Frozen honeypot corpus for "A Validated Measurement of the Adversarial-LLM Base Rate in Traffic Against Exposed LLM Infrastructure"

Frozen honeypot capture corpus underlying the manuscript "How Much of It Is Real? A Validated Measurement of the Adversarial-LLM Base Rate in Traffic Against Exposed LLM Infrastructure," submitted to IEEE Transactions on Dependable and Secure Computing. The corpus contains 79,401 events captured over a 40.95-day window of passive observation (first event 2026-04-06, last event 2026-05-16) against LLM-shaped honeypot facades. Each event is a JSON object recording the source address, geolocation and network attribution, the HTTP request, and classification metadata. Events reflect inbound traffic from external sources only; no facade responses or internal infrastructure data are included. The file is a newline-delimited JSON (NDJSON) export, provided gzip-compressed. SHA-256 of the uncompressed NDJSON: 547c3304bd21018c69a277ccf2bda5f4b646ebd49f810f16f740e6198d54805e This deposit is the frozen data source for every quantity reported in the manuscript and is provided to support independent reproduction.本冻结蜜罐捕获语料库为投稿至《IEEE Transactions on Dependable and Secure Computing》的论文《真实占比几何？针对暴露的大语言模型（LLM）基础设施流量的对抗性大语言模型（Adversarial-LLM）基准率验证测量》的支撑数据集。 该语料库包含在40.95天的被动观测周期内捕获的79401条事件（首条事件时间为2026年4月6日，末条事件时间为2026年5月16日），采集目标为仿大语言模型的蜜罐伪装节点。每条事件均为JSON对象，记录了源IP地址、地理定位信息、网络归属属性、HTTP请求内容以及分类元数据。所有事件仅反映外部来源的入站流量，未包含伪装节点的响应内容或内部基础设施数据。 该文件采用换行分隔JSON（NDJSON）格式导出，并以gzip压缩形式提供。未压缩的NDJSON文件的SHA-256哈希值为：547c3304bd21018c69a277ccf2bda5f4b646ebd49f810f16f740e6198d54805e。 本数据集为该论文中所有报告数值的冻结数据源，旨在支持研究人员独立复现相关实验结果。