Does a registry have to store every underlying data value?

No. Its role is to register metadata; observations can remain in other systems.

Does registration prove that a definition is approved?

Not by itself. The registry's status and review processes determine what that claim means.

Can two fields with similar names require different metadata items?

Yes, particularly when their meanings, units or derivation rules differ.

AI Glossary / Data engineering and architecture / Advanced

Metadata registry

Definition

A metadata registry is an information system for registering and managing metadata. It provides an organized place to identify descriptive items and record information about their meaning, representation and administration under an applicable model. It is concerned with descriptions of data rather than necessarily storing the underlying observations. Its contents and governance depend on the registry's purpose and implementation.

What it means

Teams need more than familiar field names when they exchange data. A field called duration might describe seconds, rounded minutes or time spent actively viewing a resource. A registry can hold identified descriptions that help distinguish those meanings and support reuse. In an original AI learning platform, an assistant could consult the relevant registered description before generating a calculation. The registry makes the intended meaning available, but the application must still connect the correct description to the actual field. A wrong mapping can produce an incorrect result even when the registry entry itself is clear.

How it works

Metadata items are submitted, identified and managed through the registry's chosen processes. The OECD glossary describes registration in terms of identification, provenance and quality monitoring. NCI's caDSR provides a concrete example of infrastructure supporting common data elements, forms and models using an ISO/IEC 11179-based approach. A local educational registry might record definitions, units, permitted values and review information. Those choices should be documented rather than assumed universal. Search, maintenance and review make the information usable, while explicit links connect descriptions to the datasets and systems that rely on them.

Three practical examples

A data team registers the meaning and unit of a shared duration field before multiple reports reuse it.
An AI assistant resolves an abbreviated column name through an identified metadata description instead of guessing its meaning.
An integration project compares registered definitions from two systems before deciding whether their data can be combined.

Think of it this way

Imagine an organization's reference desk maintaining an identified set of definitions for the forms people use. Staff can find what a field means, who maintains its definition and which version applies. The desk does not hold every completed form or guarantee every answer written on them. A metadata registry plays a similar organizing role for descriptions, while actual data collection and use remain separate activities.

Worked example

Consider an original fictional registry with two identified data-element descriptions. M10 defines elapsed resource duration as a non-negative number of seconds. M11 defines display duration as minutes rounded upward from elapsed seconds. Dataset A maps its duration_seconds field to M10. Dataset B maps its card_minutes field to M11. For one resource, A records 125 and B records 3. The values differ because they represent different measures, not because the resource necessarily has conflicting durations. An assistant that treats B's value as an exact measurement and multiplies it by 60 obtains 180 seconds, which is not the original 125-second observation. The registry has two descriptions even if thousands of resources use them. Mapping both fields to M10 would be incorrect despite the registry containing valid definitions. The example shows how registered meaning and correct mapping work together; it does not imply that rounded values can recover the exact underlying measurement.

Illustrative case study

In a fictional AI reporting service, two teams combine duration columns based only on their labels. A report begins treating display minutes as exact elapsed time. Reviewers use M10 and M11 to document the difference, correct the mappings and retain raw seconds where exact aggregation is needed. They add review information so future changes to the descriptions can be traced. They also test that the assistant retrieves the applicable item rather than the first name match. The scenario illustrates using a registry to support semantic consistency. It does not claim that registration alone repairs old reports or verifies every recorded resource duration.

Common misconceptions and mistakes

A registry is not automatically the database containing all the observations it describes. Nor is it simply a list of column names. Another mistake is assuming that an item is authoritative, approved or current merely because it is registered; the relevant status and governance need to be understood. Matching names are insufficient evidence of equivalent meaning. Even matching definitions can be applied incorrectly if a field is linked to the wrong item. Finally, adopting a registry model does not by itself establish complete interoperability: mappings, representations and operational practices still need to work together.

Business use

For an AI data product, register the descriptions that materially affect interpretation and connect them to actual fields. In the fictional duration service, distinguish raw measurements from rounded displays and make the mapping visible to analysts. Assign responsibility for review and changes, and identify consumers that could be affected by a revised definition. When integrating sources, compare meaning before aggregating values. Evaluate whether people and automated tools can find the correct description reliably. This makes the registry a maintained reference for decisions rather than a catalog that exists separately from the workflows it is meant to support.

Check your understanding

How many descriptions are in the toy registry?

Answer: Two.

Why can 125 and 3 both be appropriate values for the same resource?

Answer: One is elapsed seconds and the other rounded display minutes.

Can multiplying the rounded value by 60 recover the original measurement?

Answer: Not generally.

What error remains possible even with correct registry definitions?

Answer: Mapping a field to the wrong description.

Why should consumers be tracked when definitions change?

Answer: Their calculations may depend on the previous meaning.

Related terms

Metadata item
Metadata object
Semantic interoperability
data element
data dictionary
provenance

Learn next

Create two descriptions for related but different measures, giving each an identifier and definition. Map three fictional source fields to the appropriate items and explain every choice. Include one misleading field name, then show why name matching alone would produce an incorrect mapping.

Sources

OECD, Glossary of Statistical Terms, 2008, Metadata registry: https://www.oecd.org/content/dam/oecd/en/publications/reports/2008/09/oecd-glossary-of-statistical-terms_g1gh9ad7/9789264055087-en.pdf National Cancer Institute, Metadata Services for Cancer Research: https://www.cancer.gov/about-nci/organization/cbiit/vocabulary/metadata The duration definitions, mapping arithmetic, reference-desk analogy and AI reporting scenario are original educational illustrations. The simplified registry is not presented as a complete ISO implementation or a guarantee of data accuracy.

Previous: Metadata object Next: Metadata repository