Xing Chen

Large Language Models (LLMs) have shown strong potential as autonomous agents for complex data analysis. However, extending their capabilities to open-ended insight discovery remains challenging, as failures inevitably arise during multi-step reasoning and often cascade into severe downstream failures. Existing generic self-correction methods struggle to address these issues due to a lack of structural understanding of analytical failures. To tackle this, we propose the \textbf{Taxonomy-Guided Trajectory Critic (TGT-Critic)}, a novel runtime correction framework that detects and mitigates failures within the agent's execution trajectory. To ground our critic with domain-specific diagnostic capabilities, we first construct a fine-grained Agent Failure Taxonomy and an annotated dataset (InsightBench) to systematically map the root causes of analytical failures across six dimensions. By formalizing failure diagnosis as a taxonomy-grounded inference problem, TGT-Critic dynamically provides targeted revisions for both high-level planning and low-level action execution. Experimental results demonstrate that our framework effectively arrests cascading failures, significantly outperforming generic baselines and yielding a 10.52$\%$ relative gain in the quality of generated insights.

Large Language Model (LLM)-based planning agents operate under long, dynamically accumulated contexts that inevitably introduce noise. While it is commonly assumed that such noise impairs reasoning, our study shows that its primary impact lies elsewhere. Through a controlled noise injection framework on the DeepPlanning benchmark, we observe that noise does not significantly degrade the model’s ability to identify relevant information. Instead, it leads to systematic failures in utilizing that information, resulting in plans that are factually correct but violate specific user requirements. Noise also alters agent behavior, leading to reduced tool usage and fewer decision steps. To explain this discrepancy, we propose a two-stage framework for agent behavior under noise: information recognition and information utilization. Empirically, we identify distinct failure modes: some models fail to reliably identify relevant information, while others successfully identify it but fail to consistently apply it in decision-making. Furthermore, we find that selection mechanisms improve performance only when recognition is reliable, and can even be detrimental otherwise. These findings reveal a fundamental gap between recognizing and utilizing relevant information, highlighting consistent information utilization as a key bottleneck for developing robust LLM agents.