Complete multi-hop research trajectories capturing expert reasoning from plan to resolution.
Available Now · Sample Available
Deep Research
End-to-end expert research trajectories: plans, searches, source conflicts, and resolutions.
Volume
0
Volume
Scalable Volume
Coverage
Finance, Coding, Physics, Legal, Medical, Mechanical Engineering
62.5%Opus 5 · Pass@1
Targets
BrowseComp, GAIA, DeepResearch Bench
Modality & Delivery
Text→Text
Formats
JSONL (.jsonl).sft
What You Receive
Coverage
Package Structure
Differentiators
- Captures the plan, searches, source evaluation, conflict resolution, and unresolved gaps as one end-to-end trajectory.
Record Schema
action_maskarrayOptional mask identifying trainable or scoreable actions in the trajectory.
answer_aliasesarrayAccepted alternative forms of the answer.
applicable_time_windowstringTime period within which the answer and evidence are applicable.
citation_dpoobjectCitation-level preference data and sentence attribution.
competencies_exercisedarrayResearch and reasoning competencies evaluated by the task.
conflictsobjectConflicting claims, resolution strategy, and unresolved conflicts.
contamination_checkobjectAssessment of whether the task can be answered without external search.
cross_document_requirementnumberScore indicating how strongly the task requires evidence across documents.
difficulty_calibrationobjectEstimated steps and baseline accuracy measures used to calibrate difficulty.
dpoobjectChosen and rejected trajectories with preference source and margin.
evidencearrayEvidence sources with titles, URLs, excerpts, credibility, primary-source status, and conclusion usage.
evidence_utilitynumberAggregate utility score for collected evidence.
format_penaltynumberPenalty applied for format violations.
format_validbooleanWhether the response satisfies required output formatting.
gain_beyond_ragnumberEstimated performance gain beyond a retrieval-augmented generation baseline.
grading_rubricobjectRubric items and search-based, evolving, and variance settings used for grading.
grpoobjectGroup-relative optimization metadata and reward statistics.
idstringStable identifier for the RL research record.
multi_step_necessitynumberScore indicating how strongly the task requires multi-step reasoning.
noise_infoobjectNoise sources and severity associated with the research task.
outcome_rewardnumberOverall numerical reward assigned to the completed trajectory.
process_rewardsarrayStep-level rewards covering information gain, query alignment, redundancy, execution success, and value estimates.
questionstringDeep-research question presented to the agent.
reward_typestringReward formulation used for the RL record.
safety_labelsobjectRisk, harmful-content, PII, browsing, disclaimer, and refusal annotations.
task_typestringResearch task classification.
timestampstringTimestamp associated with record creation or execution.
tokenizerobjectTokenizer name, version, and rendering configuration.
tool_definitionsarrayTools available to the agent, including schemas, limits, and timeouts.
trajectoryarrayOrdered research steps containing thoughts, actions, and observations.
wrong_pathsarrayKnown unproductive queries or reasoning paths and explanations of their errors.
Results
| Model | Metric | Result |
|---|---|---|
| Opus 5 | Pass@1 |
Methodology & Quality
Methodology
SME involvement
Domain-expert curated with up to L2 review
Stressors
Multi-hop search, conflicting sources, evidence synthesis, and unresolved uncertainty.
Design criteria
- Preserve complete multi-hop research traces rather than final answers alone.
Build assumptions
Quality approach
- Legal sample trajectories include rubric-based rewards and retained comparison groups.
Known limitations
Customization & Scalability
Custom sets available. Additional volume in progress
Get Started
Request a Sample
Share your target domains and the OTS team will follow up with sample data and next steps.