[Re]train agents that research reliably.
Source quality, citation accuracy, and synthesis depth — scored on real document sets with real contradictions. Agents learn to distinguish signal from noise.
Your analysts get research they can act on.
Source quality is scored on preference for primary sources, cross-document verification, and contradiction detection. The trainer iterates until the agent consistently reaches for the strongest evidence — not just the first result.
Every claim backed by a specific source.
Page numbers, sections, dates — not vague references. Citation accuracy is scored as F1 against ground truth attributions. The training curve shows improvement iteration by iteration, measured on documents the agent hasn't seen.
Findings connected across sources into usable reports.
Synthesis is scored on the agent's ability to connect findings across multiple documents, resolve contradictions, and build coherent narratives from fragments. Not just retrieval — understanding.
Ready to train your research agents?
Source quality, citation, and synthesis — scored on your documents.