Beyond Final Answers: Auditing Trajectory-Level Hallucinations in Multi-Agent Industrial Workflows

Harshada Badave, Santosh Borse, Andrea Gomez, Harshitha Narahari, Sara Carter, Vishwa Bhatt, Aishani Rachakonda, Shuxin Lin, Dhaval Patel

arXiv:2605.24219 · 2026-05-27 공개 · arXiv · PDF

long-context multi-agent assetopsbench agentic-llm llm-audit trajectory-aware industrial-workflow trajectory-hallucination

Abstract

Large Language Models (LLMs) are increasingly deployed as autonomous agents that reason, use tools, and act over multiple steps. Yet most hallucination benchmarks still evaluate only the final output, missing failures that originate in intermediate Thought-Action-Observation steps. We present Trajel, a dataset and evaluation framework for auditing trajectory-level hallucinations in multi-agent industrial workflows. Trajel introduces a five-type hallucination taxonomy (factual, referential, logical, procedural, and scope-based) over expert-annotated agent traces from AssetOpsBench. We benchmark supervised detection models at the subtask, trajectory, and long-context levels. Our results show that the most common failure modes are missed by existing benchmarks, that nearly half of hallucinated trajectories involve multiple types at once, and that automated detectors with high binary accuracy still misclassify the subtlest types. Trajectory-aware detection significantly outperforms standard post-hoc verification, making taxonomy-grounded evaluation necessary for safer agentic deployment.

한국어 요약

📋 한 줄 요약

**[Multi-Agent 환각 평가 / Trajectory-Level]** Trajel이 5-type 환각 taxonomy(factual·referential·logical·procedural·scope)로 Thought-Action-Observation 단계의 환각 평가 — 환각 trajectory의 거의 절반이 multiple type 동반, trajectory-aware 검출이 post-hoc 대비 significant 우월.

🎯 핵심 기여도

💡 핵심 아이디어

LLM 에이전트의 환각은 최종 출력이 아니라 중간 Thought-Action-Observation 단계에서 발생·전파하므로, 5-type taxonomy로 trajectory를 fine-grained 분류하고 trajectory-aware로 검출해야 하며, 환각의 거의 절반이 multiple type을 동시에 동반하므로 단일 type 검출로는 불충분하다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: Final-only 환각 평가의 paradigm shift 제안, 5-type taxonomy로 환각의 결성도 정량화, industrial workflow에 특화된 evaluation으로 실용성, trajectory-aware detection의 우월성 입증. **한계**: AssetOpsBench 도메인 중심으로 일반 도메인 확장 검증, expert-annotation 비용·확장성, subtle type 자동 검출이 여전히 open challenge.

🚀 실용적 활용