DataClaw: Benchmark Tasks for End-to-End Data Analysis Agents

A benchmark dataset for evaluating OpenClaw-style end-to-end agents on data analysis tasks. Every task is grounded in real-world data and has a single objective gold answer. The dataset was created by GTML-LAB and was last updated on April 30, 2026.

Use Cases

• Benchmarking agent performance on tasks requiring evidence location across heterogeneous files.
• Evaluating agent capabilities in filtering and processing real-world data.
• Training end-to-end agents to handle multi-step data analysis with a single objective answer.

Strengths

• Tasks are grounded in real-world data, providing practical evaluation scenarios.
• Each task has a single objective gold answer, enabling clear performance measurement.
• The dataset was last updated on April 30, 2026.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

GTML-LAB

Freshness

Last updated 2026-04-30 07:14:23; freshness should be verified.