Finance Long-Horizon Agent Tasks
A structured dataset of expert-curated, long-horizon finance tasks designed to train and evaluate AI agents on real-world financial research, modeling, analysis, and deliverable production.
Overview
Copula Finance Expert Long-Horizon Tasks is a professional task corpus defined, constructed, and accepted by senior frontline finance practitioners.
Each datapoint is not an isolated Q&A, but a complete finance work package: the model must understand the task background and boundaries, read multiple source materials, complete verification, calculation, judgment, writing or modeling, and deliver a review-ready Word, Excel, or PowerPoint artifact. The dataset targets the core bottlenecks of real professional workflows: long-context information organization, cross-file reconciliation, financial calculation, regulatory and compliance boundaries, evidence-chain construction, and professional judgment.
Experts are drawn from sell-side research, accounting, investment, valuation, and professional financial training. They do not only provide conclusions — they define role boundaries, real workflows, critical difficulties, acceptable deliverables, and itemized scoring criteria.
Built for
- RL environments for long-horizon agent training
- Evals scored against gold deliverables
- Post-training on expert reasoning traces
Featured samples
Prompt
You are a TMT and small-/mid-cap electronics analyst at China Merchants Securities Research Institute. The current date is June 2, 2026. Creality (03388.HK), the first publicly listed consumer 3D-printing brand, has listed on the Main Board of the Hong Kong Stock Exchange and published a prospectus disclosing operating and financial data for 2022–2025. The research institute plans to initiate coverage.
Using the task attachments and public information available as of June 2, 2026, prepare an in-depth company research report for institutional investors. Assess the company’s business model, market opportunity, competitive position, profitability, and growth; independently build earnings forecasts and a valuation for 2026–2028; and provide an investment rating, target price (or reasonable valuation range), and investment recommendation. No rating direction is predetermined; the conclusion must follow from the analysis. Do not use information after the cut-off date, fabricate data, or use inside information or unverified rumors.
Deliver a single .docx file. There is no hard page limit; completeness and focus should determine length (generally 15–25 pages). The report must meet professional and compliance standards for sell-side research, including analyst information, rating methodology, risk disclosures, a disclaimer, and disclosure of interests relating to the covered company. Include forecasts for the principal financial statements in an appendix.
Gold
Space is limited; showing 3 of 25 pages.
Creality initiation report
Browsable · Document
A sell-side analyst builds a segment-level forecast, three-statement model and valuation for the first listed consumer 3D-printing brand.
Build the dataset your model is missing.
Every product here is built with domain experts and delivered to your training or evaluation spec. Tell us what you need.
Discuss your project