Licence

Everything served here is CC0 or a work of the United States Government.

The short version. Take it, use it commercially, redistribute it. You owe nothing — no share-alike, no non-commercial term, and no attribution obligation anywhere in the critical path. That is a property of the sources, and it was a selection criterion.

The data

Source	What it gives	Licence	Key?
Wikidata (SPARQL)	2,782 varieties: colour, species, origin, parentage, cross-references; 3,381 co-occurrence claims	CC0 1.0	none
eCFR, 27 CFR part 4	US labelling thresholds, quoted with citations	Public domain (17 U.S.C. 105)	none
VIVC — identifiers only	A cross-reference for 1,276 varieties	n/a — the identifier, never the content	none

Wikidata's own licensing page is explicit: "All structured data in the main, property and lexeme namespaces is made available under the Creative Commons CC0 License." Only page text is CC BY-SA, and none of it is used here. Attribution is therefore not required; it is given anyway, because naming your sources is correct regardless.

What was excluded, and why

Served live at /v1/sources, so it cannot rot in a document nobody reads.

Source	Why not
eAmbrosia (EU GI register)	Three independent blockers. Its licence is the European Commission reuse notice (Decision 2011/833/EU), a source-acknowledgement obligation that fails the public-domain bar. data.europa.eu lists exactly one distribution and its format is HTML; every candidate API path returns the application shell. And the specifications underneath are per-appellation PDFs in 24 languages.
Blending rules from Wikidata	No such property exists. Exactly one property in all of Wikidata mentions grapes and it is an identifier.
EUR-Lex (EU wine regulations)	Same Commission reuse notice as eAmbrosia.
OIV	Terms forbid "any other use than those specified above, including commercial use".
VIVC content	No reuse grant; written consent required. Identifiers only.
DBpedia	CC BY-SA 3.0 plus GFDL — share-alike reaches the whole combined work. Wikidata covers the same ground under CC0.

The code

MIT. If a plan's limits are in your way, run your own copy — or query Wikidata's SPARQL endpoint directly, which is free, public, and more capable than this service in every dimension except normalisation and retained history.

Derived fields

The normalised colour, the confidence score, the varietyId slug and the co-occurrence labelling are computed by this service, not published upstream. Each is marked derived in the response and the raw values ship alongside so the working can be checked. Those derivations are MIT with the rest of the code; the underlying facts remain CC0.