Does structural metadata tell us whether every number is true?

No. It helps organize and validate representation; accuracy needs additional evidence.

Is every possible dimension combination required?

Not necessarily. Reporting requirements and constraints determine which combinations must be supplied.

Can reference metadata themselves have structural metadata?

Yes. Their reports need defined attributes, relationships and attachment to the objects they describe.

AI Glossary / Data engineering and architecture / Advanced

Structural metadata

Definition

Structural metadata identify and describe how data and reference metadata are organized so that systems can locate, interpret and process them. In statistical exchange, they can define dimensions, measures, attributes, representations and related constraints. They establish the structure in which values have a recognizable role. This differs from reference metadata that explain matters such as methodology and quality, although both are needed for reliable use.

What it means

A sequence of numbers is difficult to interpret without knowing which value belongs to which measure, place or period. Structural metadata provide that organizing information. In an original AI learning report, the value 60 might represent participants, seconds or a percentage. A defined measure and unit help distinguish these possibilities, while dimension values identify the relevant course and month. This does not explain every limitation of the collection process. The structure gives the values their organized position; methodological explanations are still needed to assess how those values were produced and what comparisons they support.

How it works

SDMX describes structural metadata as necessary for identifying and processing data and reference metadata. Its framework organizes statistical observations through defined concepts and their roles. A data structure definition can associate dimensions, measures and attributes with their representations. In an original catalog report, a course dimension might use a maintained list of course codes, and a period dimension might use a specified monthly representation. Systems use these definitions to recognize allowed content and relationships. The exact structural model depends on the implementation, and successful structural validation does not establish that every observation is factually correct.

Three practical examples

A learning dashboard uses course and month dimensions to identify which observation each count describes.
An AI import process checks incoming codes against the expected representation before combining them with existing data.
A metadata exchange uses a defined reporting structure so explanatory notes can be attached to the correct target objects.

Think of it this way

Imagine a theater seating plan with sections, rows and seat numbers. A ticket marked B, 4, 12 becomes meaningful because the plan defines those positions. Without the plan, the symbols are difficult to interpret consistently. Structural metadata play a similar organizing role for data values. The seating plan does not tell you whether a person actually attended; that is a separate factual question about the event.

Worked example

Consider an original fictional monthly dataset with two dimensions, CourseCode and Month, and one measure, ParticipantCount. The permitted course codes are C1 and C2, and the periods in scope are January and February of a specified year. There are 2 × 2 = 4 possible course-month combinations. Suppose the dataset reports C1-January as 60, C1-February as 75 and C2-January as 40, leaving C2-February unreported. Three combinations are populated. The structural definition establishes where the fourth observation would belong, but its absence does not mean its value is zero. A record using course code X9 fails the exercise's allowed-code rule. A record using C1-January with value 600 might satisfy all the same structural rules if non-negative integers are permitted, even if the true count should be 60. This separates recognizable structure from factual accuracy. Whether all four combinations are required must be specified separately; a possible combination is not automatically a mandatory report.

Illustrative case study

In a fictional AI learning dashboard, an import process reads monthly counts but treats an unreported combination as zero. Reviewers document the dimensions, code list and reporting rules, then test the four-combination example. They distinguish missing observations from known zero counts and reject an unexpected course code for investigation. They also check whether a large but structurally valid count reflects a real event or a data error. The assistant's summaries are revised to preserve the difference between missing and zero. The scenario illustrates using structure to make data interpretable without assuming that every validly shaped record is true or that every possible coordinate must be populated.

Common misconceptions and mistakes

Structural metadata are more than a visual table layout. They can express roles, representations and constraints needed for machine processing. Another mistake is assuming that a missing cell is automatically zero because its location is known. A known coordinate does not establish an observation. Structural validity also does not prove accuracy: an allowed number can still be wrong. Finally, reference metadata have structure too. A report describing methodology may require an identified target and organized attributes, even though its content is explanatory prose rather than a numeric observation.

Business use

For an AI reporting pipeline, define the measures and dimensions before importing or comparing values. In the fictional course dataset, establish which course codes are valid, how months are represented and what reporting completeness means. Test unexpected codes, duplicate coordinates and missing combinations using the actual agreed rules. Keep methodological and quality explanations available alongside those structural definitions. When codes or structures change, inspect downstream mappings and queries. This helps a team distinguish a parsing or classification problem from a missing observation or an inaccurate count, so each issue receives the appropriate remedy.

Check your understanding

How many course-month combinations are possible in the toy scope?

Answer: Four.

How many are reported?

Answer: Three.

Does the missing C2-February observation mean zero participants?

Answer: No.

Why is X9 rejected?

Answer: It is outside the exercise's allowed course codes.

Could 600 pass a structural check while being inaccurate?

Answer: Yes.

What additional rule is needed before labeling the fourth combination an omission?

Answer: A requirement that the combination must be reported.

Related terms

Reference metadata
Data structure definition
dimension
measure
code list
structural validation

Learn next

Define a small dataset with two dimensions and one measure. List its possible coordinates, then leave one observation missing and include one invalid code. Explain which checks depend on structural metadata and which require evidence about the real-world values. State whether completeness is mandatory for every possible coordinate.

Sources

SDMX, Glossary Version 2.1, December 2020, Structural metadata and Structural validation: https://sdmx.org/wp-content/uploads/SDMX_Glossary_Version_2_1_December_2020.htm SDMX, Framework for SDMX Technical Standards, version 2.1: https://docs.sdmx.org/en/i1-doc/Sections/Section1/SDMX_2-1_SECTION_1_Framework.html The course-month structure, counts, seating-plan analogy and AI dashboard scenario are original educational illustrations. The example's code list and reporting choices are local assumptions, not universal requirements of SDMX.

Previous: Structural equation modeling Next: Structural model