Revision History

1 Overview

SDMX version 3.0 introduces new features, improvements and changes to the Standard in the following key areas:

Information Model

• Simplification and improvement of the reference metadata model
• Support for microdata
• Support for geospatial data
• Support for code list extension and discriminated union of code lists
• Improvements to structure mapping
• Improvements to code hierarchies for data discovery
• Improvements to constraints

Versioning of Structural Metadata Artefacts

• Adoption of the three-number semantic versioning standard for structural metadata artefacts (https://semver.org/)

REST Web Services Application Programming Interface (API)

• Change to a single ‘structure’ resource for structure queries simplifying the REST API specification by reducing the number of resources to five
• Improvements to data queries
• Improvements to reference metadata queries
• Support for structural metadata maintenance using HTTP PUT, POST and DELETE verbs

SOAP Web Services API

• The SOAP web services API has been deprecated with version 3.0 standardising on REST;

XML, JSON, CSV and EDI Transmission formats

• The SDMX-ML, SDMX-JSON and SDMX-CSV specifications have been extended and modified where needed to support the new features and changes such as reference metadata and microdata
• Obsolete SDMX-ML data message variants including Generic, Compact, Utility and Cross-sectional have been deprecated standardising on Structure Specific Data as the sole XML format for data exchange
• The SDMX-EDI transmission format for structures and data has been deprecated
• The organisation of structures into ‘collections’ in SDMX-ML and SDMX-JSON structure messages has been flattened and simplified
• The option to reference structures in SDMX-ML and SDMX-JSON messages using Agency, ID and Version has been deprecated with URN now exclusively used for all non-local referencing purpose

Breaking Changes

Many of the changes made are ‘breaking’ meaning that, while conversion between versions may be possible in certain circumstances, the 3.0 specification is not directly backwardly compatible with earlier versions of the Standard.

A summary of the main breaking changes is given in chapter 2.

Content of the Document

The remainder of the document provides a summary of the main changes. More detailed information can be found the SDMX 3.0 Technical Specifications, in particular:

• Section 2 – Information Model
• Section 5 – Registry Specification
• Section 6 – Technical Notes
• SDMX-TWG GitHub for the REST API and the XML, JSON and CSV formats

2 Summary of Breaking Changes in 3.0

Version 3.0 introduces breaking changes into the web services API, transmission formats and information model. A summary is given in the table below.

2.1 Web Services API

2.2 Transmission Formats

2.3 Information Model

3 Information Model

3.1 Version 3.0 Information Model

Figure 1 Version 3.0 simplified Information Model UML class diagram with ‘heat map’ illustrating the areas with most change

The schematic above is a simplified UML class diagram of the SDMX 3.0 information model illustrating the major areas of change as a ‘heat map’. Darker colours indicate where new structures have been added in version 3.0 or where structures have been significantly changed.

A number of ancillary structures including organisation schemes, process and reporting taxonomy are unchanged and have not been shown. Similarly, Organisation Scheme Map and Reporting Taxonomy Map have been omitted for simplicity. A schematic of the 2.1 model is given in Appendix A for comparison purposes.

3.2 Key Changes from Version 2.1

New Maintainable Artefacts

• Structure Map
• Representation Map
• Organisation Scheme Map
• Concept Scheme Map
• Category Scheme Map
• Reporting Taxonomy Map
• Value List
• Hierarchy
• Hierarchy Association
• Metadata Constraint
• Data Constraint
• Metadata Provision Agreement
• Metadata Provider Scheme
• Metadataset

New Identifiable Artefacts

• GeoFeatureSetCode
• GeoGridCode
• Metadata Provider

Removed Maintainable Artefacts

• Structure Set – replaced by Structure Map and the four item scheme maps
• Hierarchical Codelist – replaced by Hierarchy and Hierarchy Association
• Constraint – replaced by Data Constraint and Metadata Constraint

Changed Maintainable Artefacts

• Data Structure Definition – support for microdatasets and reference metadata linked to data
• Metadataflow – simplifies exchange of reference metadata, in particular those linked to structures
• Metadata Structure Definition – simplified model for reference metadata
• Codelist – support for codelist extension and geospatial specialised codelists (GeographicCodelist, GeoGridCodelist)
• VTL Mapping Scheme – VTL Concept Mapping Scheme removed to align the VTL / SDMX interface with the 3.0 model

New Component Representation Types

• GeospatialInformation – a string type where the value is an expression defining a set of geographical features using a purpose-designed syntax

3.3 Areas Unchanged from Version 2.1

The following areas of the information model are unchanged from version 2.1:

• Categories
• Concepts
• Data providers
• Agencies
• Data consumers
• VTL transformation and expressions – with the exception of VTL mapping scheme as already noted
• Reporting taxonomy
• Process

3.4 Reference Metadata

Reference metadata has been substantially re-designed for version 3.0 to simplify the model and better support practical use cases.

Simplify Metadata Structure Definition
The Metadata Structure Definition (MSD) has been simplified to remove target information, and the support of multiple report structures. The MSD now only contains Metadata Attributes which are used to define the structure of a report.

Figure 2 version 2.1 Metadata Structure Definition (MSD)

Figure 3 the simplified version 3.0 MSD

Change to reference metadata reported against data

Reference metadata associated with datasets, data series or observations are now reported with the data. The dataset’s DSD must reference an MSD to define the structure of its reference metadata. In practice reference metadata for data are transmitted as part of the data message. The metadata attributes are treated in a similar way to the data attributes appearing in the message at the dataset, data series or individual observation level as appropriate. In contrast to simple data attributes, metadata attributes defined by an MSD can be organised into a hierarchical structure as illustrated in Figure 3 above. For this reason, metadata attributes appear in data messages structured in the same way as metadata messages.

The SDMX-ML example below is an excerpt from a structure specific data message illustrating reporting of reference metadata with a hierarchical structure at the observation level.

For completeness, the excerpt also shows:

• OBS_STATUS – a simple observation-level data attribute
• TITLE – a multi-lingual data attribute
• SOURCE_AGENCY – a multi-value data attribute

 <Obs xsi:type="dsd:ObsType" OBS_VALUE="112" OBS_STAUS=”A” TIME_PERIOD="2010-09">
     <!—- complex multi-value and multi-lingual data attributes -->
     <Comp id="TITLE" xsi:type="ns1:TITLE_ATTRIBUTE">
         <Value>
         <common:Text xml:lang="en">Some English Text</common:Text>
         <common:Text xml:lang="fr">Quelques textes en anglais</common:Text>
         </Value>
     </Comp>
     <Comp id="SOURCE_AGENCY" xsi:type="ns1:SOURCE_AGENCY_ATTRIBUTE">
         <Value>4F0</Value>
         <Value>4D0</Value>
         <Value>CZ2</Value>
     </Comp>
 <!—- metadata attributes are reported like in metadata messages -->
 <Metadata>
     <Attribute id="COLLECTION">
         <Attribute id="METHOD">
             <Text lang="en">AAA</Text>
         </Attribute>
     </Attribute>
     <Attribute id="CONTACT">
          <Value>CONTACT 1</Value>
          <Attribute id="NAME">
             <Value>Contact 1 Name 1</Value>
      </Attribute>
      <Attribute id="NAME">
             <Value>Contact 1 Name 2</Value>
         </Attribute>
     </Attribute>
     <Attribute id="CONTACT">
           <Value>CONTACT 2</Value>
         <Attribute id="NAME">
             <Value>Contact 2 Name 1</Value>
       </Attribute>
       <Attribute id="NAME">
                 <Value>Contact 2 Name 2</Value>
                </Attribute>
           </Attribute>
       </Metadata>
 </Obs>

New - Metadata Provision Agreement

In version 2.1 a Provision Agreement could be used to report information against a Dataflow or Metadataflow. From version 3.0 this is managed by two separate structures, the Data Provision Agreement and the Metadata Provision Agreement.

Move target to Metadataflow and Metadata Provision Agreement
For reference metadata that is reported against structures, the allowable targets information which is used to specify what structures the reference metadata can be reported against, has moved to the Metadataflow and can be further refined in the Metadata Provision Agreement.

Add maintainable properties to reference metadata
A Metadataset now has mandatory identification information, (owner id, id, version) enabling metadata providers to uniquely identify their reports for create, update or delete maintenance operations.

3.5 Microdata Exchange

Several changes have been made the Data Structure Definition to support microdata use cases in addition to aggregated time series.

Multiple measures measures are a common characteristic of microdatasets. To support this use case, the MeasureDimension has been deprecated and replaced with the option to define zero or more measures. Measures now act like any other component in that they use concepts, can have their own local coded or uncoded representation defined within the Data Structure Definition, and can be either mandatory or conditional. Creating a measure with the “MEASURE” concept role applied emulates the version 2.1 MeasureDimension behaviour as illustrated in the SDMX-ML example below:

 <str:MeasureList id=”MeasureDescriptor”>
     <str:Measure id=”OBS_VALUE” minOccurs=”1” maxOccurs=”1” usage=”mandatory” > <str:ConceptIdentity>
         <Ref id=”OBS_VALUE” maintainableParentID=”CONCEPTS” agencyID=”SDMX” maintainableParentVersion=”1.0.0” />
     </str:ConceptIdentity>
     <str:LocalRepresentation>
         <str:TextFormat textType=”String” isMultiLingual=”true” />
     </str:LocalRepresentation>
     <str:ConceptRole>
         <Ref id=”MEASURE” maintainableParentID=”SDMX_CONCEPT_ROLES” agencyID=”SDMX” maintainableParentVersion=”1.0.0” />
     </str:ConceptRole>
  </str:Measure>
         ...
     <str:Measure>
 </str:MeasureList>

Multi-value measures and attributes
Both measures and attributes have been extended with the option to take ‘arrays’ of 193 multiple coded or uncoded values. This supports use cases like multiple observation 194 status flags. New minOccurs and maxOccurs properties define the valid number of 195 values. The usage property separately defines whether the measure or attribute is mandatory or optional. In the SDMX-ML measure example above, the properties minOccurs=”1” maxOccurs=”1” usage=”mandatory” specify that OBS_VALUE must be 198 reported, and can only consist of a single value.

Attributes relationship to measures
In addition to attaching attributes to a specific level within the dataset, their relationship 202 to measures can also be defined.

Value lists
Value lists help in modelling microdata by providing an enumeration similar to code lists 206 but allowing any string values without being restricted to the rules of SDMX identifiers. That allows ValueItems (the equivalent to Code) to contain symbols like ‘¥’ and ‘€’, but also means they are not identifiable.

3.6 Geospatial Data Exchange

The version 3.0 model has been extended to provide explicit support for geospatial data.

GeospatialInformation type
A new GeospatialInformation string type has been added which can be used as the representation for any dimension, attribute or measure component. The value which is a string expression conforming to the syntax defined in Section 6 of the technical specifications precisely defines a ‘Geo Feature Set’ – a collection of geographical features like points, lines or polygons. Its use is recommended in conjunction with the “GEO_FEATURE_SET” concept role.

Geospatial code lists

Two new specialised types of code list have been added where the definition of each code includes additional geospatial information in addition to the standard ID, name and description:

• GeographicCodelist – each item includes an element to represent a specific Geo Feature Set which is described using the same expression syntax as for GeospatialInformation type.
• GeoGridCodelist – A code list defining a geographical grid composed of cells representing regular squared portions of the Earth. Each item references a cell within the grid.

3.7 Structure Mapping

The Structure Set in version 2.1 is a container for many mapping structures including Data Structure Map, Codelist Map and Concept Map. For version 3.0 the Structure Set artefact has been deprecated and replaced with a number of new maintainables giving better flexibility and reusability, specifically: Structure Map, Concept Scheme Map, Representation Map, Reporting Taxonomy Map, Category Scheme Map and Organisation Scheme Map.

The version 2.1 Codelist Map been replaced with Representation Map which allows mappings to be defined between any combination of Code Lists, Value Lists and noncoded representations such as text strings and numbers.

Many-to-many source and target components
Structure mapping rules may be defined with both multiple source components and multiple target components in contrast to version 2.1 where only one source and target was allowed. That supports many-to-many (n-n) mapping use cases where the output of a mapping rule may be dependent on the combination of a number of input components. For instance:

Set the output component INDICATOR=”DE_A” if the input components are FREQ=”A” and REF_AREA=”DE”.

Similarly, an n-n rule may also set the values of any number of output components:

Set the output components FREQ=”A”, REF_AREA=”DE” if the input component INDICATOR=”DE_A”.

Fixed source and target
The Structure Map may now define input or output components which have a fixed value.

Time representations mapping
Non SDMX time representations may now be described in a Structure Map, allowing them to be mapped into SDMX time formats.

Regular expression and substring mappings
All item maps allow the use of regular expressions and substrings to match source values, specifically: Concept Scheme Map, Reporting Taxonomy Map, Category Scheme Map and Organisation Scheme Map.

Item maps validity period
Item maps may further define the period for which the mapping is valid, meaning the mapping rule will only be applied if the row of information being mapped is within the period.

3.8 Constraints

Constraints in version 3.0 are modelled using two separate artefacts which replace the version 2.1 content сonstraint:

• data сonstraint for data; and
• metadata сonstraint for reference metadata.

Metadata сonstraint differs from its data counterpart in having a simplified cube region model better suited to reference metadata reporting use cases and not carrying details of the constrained targets – that information instead being defined directly within the metadataflow and Metadata Provision Agreement. Thus, metadata related constraints only specify сonstraints to the values of metadata attributes.

The ‘%’ wildcard character can now be used when defining cube region сonstraints to match multiple codes with a single expression, for instance for economic activity, ISIC4_% matches all codes beginning with ‘ISIC4_’ avoiding the need to maintain an explicit list.

The validity period definition has been moved from the сonstraint to the individual constraining terms, specifically CubeRegion, DataKeySet and MetadataTargetRegion providing more granular control.

Attachment сonstraints have been deprecated due to a lack of use cases.

3.9 Code List Extension

In addition to the two new specialised geospatial forms, the option has been added to define a code list as an extension of, or by inheriting codes from, other lists. An optional prefix can be added to inherited codes to disambiguate duplicates.

This feature allows new code lists to be easily derived from existing lists without the need to make and manually maintain copies. When querying for extended code list structures using the REST API, the option has been added to retrieve either the definition or the materialised list. Traditional literal lists of codes continue to be supported.

3.10 Discriminated Union of Code Lists

Combining code list extension with wildcarded сonstraints solves the discriminated union of code lists problem where a classification or breakdown has multiple “variants” which are all valid but mutually exclusive. A common example is economic activity where several alternative classification schemes are in use including ISIC revisions 1 to 4 and NACE as used in the European Community.

3.11 Code Hierarchies

Code hierarchies allow the definition of complex hierarchies of codes from potentially multiple lists for data discovery purposes. Hierarchical Codelist has been deprecated and replaced by two new artefacts: Hierarchy – the actual hierarchy of codes, and Hierarchy Association links hierarchies directly to any other identifiable object, a capability missing from the version 2.1 model. Further, the linkage can be within a particular context, for instance linking a hierarchy to a dimension within the context of a specific Dataflow (dimension REF_AREA in the context of the ECB:EXR Dataflow).

4 Versioning of Structural Metadata Artefacts

Version 3.0 adopts semantic versioning principles for versioning of metadata artefacts following the rules set out at https://semver.org However, this is not mandatory, and organisations may continue to use the pre-existing two-digit versioning strategy, or not to version artefacts by omitting the version property. The version number no longer defaults to 1.0 if not explicitly set.

Semantic version numbers are three digits:

MAJOR.MINOR.PATCH

Where

• The first digit (major) indicates that changes (either new features or bug fixes) are not backward compatible.
• The second digit (minor) indicates that features have been added in a backward compatible manner.
• The third digit (patch) indicates that bugs have been fixed in a backward compatible manner.

Examples:

SDMX:CL_AREA(1.0.0)
SDMX:CL_AREA(2.3.2)

Dependency management

Additional constructs are possible for dependency management when referencing structures. For instance:

2.3+.1               Means the currently latest available version >= “2.3.1” and < “3.0.0” (all backwards compatible versions >= “2.3.1”).
2+.3.1               Means the currently latest available version >= “2.3.1” (even if not backwards compatible).

Draft structures

A key principle is that semantically versioned structures are immutable and must not be changed without a corresponding change to the version number, except where explicitly marked as draft using extensions to the version number.

MAJOR.MINOR.PATCH-EXTENSION

1.10.0-draft                Means that version 1.10.0 is still being modified and may change – equivalent to setting isFinal=false in SDMX 2.1.
1.10.0-unstable          Alternative to -draft.
1.10.0-notfinal            Alternative to -draft.

The SDMX 2.1 isFinal property is deprecated in 3.0.

5 REST Web Services API

5.1 Simplified list of resources

The version 3.0 REST API has just five main resources:

• structure
• data
• schema
• availability
• metadata

All structure and item queries have been organised under the structure resource in contrast to the version 2.1 API which specified a separate resource for each structure.

This and changes in the URLs and query parameters on the data, availability and metadata resources means that, with the exception of schema queries, the version 3.0 API is not backwardly compatible.

5.2 Improved data queries

Data queries have been changed to provide more granular selections from contexts wider than just a Dataflow.

Extend the context of data retrieval
Version 2.1 data queries always retrieved data from a single specific Dataflow. In version 3.0, the query context may be specified as:

• Dataflow;
• Data Structure Definition – i.e., all Dataflows that use it; or
• Provision Agreement – i.e., all Dataflows associated with it.

Data queries may also search across datasets, for instance “retrieve all data about a country”.

Component-based filters
Expressions filtering on individual components can now be included as part of the data query URL.

/data/dataflow/ESTAT/ICP?c[REF_AREA]=CH&c[CONF_STATUS]=F

Support for operators
Filter expressions can also include operators.

/data/dataflow/ESTAT/ICP?c[REF_AREA]=DE&c[ICP_ITEM]=sw:01&c[TIME_PERIOD]=ge:2015

Operators include:

Support for multiple keys
Queries can now specify multiple series keys.

/data/dataflow/ESTAT/ICP/1.0.0/M…A.ANR,M…A.INX,M…B.CTG

5.3 Improved reference metadata queries

Reference metadata queries have been improved with a number of new options to retrieve metadata reports.

Get metadata reports by ID

/metadata/metadataset/ESTAT/QUALITY_REPORT/1.0.0

Get metadata reports by Dataflow

/metadata/metadataflow/ECB/METHODOLOGY/*/FR2

Get metadata reports about a Data Structure Definition

/metadata/structure/datastructure/BIS/BIS_CBS/1.0

5.4 Structural metadata maintenance

Support has been added for maintenance of structural metadata.

HTTP verbs PUT, POST and DELETE may be used to submit SDMX-ML or SDMX-JSON structure messages to an SDMX registry for the purposes of adding, updating or deleting structural metadata artefacts.

6 XML, JSON, CSV and EDI Transmission formats

6.1 SDMX-ML

The SDMX-ML XML messages have been modified and updated for version 3.0. While they broadly follow the same principles, there have been significant changes which break backward compatibility.

Structure message
The SDMX-ML structure message is used for transmission of structural metadata. It closely reflects the SDMX information model and has therefore been significantly updated for version 3.0 with the addition of new structures, modifications where structures have changed, and removal of deprecated structures like Structure Set.

Additionally, the way the individual artefacts are organised into ‘collections’ within the message has been significantly revised with a simpler flat structure adopted as set out in the following table:

No changes have been made to the way the following artefacts are organised in the structure message:

From version 3.0, collections can appear in any order within a structure message.

Data messages

All legacy SDMX-ML data messages have been deprecated with the exception of Structure Specific Data which becomes the sole standard format for transmission of SDMX data in XML in version 3.0.

Specifically, the following data messages are not supported in version 3.0:

• SDMX-ML 1.0/2.0 Generic (time-series) data message
• SDMX-ML 1.0/2.0 Compact (time-series) data message
• SDMX-ML 1.0/2.0 Utility (time-series) data message
• SDMX-ML 1.0/2.0 Cross-Sectional data message
• SDMX-ML 2.1 Generic data messages (for observations, time-series and crosssectional data)

The Structure Specific Data message has been extended to support the transmission of microdata sets, in particular those with multiple measures and array values for measures and attributes.

As detailed in paragraph 3.4, the message now additionally allows data’s reference metadata to be reported as an integral part of the dataset. Like data attributes, these metadata attributes are included in the data message at the dataset, series or observation level as appropriate.

The time series variant of the Structure Specific Data message is no longer used.

Reference metadata message
The Generic Metadata message remains the standard format for transmission of reference metadata sets in XML but has been modified to support the revised version 3.0 reference metadata model.

Registry structural metadata ‘query’ messages
As a consequence of the deprecation of the SOAP API and standardisation on REST, the structural metadata ‘query’ messages have all been removed. In version 3.0, querying an SDMX Registry for structural metadata is performed solely using REST GET.

Structure referencing
The option to reference structures using Agency, ID and Version has been removed. From SDMX version 3.0 URN is used for all referencing purposes with the exception of local references such as where groups reference dimensions within a DSD.

6.2 SDMX-JSON

Like SDMX-ML, the SDMX-JSON messages have been significantly modified and updated for version 3.0. They are not backwardly compatible with version 2.1.

Structure message
The SDMX-JSON structure message closely replicates the SDMX-ML equivalent. Like that of SDMX-ML it has been updated to align it with the version 3.0 information model with addition, deletion and modification of artefacts as required. The organisation of the structure collections has also been revised as detailed in paragraph 6.1.

Data message
The SDMX-JSON data message has similarly be updated. Additional changes have been made to allow a single message to carry data from multiple datasets with potentially different Data Structure Definitions to support REST data queries of the form “retrieve all data about a country”. For this reason, the version 3.0 SDMX-JSON is not backwardly compatible with version 2.1 data messages. Support has been added for the transmission of microdata and reporting of reference metadata on data as an integral part of the dataset.

Reference metadata message
The SDMX-JSON metadata message has also been updated to support the version 3.0 reference metadata and Metadataset specifications.

Structure referencing
As for SDMX-ML, the option to reference structures using Agency, ID and Version has been removed with URN used for all non-local referencing purposes.

6.3 SDMX-CSV

CSV in SDMX is used transmission of data and reference metadata only.

Data message
The SDMX-CSV data message has been modified to align with the version 3.0 information model, support the enhanced REST API and ensure that data can be freely converted to and from the XML and JSON formats without loss. These changes include:

• An additional column identifying the type if the artefact defining the structure of the data: “dataflow”, “datastructure” or “dataprovision”;
• A column for the structure artefact’s identification of the form

ESTAT:NA_MAIN(1.6.0) which replaces the dataflow identifier in version 2.1; and

• A column for the dataset action: information, append, replace or delete, which is consistent with both the the SDMX-ML and SDMX-JSON data messages.

Reference metadata message
The SDMX-CSV metadata message is new for version 3.0 and, like the SDMX-ML and SDMX-JSON equivalents, is used for the transmission reference metadata sets.

6.4 EDI deprecation

The EDI format for transmission of both structures and data has been deprecated. Version 3.0 is therefore not backwardly compatible with legacy EDI messages.

Appendix A – Version 2.1 Information Model