Improving Discoverability, Usability and Governance of Priority Agency Data
Overview
To improve the discoverability, usability, and governance of these priority agency data assets agencies shall apply the documenting techniques illustrated below.
Source
Category
Keywords
Details
This page documents metadata priorities for federal agency data assets. It describes specific keyword and field requirements tied to federal policy directives, and shows how those requirements map to DCAT-US v3.0 fields.
Note on DCAT-US version: The original version of this page was written for DCAT-US v1.1. Field guidance has been updated to reflect DCAT-US v3.0. Where fields have changed between versions, both the original v1.1 field and the v3.0 equivalent are noted.
COVID-19 Response Related Data Assets
Policy background
In April 2020, OMB Memorandum M-20-16 directed agencies to prioritize COVID-19 response data as their highest priority data asset during the public health emergency.
Current status: The federal public health emergency for COVID-19 ended in May 2023. M-20-16 is no longer actively enforced as a current data prioritization requirement. Agencies that documented COVID-19 datasets under M-20-16 do not need to remove existing keywords – they remain useful for discoverability – but there is no active policy mandate requiring new COVID-19 prioritization.
What to do with your metadata
The keyword field is the only field specifically required under M-20-16. Include COVID-19 and coronavirus as keywords on any dataset related to COVID-19 response. Additional descriptive keywords are encouraged – for example viral-testing, CARES-Act, or SARS-CoV-2.
The keyword field works the same way in DCAT-US v3.0 as it did in v1.1. No changes are needed to your existing COVID-19 keyword values.
Data Assets to Fuel AI Research and Development
Policy background
In February 2019, Executive Order 13859 directed agencies to improve data inventory documentation to enable discovery and usability of federal data assets for AI research, and to tag datasets suitable for AI research with standardized keywords.
Current status: EO 13859 remains on the books. The AI policy landscape has evolved significantly since 2019 – EO 14179 is the current governing AI executive order, issued January 2025. The usg-artificial-intelligence keyword convention established under EO 13859 remains a useful and recognized practice. Agencies should continue applying it and consult current OMB guidance for any updated requirements under EO 14179.
What to do with your metadata
Keywords are the primary requirement under EO 13859. Include usg-artificial-intelligence on any dataset suitable for AI research. If the dataset is intended as training data, also include usg-ai-training-data. Additional descriptive keywords are encouraged – for example machine-learning, natural-language-processing, or AI. The keyword field works the same way in v3.0 as it did in v1.1.
Contact information is also required for AI datasets. Include a contact who can discuss restrictions or controls on the dataset with AI researchers. In v1.1 this was expressed using a vCard-style format. In v3.0 contactPoint uses a Kind object – provide at minimum a fn (full name or team name) and a hasEmail value formatted with a mailto: prefix. You can optionally add a position title and organization name. See the Dataset fields page for the full contactPoint format.
Data quality was required under EO 13859 to indicate whether a dataset meets the agency’s Information Quality Guidelines. In v1.1 this was expressed using the dataQuality boolean field (set to true or false). That field is not in the v3.0 schema. In v3.0 use hasQualityMeasurement instead – this is a structured object that captures the quality metric, the measured value, and the unit of measure. See the QualityMeasurement class technical reference for field details.
References and documentation were required under EO 13859 if publications, model documentation, or other references exist for the dataset. In v1.1 this was expressed using the references field (an array of URLs). That field is not in the v3.0 schema. In v3.0 use one of two replacements depending on the type of link. Use isReferencedBy for publications or papers that cite or use the dataset – this is a simple array of URL strings. Use page for documentation, model cards, or technical references about the dataset – this is an array of Document objects, each with a title and an accessURL or downloadURL. See the Dataset fields page for details on both fields.
Summary of field changes for priority datasets
The following v1.1 fields used in priority dataset documentation have changed in v3.0.
keyword – No change. Same field, same format in both versions.
contactPoint – Same field name. Format updated in v3.0 – use a Kind object with fn and hasEmail. Can now be an array for multiple contacts.
dataQuality – Replaced in v3.0 by hasQualityMeasurement. Use a structured QualityMeasurement object instead of a boolean value.
references – Replaced in v3.0 by two more specific fields. Use isReferencedBy for things that cite the dataset, and page for documentation about the dataset.