lbox/kcl

KCL（韩国标准法律基准）数据集包含两个子集：KCL-Essay和KCL-MCQA。KCL-Essay包含169个开放式问题，550个支持性先例和2,739个实例级评分标准，用于评估模型的开放式生成能力。KCL-MCQA包含283个五选一选择题和1,103个支持性先例，用于评估模型的选择题回答能力。数据集的设计目的是区分知识覆盖和基于证据的推理能力，适用于法律检索增强生成（RAG）研究和自动化评分。数据集语言为韩语，主要用于问答任务，标签为法律领域，规模小于1K，许可证为CC-BY-NC-4.0。

The KCL (Korean Canonical Legal Benchmark) dataset consists of two subsets: KCL-Essay and KCL-MCQA. KCL-Essay includes 169 open-ended questions, 550 supporting precedents, and 2,739 instance-level rubrics for evaluating open-ended generation capabilities. KCL-MCQA includes 283 five-choice multiple-choice questions and 1,103 supporting precedents for evaluating multiple-choice question answering capabilities. The dataset is designed to disentangle knowledge coverage from evidence-grounded reasoning, making it suitable for legal Retrieval-Augmented Generation (RAG) research and automated scoring. The dataset is in Korean, primarily intended for question-answering tasks, tagged under the legal domain, with a size category of n<1K, and licensed under CC-BY-NC-4.0.