Supervised learning examples
Instructions, relevant context, and carefully reviewed responses organized for a defined learning objective.
Discuss this formatOur products
Data built around capability.
The right format depends on the behavior you want to develop. Explore the building blocks for training, feedback, and evaluation, then work with us to define a suitable brief.
Evaluation task sets
Held-out problems and acceptance criteria that help assess a specific capability or workflow.
Discuss this format
Rubrics & verifiers
Explicit grading guidance and checks for outputs whose quality cannot be captured by surface similarity alone.
Discuss this format
Preference data
Comparisons that record which response better satisfies the task, with consistent criteria for the choice.
Discuss this formatTool-use environments
Tasks, tool interfaces, and observable state changes for agents learning to act across connected systems.
Discuss this formatComputer-use demonstrations
Recorded steps through browser and desktop workflows, paired with the task context and completion conditions.
Discuss this formatSoftware tasks
Implementation briefs, debugging examples, and tests that connect a coding request to verifiable behavior.
Discuss this formatDomain-specific tasks
Material shaped around the terminology, constraints, and review practices of a professional field.
Discuss this formatResearch workflows
Evidence-gathering and synthesis tasks with attention to source quality, uncertainty, and attribution.
Discuss this formatMultimodal tasks
Learning and evaluation material for questions that depend on more than a single form of input.
Discuss this formatFailure analysis
A structured review of errors that helps identify where additional examples or a different task design may be useful.
Discuss this formatCustom data programs
A scoped combination of task design, data preparation, review, and documentation for a particular development goal.
Discuss this formatA useful dataset starts with a precise brief.
Share the capability, domain, preferred format, and constraints. We can then discuss a scope that fits the work.