Provenance / Chain of custody
Every dataset should answer one question first.
Where did it come from? Scraped data can't answer. Ours can: every session is documented from source to delivery, and the record ships with the data. Verifiable, auditable, complete.
Eight links · Every dataset
Source verification
The facility, cohort, or expert panel is identified and its authority to share confirmed.
Collection method
Rig placement, instrumentation, sampling rules, and operator procedures on record.
Consent & rights
Signed opt-ins, compensation terms, and ownership agreements for every participant.
Transformations
Every processing step logged: format shifts, filtering, resampling, corrections.
Anonymization
PII/PHI removal method and verification results, including k-anonymity checks.
Annotation history
Who labeled what, under which schema and rubric version, with agreement scores.
Quality checks
Gold-task calibration, spot audits, and known-limitation notes per batch.
Delivery records
What shipped, to whom, under which license: the closing link of the chain.
Context is data
Data without context is incomplete.
Metadata is designed around the model and task the data will support: forty-plus fields per session, from device context to known limitations.
Certified delivery
A certificate with every package.
Each delivery includes a signed provenance certificate: the chain record, consent basis, de-identification report, quality metrics, and permitted-use terms. Your legal and safety teams get answers, not assurances.
