基准与评测Benchmarks & evaluation
把真实金融研究问题转化为可执行任务，提供明确评分标准与可追溯的过程记录。Turn real financial research questions into executable tasks, explicit criteria, and inspectable trajectories.
查看已发布基准See our published benchmarkPROPHETLAB · APPLIED INTELLIGENCE
以史为鉴，预测万物。Learn from the past.Predict what's next.
我们以专业判断为起点，以真实数据为依据，聚焦于历史回放与未来预测，在回测中构建模拟环境、评估模型能力、检验模型成果。Starting with expert judgment and grounded in real data, we focus on replaying the past and forecasting the future. Through backtesting, we build simulated environments, evaluate model capabilities, and test their results.
我们的关注OUR FOCUS
预测Prediction/推理Reasoning/决策Decision-making
EVIDENCE BEFORE CONFIDENCE.01我们构建什么WHAT WE BUILD
从数据到评测， 从回测到预测。From data to evaluation. From backtests to forecasts.
预测能力需要的不只是一个答案。数据、任务环境和专业评价，必须协同工作。Predictive capability demands more than an answer. Data, task environments, and expert evaluation need to work together.
数据与任务环境Data & task environments
围绕真实研究流程组织信息、工具与时间边界，为评测和智能体工作提供基础。Structure information, tools, and time boundaries around genuine research workflows.
洽谈数据合作Discuss data access按需求对接 · 暂无在线购买Access by arrangement · no self-service checkout专业判断与协作Expert judgment & collaboration
通过 Council of Fellows，连接领域专家、命题与独立评审，把金融工作中的真实难题带进评测。Council of Fellows connects domain experts, task design, and independent review to bring real challenges from financial work into evaluation.
了解专家合作Explore expert collaboration02已发布基准OUR FIRST PUBLIC BENCHMARK
FINANCE · AGENT EVALUATIONInvestor's
Last Exam v1
让智能体回到过去，面对真正的股票研究任务。共享的数据环境、严格的信息时点，以及可以逐步复核的研究过程。Take agents back in time to tackle real equity research tasks. A shared data environment, enforced information cutoffs, and work that can be examined step by step.
进入 Investor's Last Exam v1Explore Investor's Last Exam v1阅读研究文章Read the research story →浏览全部基准Browse all benchmarks →03与专家共同构建BUILT WITH EXPERTS
专业的事， 由专业的人来做。Professional work, in expert hands.
专业知识不只存在于答案里，也存在于怎样提出问题、寻找证据和识别错误。我们希望把这些判断带进 AI 的评测与改进。Expertise lies not only in answers, but in how people frame questions, find evidence, and identify errors. We want those judgments to shape how AI is evaluated and improved.
01提出有价值的问题Define meaningful questions
02建立有依据的标准Set evidence-based criteria
03独立复核模型表现Review model work independently
04关于 ProphetLabABOUT PROPHETLAB
理解变化， 建立判断。Understand change. Build judgment.
我们是一家专注预测、推理与决策的 AI 科技公司。金融市场，是我们验证能力的第一个高要求场景。We are an AI company focused on prediction, reasoning, and decision-making. Financial markets are our first demanding proving ground.
我们围绕多智能体、世界模型与大模型后训练开展工作，并将数据、专业经验和可验证的评价连接起来。Our work spans multi-agent systems, world models, and post-training, connecting data and professional experience with verifiable evaluation.
探索与我们共事的机会Explore working with usLET’S BUILD SOMETHING THAT HOLDS UP.