BTF-2/BTF-2

BTF-2（Bench to the Future 2）是一个包含1,417个过去预测问题的数据集，带有预测、解决结果和研究摘要，旨在无后见之偏或网络数据泄露的情况下，对预测代理进行可重复评估。问题最初于2025年10月提出，并于2025年12月解决。数据包括二元预测问题，解决分布为1,030个否（0）和387个是（1）。主题涵盖监管和政策行动（约23%）、美国政府及公共政策（约13%）、宏观经济和市场（约12%）、国际安全与外交（约11%）、法院案件和调查（约10%），以及加沙战争外交、极端天气、伊朗核计划、商业太空发射、COP30气候谈判和选举等较小集群。许可证为CC-BY-NC-4.0（用于研究和学术用途；商业用途需获得许可）。数据集模式包括问题ID、问题文本、解决标准、背景、研究摘要、当前日期、解决结果、解决解释、SOTA预测概率和SOTA摘要理由等列。伴生文件包含2025年10月抓取的网页索引，用于支持预测。BTF-2 (Bench to the Future 2) is a dataset of 1,417 pastcasting questions with forecasts, resolutions, and research summaries, designed for reproducible evaluation of forecasting agents without hindsight bias or web-data leakage. The questions were originally asked in October 2025 and resolved in December 2025. It includes binary forecasting questions with a resolution distribution of 1,030 No (0) and 387 Yes (1). Topics cover regulatory and policy actions (~23%), US government and public policy (~13%), macroeconomics and markets (~12%), international security and diplomacy (~11%), court cases and investigations (~10%), and smaller clusters such as Gaza war diplomacy, extreme weather, Irans nuclear program, commercial space launches, COP30 climate negotiations, and elections. The license is CC-BY-NC-4.0 (for research and academic use; commercial use requires permission). The schema includes columns like question_id, question, resolution_criteria, background, research_summary, present_date, resolution, resolution_explanation, sota_forecast_probability, and sota_summary_rationale. A companion file provides scraped web pages from October 2025 to support forecasting.