Data sources

Every dataset Open Label reads, how current it is, and what it leaves out.

All of it is public, and none of it needs an account.

Source	Used by	Current to	Notes
DailyMed SPL	Check, Watch, search, diff, Skills	Live	Current labels by XML, past versions from the zipped archive
Medicare Part D Prescribers by Provider and Drug	Unseen	2024 (published May 2026)	Medicare patients only, pharmacy-dispensed drugs only
Open Payments general payments	Unseen	2023, 2024, 2025	Payments below a small yearly threshold need not be reported
NPPES NPI Registry	Unseen	Live	Practice address and credential
ClinicalTrials.gov	Unseen	Live	No NPIs; matched by name and state

Pinned datasets

CMS publishes each year of Part D and Open Payments as its own dataset with its own id. Those ids are pinned in packages/unseen/src/partd.ts and packages/unseen/src/payments.ts, and were checked against the live APIs: the Part D file name ends in DY24, data year 2024, and each Open Payments id returned rows for its own program year.

When CMS publishes the next year of either, add its id there.

What the model sees

Check sends the model the piece and, for each claim, the twelve label sentences retrieval chose. Unseen sends the model only your question. Neither sends patient data, because none of these sources contain any.