Experimental data.
Built for your models.

License unpublished assay results, instrument data and experimental records for pre-training, post-training and evaluation. We source and prepare the data around your requirements, with methods, provenance and permitted uses documented.

Define the question.
Specify the data.

Start with the scientific task and the gap in your existing data. Together, we define what to source, how to assess it and how you can use it.

Scientific task

Define the question, the measurements that answer it and how success will be measured.

Data specification

Specify assay and data types, volume, coverage, controls and quality requirements.

Permitted use

Agree training or evaluation use, recipients, exclusivity and retention.

Data for each stage
of model development.

License existing lab records, source what is missing, and test performance on scientific tasks. Lab environments for agents are in development.

Which offering supports each stage of model development
OfferingPre-trainingPost-trainingEvaluation
Licensed datasetsUnpublished lab records
Custom sourcingExisting records or new experiments
Evaluation setsHeld-out scientific tasks and scoring
Environments SoonSimulated lab workflows

Uses depend on the agreed specification and licence. Offerings can be combined.

Licensed datasets

Experimental records.
Prepared for training.

License assay results, instrument files and experimental records from authorised laboratories and universities. Positive, negative and inconclusive outcomes stay linked to their methods, samples and controls. We de-identify, deduplicate and quality-check each dataset against a specification that sets its permitted use, labels and acceptance criteria.

Explore data types
Dataset documentation
Scattered source records becoming the ordered rows of a prepared dataset
Source
Laboratory, instruments, period, coverage
Rights
Permitted uses, exclusivity, retention
Preparation
De-identification, method linking, QC checks
Delivery
Native files, release version, agreed updates

Each release includes dataset documentation and the agreed licence terms.

Custom sourcing

Close the gaps
in your training data.

Commission data for a specific assay, target class or scientific task. We identify authorised laboratories and arrange exports of existing records or, where an archive cannot answer the question, a targeted experimental programme with qualified CROs and labs. The brief sets methods, controls and repeats, licensing permissions, coverage, acceptance criteria and delivery terms.

Discuss custom sourcing
Sourcing brief
A dataset with a precise gap, and a sourced piece of exactly that shape
  1. BriefScientific question, coverage and acceptance criteria
  2. LaboratoriesExisting records or new experiments
  3. DeliveryPrepared, documented dataset

The sourcing brief defines what to collect and how the result will be assessed.

Evaluation sets

Know what your model
can actually do.

Build held-out evaluation sets from real lab work, with task inputs, reference outcomes or scoring rules reviewed by scientific experts, and documented splits. We check for contamination against the training sources available for review and record the limits of those checks, with versioned releases for comparing models.

Discuss an evaluation
Three questions per task
Training data on the left; on the right, held-out tasks kept apart, each with its reference outcome
Evidence
Does the answer trace back to the experimental record?
Reasoning
Does the interpretation account for methods and controls?
Limits
Does it distinguish what is known from what remains uncertain?

Task coverage, scoring methods and contamination checks are defined with your team.

Environments Soon

Real lab workflows.
Runnable tasks.

In development: replay tasks and simulated lab workflows for training and evaluating agents, with scientific constraints and traceable outcomes. Licensed records will provide the starting context; instrument and lab-software interfaces, protocols and verifiers will define what the agent can do and how success is measured.

Discuss planned environments
Define before the agent runs
A path through a task: a starting state, two tool actions and a checked end state
  1. Start stateSeeded with licensed lab records
  2. ToolsInstrument and lab-software interfaces
  3. VerifierVerifiable end states and trajectory logging

Simulation fidelity, validation and access are agreed per programme.

Built around
scientific tasks.

Match the data to the capability you want to train or evaluate.

Documented rights.
Controlled access.

Define the source, permitted use and handling of the data before delivery.

Source and provenance

Record the data’s origin, laboratory and instruments, and verify the provider’s authority to license it.

De-identification

Agree the identifiers to remove or replace, such as patient, compound and sample codes, the records to exclude and the checks required before lab review.

Permitted uses

Specify training and evaluation rights, recipients, exclusivity and retention in the agreement.

Release approval

After the agreed checks, the laboratory’s authorised representative approves the records, recipients and access conditions for each sample or release.

What does your
model need next?

Share the scientific task, data requirements and intended use.

Discuss your requirements
What is available, and when can we receive it?

The proposal identifies available records, additional sourcing or experiments, volume, assay and sample coverage, date range, preparation work, delivery format and timing.

Can you source both existing and newly generated data?

Yes. We license existing laboratory records and can commission new experiments with qualified CROs and labs. Feasibility, rights, cost and timing depend on the project.

Can we see data before we commit?

Subject to laboratory approval, we can arrange access to a prepared sample before a full-dataset commitment. It includes records, field definitions, methods, source and date coverage, preparation notes and known limitations. Its size, selection, assessment rights and any fee are agreed before access.

Who may use a licensed dataset?

The licence identifies the permitted recipients, uses and access locations. Access by group companies or contractors, and any onward sharing, must be expressly permitted.

What does an evaluation delivery include?

The scope identifies task inputs, reference outcomes or scoring rules, dataset and scoring versions, split boundaries and the limits of overlap checks. Evaluation execution and reporting are included only if stated in the scope.