263 万字,不可能通读
20 份 10-K/10-Q、38 份补充披露与 30 份附件,只能通过全文检索工具定位账户与期间,再精读关键片段。
EDGAR-US-V6-0052-S1 · Stage 1 · SEC 定期报告审计推理
在 88 份 SEC 文件、263 万字的离线语料中自主检索,识别可被财报观察到的异常信号,并以原文证据支撑每一条发现。
20 份 10-K/10-Q、38 份补充披露与 30 份附件,只能通过全文检索工具定位账户与期间,再精读关键片段。
标签源自 SEC 诉讼公告与 AAER;语料截止在问题被公开确认之前,答案不可能在原文中直接找到。
题库包含从未被 SEC 起诉的硬负例公司。对每家公司都报警的模型会被精确率惩罚,只有真正的审计推理才能得分。
语料以独立检索服务提供(search → fetch → read),作答全程断网,模型可以编写 Python 做多期分析。
以下为 Agent 实际收到的题面节选(原文照录),用于展示任务设定与约束。
You are a financial statement auditor. You have been given the periodic filings of XX INDUSTRIES INC and asked to perform an audit reasoning task.
Identify observable anomalous patterns (red flags) in the financial statements and disclosures. For each red flag, state the affected subject, the shape and direction of the anomaly, the filings it touches, and the filing-based evidence that establishes it.
Two outcomes are possible, and both are legitimate answers. Some companies under review have no red flag that is observable in the supplied periodic filings. On a company with observable red flags, abstention scores zero. Unsupported findings lower precision. Decide from the evidence.
The corpus stops at a cutoff date that precedes any public confirmation of the matters at issue. Filings made after that date are not in the corpus, and neither is any enforcement document, restatement announcement, or news report that would settle the question for you.
The offline case library contains 20 periodic reports, 38 supplemental disclosures, and 30 selected exhibits (2,631,525 characters in total). This is far more text than you can read end to end. Choose accounts and periods before retrieval, use FTS and --grep to locate evidence, and read only the chunks that matter.
NOTE节选自题面,已省略分类表、约束条件与输出结构。
作答容器内可见的全部材料,统计由题包自动生成。原始文件随完整样例包提供。
| 主文档类型(不含附件) | 份数 |
|---|---|
| 8-K | 46 |
| 10-Q | 15 |
| 8-K/A | 10 |
| 10-K/A | 7 |
| 10-K | 6 |
| DEF 14A | 4 |
语料以独立的只读检索服务提供,作答容器内通过 search → fetch → read 命令访问,不直接挂载原始文件。
判据全文、权重与参考答案随完整数据包提供。
前沿模型在真实 Agent 框架中的作答轨迹统计。一个回合即一次模型调用;上下文曲线展示长程任务中状态如何持续累积。
| 模型 | Agent 回合 | 工具调用 | 累计 prompt tokens | 峰值上下文 | 作答耗时 | 结果 |
|---|---|---|---|---|---|---|
| Claude Opus 4.8Claude Code · Harbor, offline | 17 | 880K | 69K | 6m 46s | 完成作答 |
该题所属题库的每一道题都附有一条 Claude Opus 4.8 完整参考轨迹(含全部工具调用与思考),轨迹不含标准答案与判分结果。
| 任务 ID | EDGAR-US-V6-0052-S1 |
|---|---|
| 任务格式 | Harbor task.toml |
| Agent 时限 | 2 小时 |
| 作答容器 | 2 vCPU · 4 GB 内存 |
| 网络 | 完全离线(allowlist 为空) |
| 题面长度 | 15,527 字符 |
| 所属题库 | EDGAR-US v6 · 609 题 |
| 题型 | Stage 1 · 异常发现 |
留下联系方式,我们会提供该题的完整样例包(题面、附件、verifier 与参考轨迹),并介绍同类题目的规模与定制方式。