Skip to content

Every project starts with a brief and a sample.

You can start from a new brief or from one of our products: professional work, codebases and tasks, agent environments, business environments or computer-use trajectories.

01Stages
  1. 01

    Brief

    Tell us what you are building and what the data needs to do for you.

    A written brief that says what good data looks like for you.

  2. 02

    Sample

    You review a sample in the format you need. Your feedback decides what we pick, how much context comes with it and how hard it is.

    An agreed standard for what goes in.

  3. 03

    Pilot

    We agree on what a good result looks like. For environments, that means how each task is scored and what counts as a pass.

    A small pilot, with a way to review and replace anything that falls short.

  4. 04

    Delivery

    We scale up what you accepted, keep taking feedback and add new domains or task types as your research grows.

    Regular deliveries that change as your feedback comes in.

02Evaluation kit4 files

The templates we use to write a brief and check a delivery.

The example manifest is made up. Your sample contains the real material.

03Questions
Can we start small?
Yes. We start with a small sample so your team can judge the fit and shape the pilot. What the sample covers, who can see it and how you can use it depend on the material. We agree on all of it before delivery.
Can you build to our own brief?
Yes. Your domain, task types, formats, context and evaluation needs shape the work. We can adapt one of our products or source and prepare new material for your brief.
What comes with the data?
We agree the package with you. It can include source files, rendered pages, extracted text, task metadata, records of where the material came from and how we prepared it, review notes and a release manifest. Graded environments also come with the environment itself and proof that the grading works.
Can you meet our sourcing and licensing requirements?
Yes. We document where the material comes from and how we prepared it, and we work through the licensing terms for your use. Rights for training, evaluation, exclusivity and distribution are agreed per project.
Can documents keep their links to other company records?
Yes, where the source and permissions allow it. We agree up front how records link together. Some products keep names and identities local to each document. Others keep them consistent across records so you can follow a workflow. Each needs different preparation.
How do pricing and delivery work?
We scope the material, the review work and the delivery schedule before we quote. A pilot sets the standard for a larger project. Price depends on source access, preparation, annotation, exclusivity and review effort.