Dataset

The foundational properties that uniquely identify and describe the dataset to both users and machines.

• name: The canonical title of the dataset (e.g., "Global Surface Temperature Anomalies").
• description: A detailed, plain-text summary of the dataset's contents, collection methodology, and purpose.
• identifier: A unique code, such as a Digital Object Identifier (DOI), used for persistent citation.
• url: The landing page where a user can access the dataset.

Properties that define the dataset's scope in time and geography, critical for filtering in data catalogs.

• temporalCoverage: The time period the data represents, expressed as an ISO 8601 interval (e.g., "1880-01-01/2023-12-31").
• spatialCoverage: The geographic region the data covers, defined using a Place type with a geo property or a named administrative area.

The distribution property links to one or more DataDownload nodes, specifying how to obtain the raw data.

• contentUrl: The direct download link for the file.
• encodingFormat: The MIME type of the file (e.g., "text/csv", "application/zip").
• contentSize: The file size in bytes, helping users assess download feasibility.

Properties that establish the dataset's authority and lineage, directly supporting algorithmic trust signals.

• creator: The Organization or Person that originally produced the data.
• publisher: The entity responsible for making the data available, which may differ from the creator.
• isBasedOn: Links to a source dataset if this is a derivative work, establishing a clear lineage.
• citation: A scholarly reference format for crediting the dataset in academic work.

Properties that describe the internal schema of a tabular dataset, making it machine-readable.

• variableMeasured: An array of PropertyValue nodes, each defining a column's name, expected data type, and unit of measurement.
• includedInDataCatalog: A direct link to the DataCatalog that indexes this dataset, establishing its containment hierarchy.

Properties that communicate the legal and technical boundaries of data usage.

• license: A URL to the specific legal license governing reuse (e.g., Creative Commons BY 4.0).
• conditionsOfAccess: Textual description of non-license requirements, such as "Access requires institutional login."
• measurementTechnique: The instrument or methodology used to collect the data, providing critical context for assessing fitness-for-purpose.

The Dataset type enables explicit provenance tracking through properties like creator, publisher, isBasedOn, and sameAs. This creates a verifiable chain of custody for data used in AI training. When a language model cites a dataset, the structured metadata provides the canonical source, version, and license, enabling citation integrity scoring and reducing the risk of models training on misattributed or deprecated data.

Use temporalCoverage and spatialCoverage properties to define the precise scope of your data. This is critical for geospatial and time-series datasets where relevance depends on location and period. For example:

• temporalCoverage: "2020-01-01/2023-12-31"
• spatialCoverage: A Place or GeoShape entity This allows search engines to filter datasets by geographic region and time range, dramatically improving precision for domain-specific queries.

The distribution property, using the DataDownload type, allows you to specify multiple access points and file formats for the same dataset. You can define:

• contentUrl: Direct download link
• encodingFormat: "CSV", "Parquet", "JSON"
• contentSize: File size in bytes This enables crawlers to understand data accessibility and format availability without human interpretation, supporting automated data pipeline integration.

The license property provides a machine-readable reference to the legal terms governing data usage. By linking to a canonical license URI (e.g., a Creative Commons deed or a custom license page), you explicitly define permissible use cases for AI training and redistribution. This is a critical signal for data governance and compliance automation, allowing crawlers to filter datasets by license type before ingestion.

The sameAs property on a Dataset entity links it to external canonical identifiers, such as a DOI (Digital Object Identifier), a Wikidata entry, or a repository record on Zenodo or Figshare. This performs explicit entity reconciliation, consolidating authority signals and ensuring that search engines treat all references to the same dataset as a single, unified entity rather than fragmented duplicates.