JSON-LD

JSON-LD expresses linked data in JSON while using a context to map terms to identifiers. It is widely used for web-page structured data and can be processed as RDF. The context is part of the interpretation: the same short property name can mean different things under different mappings. Keeping that context makes the data more portable than treating property strings as self-explanatory.

Valid JSON-LD does not establish that a publisher's assertions are true or that a consumer applies the intended vocabulary correctly. Remote contexts and external identifiers can change or become unavailable. Retain the original document, its context, retrieval date, and any normalized representation used for analysis. A parser's successful expansion is a syntax check, not a factual audit.

A fictional API returns an object with a key called status. Under the publisher's context, status maps to a property in a drug-regulatory vocabulary, and the value is a string. In a revised version of the API, the publisher changes the context so that status maps to a property in a generic project-management vocabulary. The JSON bytes are unchanged.

A consumer that processes the new context gets different RDF for the same JSON. A consumer that cached the old context keeps the old meaning. A reviewer reconstructing a past decision needs the document, the context in force when the document was received and the processing date. The record therefore stores the received bytes, the resolved context as retrieved, a hash of each and the retrieval time. If the context is remote and later disappears, the stored copy is the only evidence of what the keys meant.

JSON-LD is not a schema language. It does not say which properties are required or what values are allowed. Those constraints come from another layer, such as a shapes language or an application profile, and passing a JSON-LD parser is not validation against them.

It is also not a different data model from RDF. It is a syntax for it, with some extra features, and a document that uses features with no RDF equivalent loses them on conversion. Finally, remote contexts introduce a dependency on a server. For documents that must remain readable, bundle or pin the contexts they need, and record their hashes.

JSON-LD documents age through their contexts. A context may be edited in place, moved to a new URL, split into versions or removed along with the site that hosted it. Some processors cache contexts aggressively, which can hide a change for a while and then reveal it all at once when the cache expires. Publishers who care about stability version their contexts at distinct URLs and keep old versions available. Consumers who care about reproducibility store the context they used. The specification also has a mode for documents that declare which processing rules they were written for, and documents should state it when it matters.