For labs
Environments with a past.
Digital twins of the industry software real businesses ran, seeded with years of de-identified history, with tasks drawn from what actually happened and verifiers that check against the outside world. And the corpora underneath them, licensed properly.
What is in the box
Four deliverables from one source, and the paper that travels with them.
- Read more
01
Corpora
One business, every tool, outcomes attached, entity-consistent across sources.
- Read more
02
Expert trajectories
How the work was done, step by step, inside the software.
- Read more
03
RL environments
Digital twins of the industry software, seeded with real history, with verifiers.
- Read more
04
Evals
Held-out slices scored with the same verifiers, refreshed as records arrive.
- Read more
05
Provenance
Licence chain, method document, exclusions annex, watermark, jurisdiction.
Get data
Start with a sample.
Tell us the vertical, the software, and whether you want corpora, environments or evals. We return the current inventory and a sample under NDA.
- A twenty-point sample from the corpus you are evaluating, de-identified to the same standard as the delivery, under NDA. Two weeks from request.
- Per corpus, per environment, per eval slice, or the whole inventory in a vertical. Exclusive or shared, priced accordingly.
- Structured exports per system with schema and join map; environments as containers with an API or MCP surface and a rendered UI; evals as task suites with graders and baselines.
- Training and evaluation use under licence. No resale, no re-identification, no redistribution. Retraction and refresh on a cadence set in the agreement.
Questions
The ones a data team and its counsel ask in the first call.
Those vendors produce data about work by asking experts to write it down. We license the record of the work itself, from the business that did it, and build the environment around that record. The experts in our trajectories are the staff who actually ran the software.
No. They are rebuilt from the data model and the workflows the records exhibit, so a task that required a particular screen, field or state transition can be reproduced. No vendor binaries or proprietary assets are redistributed.
Yes. The twin exposes the full state, so a lab can author tasks and graders against it. We also build task families to a specification and hold out eval slices for them.
Clinic and revenue-cycle corpora contain protected health information at source. They go through HIPAA Expert Determination rather than Safe Harbor, so date intervals survive, and they are delivered only to labs in the United States and allied jurisdictions.