Where did it come from?
Every dataset should answer that question first. Ours carry a complete record, from the agreement that allowed recording to the license it shipped under.
Eight links, recorded for every dataset.
Source
Who holds the data, and their authority to share it.
Collection
Rigs, instruments, sampling rules and procedures.
Consent and rights
Signed opt-ins, payment and ownership terms.
Transformations
Every processing step, logged.
Anonymization
How personal data was removed, and the checks that confirmed it.
Annotation
Who labeled what, under which schema and rubric.
Quality
Gold-task calibration, spot audits and known limitations.
Delivery
What shipped, to whom, under which license.
Source
Who holds the data, and their authority to share it.
Collection
Rigs, instruments, sampling rules and procedures.
Consent and rights
Signed opt-ins, payment and ownership terms.
Transformations
Every processing step, logged.
Anonymization
How personal data was removed, and the checks that confirmed it.
Annotation
Who labeled what, under which schema and rubric.
Quality
Gold-task calibration, spot audits and known limitations.
Delivery
What shipped, to whom, under which license.
A signed record ships with every dataset.
- The chain record
- The consent basis
- The de-identification report
- Quality measures
- Permitted-use terms
Everyone recorded agreed to it.
Every worker, clinician and expert opts in, is paid, and can withdraw. Nothing is recorded without a signed agreement.
Scraped data cannot answer the first question.