Dataset license summary
Draft of 2026-10-15. This is a plain-language summary; the full license text ships with each dataset and governs if the two differ. Text in [square brackets] is for counsel to complete.
Paid datasets
You may
- use the dataset inside your organisation, including for commercial products;
- test, evaluate, train and fine-tune models and software with it;
- modify it and combine it with your own data;
- quote short excerpts in papers, documentation and bug reports, naming CohortLoom as the source.
You may not
- resell, publish or redistribute the dataset, or a substantial part of it, as a dataset;
- present it as real patient data, de-identified data, or clinically validated data;
- use it to make decisions about the care of real people.
The free sample
The sample may be used to evaluate CohortLoom and for non-commercial research, and shared with credit to CohortLoom. [Whether the sample uses a standard open license is to be decided before launch.]
What the data is
- Synthetic by construction: simulated patients, with documents written by language models from the simulated charts. No real patient data and no PHI.
- Codes: diagnoses use ICD-10-CM, a public-domain code set published by the US government.
- Provenance: each dataset includes a provenance record and a checksum manifest. The writing models are never trained on Claude or Gemini outputs.
No warranty
Datasets are for testing, development and research, and are provided "as is". Machine-written documents can contain clinical errors; that's expected in test data and is why each file ships with an answer key. [Warranty and liability wording to be set by counsel.]
Questions
Write to [email protected].