Licensing, provenance and quality — before listing.
Buyers need to know they can use what they get. These are the standards a dataset must meet before it is listed, and what we show when it does not yet meet them.
Rights you must document
- You own the data or hold a license that permits redistribution
- The redistribution scope: single records, paid access, bulk
- Required attribution text
- Any third-party terms embedded in the data
- Restrictions such as no model training or no resale
- A named contact for rights questions
How rights map to what buyers can do
Every dataset's license carries one redistribution value. It is enforced in code: downloads and private connections are refused unless rights are confirmed.
| Value | Meaning | Effect on access |
|---|---|---|
| public | Free to redistribute with attribution | Single records served; bulk channels possible once confirmed |
| paid | Redistribution under paid access | Served to entitled plans; bulk only under agreement |
| restricted | Third-party or contractual limits | Records not served; no download |
| unknown | Rights not yet established | No download until clarified |
Fields every record must carry
| Field | Why it matters |
|---|---|
| Provider record id | So users can cite and re-fetch the exact record. |
| Upstream source | Where the provider got it (for example a court or agency). |
| Source / canonical URL | So users can verify against the original. |
| License id | Which terms govern this specific record. |
| Retrieved / effective dates | So users know how current it is. |
| Derived flag + generator | So AI or editorial output is never mistaken for source. |
Quality standards
Verified counts or none
Record counts and date ranges are published only after verification. Unverified values are shown as “Not yet verified”.
Known limitations
Every listing includes quality notes: gaps, duplicates, OCR issues, coverage bias.
Honest access status
Each access method is available, awaiting credentials, awaiting rights confirmation, or not offered.
Update cadence
A stated update frequency, and change notices when the schema changes.
Separated layers
Source text and facts kept apart from AI or editorial enrichment, each with its own provenance and license id.
Stable ids
Record ids that do not change between updates, or a mapping when they must.
Integration requirements
For API and connector publishing. File publishing has its own delivery checklist.
- A dedicated, read-only service credential for Data.co
- Single-record fetch by stable id
- Search with filters and deterministic pagination, page size up to 100
- No endpoint that enumerates or exports the whole corpus to Data.co
- Per-token rate limits you are comfortable with
- Request logging keyed by the X-Request-Id Data.co sends, for reconciliation
- Clear error responses for not found, unauthorized and rate limited
What we will not list
Scraped data without rights, personal data without a lawful basis, data whose license forbids redistribution, and synthetic data presented as real. Demo fixtures used for previews are always labeled as demo and are never billed.
Why the bar is set here
Buyers increasingly put marketplace data in front of AI systems that quote it to end users. If a record's license, origin or derived status is unclear, the buyer cannot tell whether it may be shown, cited or stored. Record-level provenance and explicit rights are what make a dataset usable in that setting, which makes them commercial requirements as much as legal ones.
Questions buyers ask
Can a dataset be listed before every access method is ready?
Yes, with each method's real status shown. That is how LAW.co was listed before its service credential arrived, and how its statute reference table and bulk downloads are listed today.
Do you audit provider data?
We review samples, schema and provenance during onboarding and spot-check after listing. We do not certify accuracy; providers state their own quality notes.
What if my data includes publisher headnotes or other third-party material?
That material is excluded from what Data.co serves until its rights are cleared. LAW.co's connector, for example, drops reporter head matter whose rights are unconfirmed.
Do I need to provide record counts?
Not to list. Counts are shown only once verified; until then the listing says not yet verified, which buyers understand better than an estimate.
Can I restrict model training on my data?
Yes. Restrictions such as no training or no bulk extraction are part of the license and are shown to buyers and returned as access restrictions with records.
What happens if my rights change after listing?
Tell us. The affected channels are switched to the right status, bulk deliveries stop, and buyers see the change on the dataset page.
Do you accept data with personal information?
Only with a documented lawful basis and license terms that allow it, and only the fields needed. Case law naming parties and judges is assessed as part of the rights review.
List your data on Data.co.
We review every provider for licensing, provenance and quality. You keep control of terms and pricing.