Benchmark RadarDRFLOW
DRFLOW evaluates an agent's ability to predict personalized workflows, sequences of action-steps, from heterogeneous sources across five domains. It includes 100 tasks with reference workflow steps and multiple diagnostic metrics.
- Released
- 2026-06-16
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Deep research systems are typically evaluated on report generation, but enterprise tasks often require actionable workflows. DRFLOW addresses this gap by assessing workflow prediction, offering metrics for factual grounding, step recovery, and personalization, which can guide development of more practical agents.
Motivation
Deep research (DR) systems are increasingly used for complex information-seeking tasks, but existing works mainly focus on generating reports and summaries.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.