and refinements. We derive the rubrics based on an automatically constructed DRA Failure Taxonomy, which systematically classifies agent failures into five major categories and thirteen sub-categories. We present DeepVerifier, a rubrics-based outcome reward verifier that leverages the asymmetry of verification and outperforms vanilla agent-as-judge and LLM judge baselines by 12%-48% in meta-evaluation F1 score. To enable practical self-evolution, DeepVerifier integrates as a plug-and-play module during test-time inference. The verifier produces detailed rubric-based feedback, which is fed back to the agent for iterative bootstrapping, refining responses without additional training. This test-time scaling delivers 8%-11% accuracy gains on challenging subsets of GAIA and XBench-DeepResearch when powered by capable closed-source LLMs. Finally, to support open-source advancement, we release DeepVerifier-4K, a curated supervised fine-tuning dataset of 4,646 high-quality agent steps focused on DRA verification. These examples emphasize reflection and self-critique, enabling open models to develop robust verification capabilities.

Figures (3)

• Figure 1: Upper: Inference-time scaling of verification on the full GAIA development set ($n = 165$). Lower: Performance comparison between DeepVerifier-8B fine-tuned on our dataset and other open-sourced models after 10 rounds of verification & feedback on the full GAIA development set.
• Figure 2: Overview of DeepVerifier, which decomposes complex verification problems into smaller, simpler sub-questions leveraging the asymmetry of verification, and provides corrective feedback for the DRA to retry when the answer is considered incorrect.
• Figure 3: DRA failure taxonomy that categorizes 555 agent failures into five major classes and thirteen subclasses.