Skip to content
All products

Research agents

Deep research training data.

Research tasks for gathering evidence, comparing sources, and producing analysis with verifiable citations.

Discuss your use case

Connect research claims to evidence.

Research agents need to find evidence, compare sources, resolve gaps, and write an answer that readers can verify. Training examples should retain the decisions and sources behind the final analysis.

We build research tasks around your questions and review standards. Each task connects the objective, supporting evidence, and expected deliverable.

Represent research as a sequence of decisions

Define the question and the evidence needed to answer it. Include incomplete or conflicting sources and show when further research is necessary.

  • Record the objective and evidence requirements.
  • Compare sources rather than collecting links alone.
  • Preserve decisions to refine or extend the investigation.

Review the connection between claims and evidence

Check whether each citation supports its claim. Review coverage, synthesis, acknowledged limitations, and usefulness to the intended reader.

  • Check citations at the claim level.
  • Distinguish sourced findings from inference.
  • Identify omissions and unsupported conclusions.

What a program can include

Scope can include:

  • Long-horizon research tasks and source context
  • Expert research trajectories where scoped
  • Grounded answers with source references
  • Rubrics for evidence quality and synthesis

Illustrative task / Not a production record

Make the objective concrete.

Objective
Compare two technical approaches against a set of project requirements.
Context
The question, required comparison dimensions, and available source material.
Review
Assess whether the analysis covers the requirements, uses supporting evidence, and explains uncertainty or tradeoffs.

Common questions

How do you assess a research answer beyond factual accuracy?

Review source relevance, coverage of the question, support for individual claims, handling of conflicting evidence, and the clarity of the synthesis. The exact criteria should follow the task.

Can research tasks include tool and browser use?

Yes. Tool-calling and browser workflows are related parts of the RDL offering. Agree on the available tools, evidence sources, and environment during scoping.

Further reading

Built around your model

What does your model need to do?

Share representative tasks and examples of a good result. We’ll scope the data and evaluation with you.

Discuss your use case