Sign in to view source links and access this dataset
Description
FDAbench-Full is a benchmark dataset designed for evaluating data agents in multi-source analytical scenarios. It contains 2,007 diverse tasks across different data sources, domains, difficulty levels, and task types. The dataset was created by FDAbench2026 and was last updated on May 20, 2026.
Use Cases
Benchmarking data agent performance based on the described multi-source analytical tasks.
Evaluating agent-expert collaboration frameworks based on the dataset's generation methodology.
Training data agents across different domains and difficulty levels as described.
Researching task diversity and complexity in data agent evaluation.
Strengths
Contains 2,007 diverse tasks for evaluation.
Designed for multi-source data analytical scenarios.
Tasks span different domains, difficulty levels, and types.
Limitations
Column-level documentation is absent; field semantics must be inferred after download.
Row count and file size are unknown, which may limit suitability assessment.
Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
Source
FDAbench2026
Collection Method
Constructed via an agent-expert collaboration dataset generation framework.
Freshness
Last updated 2026-05-20 06:07:38; freshness should be verified.
License is unknown; users should verify terms of use before downloading.