tldc/diligence-bench

DiligenceBench是一个包含50项分析任务的基准数据集，涵盖美国大型加速股票，涉及能源、银行、生物技术、保险、技术、房地产投资信托基金（REITs）、餐饮、工业和公用事业等多个领域。数据集包含48个不同发行方的数据，分布在9个行业。每个任务包含三个部分，平均38个二元标准，涉及事实准确性、分析推理和风险意识。数据集旨在评估模型在金融领域的分析能力，如撰写卖方或买方报告、信用委员会备忘录、并购尽职调查、私人信贷交易承销、重组咨询、季度收益预览和反应笔记、法务会计工作流以及激进做多或做空论点备忘录等任务。

DiligenceBench is a 50-item benchmark of analytical tasks across large-accelerated US equities spanning energy, banking, biotech, insurance, technology, REITs, restaurants, industrials, and utilities. The dataset includes data from 48 distinct issuers across 9 sectors. Each task consists of three sections with an average of 38 binary criteria, covering factual accuracy, analytical reasoning, and risk awareness. The dataset is designed to evaluate models analytical capabilities in finance, such as initiating sell-side or buy-side coverage, writing credit committee memos, staging M&A diligence, underwriting private credit transactions, producing restructuring-advisory workstreams, writing quarterly earnings preview and reaction notes, running forensic accounting workflows, and producing activist short or long thesis memos.