| --- |
| pretty_name: LongHorizon Full Data-Agent Benchmarks |
| task_categories: |
| - question-answering |
| - tabular-classification |
| - tabular-regression |
| - text-generation |
| tags: |
| - agents |
| - data-analysis |
| - benchmark |
| --- |
| |
| # LongHorizon Full Data-Agent Benchmarks |
|
|
| Companion data artifacts for five complete evaluation tracks: |
|
|
| - DataSciBench full55 / 167 metric entries |
| - DABStep full450 |
| - DABStep-Research full100 |
| - DSBench Modeling full74 |
| - LongDS full68 / 2,225 turns |
|
|
| The companion GitHub repository contains processed manifests, evaluation code, |
| historical API ReAct baseline code, download/preparation tools, and the frozen |
| source lock. `artifact_manifest.json` records every uploaded object's size, |
| SHA-256, installation path, and archive contents. |
|
|
| DataSciBench runtime trees and DSBench extracted directories are intentionally |
| not duplicated here. They are reconstructed from the compact source artifacts. |
| DABStep leaderboard submission history is excluded; all 450 task definitions |
| and all seven context files are included. LongDS is split by dataset family so |
| downloads can resume without a monolithic 19 GB archive. |
|
|
| The DABStep project metric uses six official public-dev references and 444 |
| DataCOPE Genesis proxy references. It is not an official hidden-leaderboard |
| full-test score. |
|
|
|
|
