Benchmark
EDA Benchmark
Test how accurately and consistently AI agents solve data-analysis problems.
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
Ten synthetic tasks spanning data cleaning, aggregation, anomaly detection, time-series analysis, and reconstruction of events.
Methodology
Agents return structured answers checked against task-specific ground truth. Five runs per task measure accuracy and variation; a reliability-adjusted score penalizes inconsistent performance. Models use their provider-specific harness or opencode.