---
title: "From Noisy Traces to Root Causes: Structural Trajectory Analysis and Causal Extraction for Agent Optimization"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.07702v1"
permalink: "https://feed.astralor.com/items/itm_7f2f393f535edcbb7cee756d"
published: "2026-07-08T17:57:27.000Z"
score: 77
channel: "tech"
tags: ["Agent Optimization","STRACE","Causal Extraction","Formal verification","Failure Pattern Mining"]
---

# From Noisy Traces to Root Causes: Structural Trajectory Analysis and Causal Extraction for Agent Optimization

## AI commentary

> This paper directly tackles the pain point of 'noisy trajectories undermining agent optimization.' The causal localization approach in STRACE is clever and practical. For teams or researchers developing autonomous agents—especially those facing complex multi-step tasks with debugging difficulties—this is one of the most noteworthy recent works. I recommend reading the full paper and trying the open-source code.

## At a glance

Long-horizon agent optimization often relies on LLM-based reflection, but real execution traces are redundant and heterogeneous, causing inefficient optimization and overfitting. STRACE constructs high-signal-noise optimization contexts via batch-level failure pattern mining and within-trajectory causal localization. On the VeruSAGE-Bench formal verification task, it boosts the success rate of human-expert-designed agents from 42.5% to 58.5% (1.4× improvement). Code is open-sourced.

## Key points

- STRACE filters redundant traces at the batch level and retains representative failures, addressing heterogeneity and redundancy in large trace collections.
- Within each trajectory, it performs causal localization over a textual dependency graph to remove non-causal steps and identify the true root cause module.
- On VeruSAGE-Bench, STRACE significantly outperforms standard context-filtering baselines, achieving a 1.4× improvement in success rate.
- The framework is open-sourced and suitable for automated debugging and optimization of long-horizon agents.

## Full text

The optimization of long-horizon agents increasingly relies on reflection-based mechanisms, where a large language model (LLM) acts as an optimizer to diagnose agent failures and improve agent policies. However, real execution traces are difficult to use directly for optimization: large trace collections are often redundant and heterogeneous, making optimization inefficient and prone to overfitting to low-value failures; meanwhile, each individual trajectory also contains many irrelevant steps, while naive context reduction methods such as truncation or sliding windows can discard causally important evidence and produce misleading optimization signals. To resolve this dilemma, we introduce STRACE (Structural TRajectory Analysis and Causal Extraction), a framework that constructs high signal-noise optimization contexts for more precise and effective optimization. At the batch level, STRACE mines failure patterns to filter redundant traces and retain representative failures; within each selected trace, it performs causal localization over a textual dependency graph to remove non-causal steps and identify the true root-cause module for optimization. Empirical results demonstrate that STRACE significantly outperforms standard context-filtering baselines. Notably, on a challenging formal verification task (VeruSAGE-Bench), it successfully optimizes human-expert designed agents, delivering $1.4\times$ success-rate improvement (42.5% to 58.5%). The code is available at https://github.com/moomight/STRACE .

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_7f2f393f535edcbb7cee756d