Compliance first: We never take HIPAA/PHI data or personal customer data. Everything is de-identified before it leaves your control.
Home / Guidelines / Open data policy guideline 28
Open data policy guideline 28

Data Quality: Open Data Policy Guideline 28

Data Quality: Open Data Policy Guideline 28

Guideline 28 treats data release as an ongoing, iterative process: keep published data accurate and current, document known problems, avoid parallel data streams, and give every dataset clear metadata and a contact.

Publishing data exposes its flaws. This guideline asks governments to treat that exposure as a feedback loop rather than a reason to wait for perfection.

What the guideline means

Release alone does not produce good data. Guideline 28 asks open data policies to build processes that keep published data accurate, current and accessible. Release should be iterative: once sensitive information is addressed, publish, then improve and update as the data grows.

Where quality problems are serious, they should be documented so users are not misled. The guideline warns against separate public reporting streams that diverge from the data government actually uses, since parallel versions tend to drift. Each update should carry complete metadata (Guideline 13), including a visible contact, and respond to issues raised through a prominent feedback channel.

Model policy language patterns

Paraphrased quality provisions:

  • Departments shall review published datasets regularly for accuracy and completeness.
  • Each dataset shall name a responsible contact and include documentation of known limitations.
  • Data shall be published from authoritative source systems rather than separately maintained copies.
  • The oversight body shall report on data quality and corrective actions in its annual report.

Where it appears in adopted policies

Quality language appears in many policies in this index, including:

Annual reports from New York State, Chicago and Oakland during this period are useful case studies and are linked below.

Common pitfalls

  • Waiting for perfect data. Holding datasets back until they are flawless means they are never released.
  • Undocumented flaws. Publishing known-bad data without caveats spreads misinformation.
  • Shadow datasets. A "public version" maintained by hand diverges from the operational system.
  • Anonymous datasets. Without a named contact, users cannot report errors and publishers never learn about them.
  • Quality reviews that never reach users. If corrections are made silently, people who already downloaded the data keep using the flawed version; a change log fixes this.

A practical checklist for quality provisions

  • Publish from authoritative source systems, not hand-maintained copies.
  • Name a contact for every dataset.
  • Document known limitations, gaps and changes in methodology.
  • Review high-use datasets on a schedule for accuracy and completeness.
  • Route public error reports to data owners and track resolution.
  • Group related datasets and keep definitions consistent across them.
  • Report recurring quality issues and fixes in the annual open data report.

Why this matters for your business data

Quality is where buyers discount or reward a dataset. AI and machine-learning teams test extracts for completeness, consistency, duplicates and drift before they pay. Data pulled directly from the system a business actually runs on, with a short note on known gaps, is usually worth more than a polished spreadsheet someone assembled by hand.

That is the private-sector version of Guideline 28: publish from the source, document the limits, fix issues iteratively. How to value my data lists the quality signals buyers check, and the valuation calculator factors them into an estimate.

We never take HIPAA/PHI data or personal customer data. Everything is de-identified before it leaves your control.

Our compliance promise: We never take HIPAA/PHI data or personal customer data. Everything is de-identified before it leaves your control.

Frequently asked questions

Should governments wait until data is clean before publishing?

The guideline recommends against waiting for perfection. Once sensitive information is addressed, publish, document known issues, and improve iteratively. Public use often surfaces errors faster than internal review.

What is a parallel data stream and why is it a problem?

It is a separate copy of data maintained specifically for publication rather than drawn from the system government actually uses. Such copies drift out of sync, creating conflicting numbers. Publishing from authoritative sources avoids this.

What metadata supports data quality?

At minimum: a description, field definitions, update frequency, the date of last update, known limitations, and a named contact. These let users judge fitness for purpose and report problems.

Policies in the index that reflect guideline 28

PlaceAdoptedInstrumentGuidelines
Chicago, ILDec 10, 2012Executive order34678101113
South Bend, INAug 22, 2013Executive order237810182528
Louisville, KYOct 11, 2013Executive order (No. 1, Series 2013)1234781018
Oakland, CAOct 15, 2013Council resolution (No. 84659)2345781011
Honolulu, HINov 27, 2013Ordinance237810111820
Pittsburgh, PAMar 19, 2014Ordinance12345678
Hartford, CTMar 26, 2014Executive order34789182224
Minneapolis, MNAug 1, 2014Council-adopted policy34568131819
Houston, TXSep 18, 2014Administrative policy (AP 8-7)234567810
Suffolk County, NYJul 31, 2015Administrative policy78131820222426
Jackson, MSSep 1, 2015Executive order (No. 2015-06)12345678
Cambridge, MASep 21, 2015Ordinance12345678
Asheville, NCOct 13, 2015Council resolution13489101118
Tulsa, OKDec 21, 2015Executive order123458910
Las Vegas, NVApr 12, 2016Administrative policy (City Manager policy CM510)12345678
Saint Paul, MNApr 27, 2016Council resolution12578101113
Milwaukee, WIJul 14, 2016Council resolution12356789
Bellevue, WAAug 1, 2016Management policy15678101118
Naperville, ILNov 15, 2016Council resolution45789101112
Fort Collins, COFeb 7, 2017Council resolution12345678

Sources

  1. Sunlight Foundation Open Data Policy Guidelines (historical reference)
  2. Open NY One Year Report (2014)
  3. Chicago: Open Data Annual Report (2013)
  4. Oakland: Toward Collaborative Transparency (2014)
  5. Philadelphia: New Opportunities for Data Publishing (2013)

Updated December 31, 2024.