PAPER / ARXIV:2609.08094
Dingying Liu, Yunshun Zhong, Wentao Zhang, Yiyuan Li
RESUMO
LLMs are increasingly deployed in public-sector settings, where incorrect guidance can cause irreversible harm. We introduce CIVI, the first framework for diagnosing search agent failures in civic information, spanning cross-national government contexts. We evaluate ten frontier search agents and find none matches an attentive human baseline. Our ARISE method decomposes agentic search failures into four mutually exclusive modes and attributes 72.1% of observed failures to retrieval-bound causes rather than gaps in parametric knowledge.
NO MESMO MAPA