Data point model — the layer everyone skips
The data point model is the layer most reporting projects skip, and the layer most reporting failures come from. The European Banking Authority describes its DPM as a structured representation of the data: every business concept, the relations between them and the validation rules, with the technical detail needed to build a reporting solution. Put simply, the model is the meaning and the XBRL taxonomy is only the encoding of it. Since the EBA’s reporting framework 4.0 the model itself has changed — DPM 2.0, a rebuilt glossary, a new validation-rule language and an xBRL-CSV-only format. This piece explains what a data point is, what changed, and how to map to it in practice, with three worked cases.
1. A cell is not a cell
A reported figure is not identified by its position in a template. It is identified by a combination: a main property (the metric — a carrying amount, a number of transactions) plus the properties and items that qualify it, such as counterparty sector, country, currency, maturity band or accounting portfolio. The template is a human-readable view of those combinations, not their definition. Each combination has a data point ID, and that ID — not the row and column — is what the supervisor stores.
Two consequences follow:
- The same data point can appear in more than one template. If it does, it must be the same number, and cross-template validation rules will check it.
- Visually similar cells can be different data points. Two rows with the same label in different templates may differ in one qualifier, and mapping both to one source query is a silent error.
2. Two layers: semantics and syntax
The EBA describes the DPM in two layers. The semantic layer, also called the glossary, holds the content used to describe reporting requirements — the concepts and their definitions. The syntactic layer is the structure that organises and stores that content: the standard itself, its database and the tables built from it.
Both layers were rebuilt in recent years. The EBA and EIOPA jointly upgraded the syntactic standard from DPM 1.0 to DPM 2.0, used from release 4.0. In parallel, the EBA’s DPM Quality Review project rewrote the glossary, starting from every concept in the data dictionary up to release 3.2.
The vocabulary changed with it. What DPM 1.0 called domains, dimensions, domain members and hierarchies are, in the DPM 2.0 glossary, categories, properties, items and subcategories. Anyone reading older mapping documents or vendor screens will meet both sets of terms.
3. What DPM 2.0 adds
The EBA’s FAQ on its reporting innovations lists three groups of new features:
- Glossary management. The glossary can now evolve: categories, items and properties can be reorganised, supercategories built from existing ones, and items, properties and subcategories versioned. More complex items can be built from simpler parts — the “compound items” flagged in the EBA’s conversion file.
- Template enhancements. Template headers are versioned as well as templates. Templates can be linked to describe relational models, and new groupings support automation.
- Validation and calculation. Rules are machine-readable, so validation engines can process them without manual interpretation.
One constraint has a visible effect on templates: DPM 2.0 does not allow identical data points in the same template. Where an older template repeated a data point, the EBA split it into several technical templates. The printed layout can look unchanged while the technical structure underneath has more pieces.
4. The transition, release by release
| Release | What changed |
|---|---|
| 4.0 (published December 2024) | Taxonomies generated from the DPM 2.0 database under taxonomy architecture 2.0. Files submitted to the EBA must refer to them. New semantics for the basic domains (metrics, main category, base, portfolios, counterparties, liquidity) in the modules updated in that release. DPM-XL becomes the validation-rule language. |
| 4.0 and 4.1 | Transition period: the EBA published both DPM 1.0 and DPM 2.0 databases for compatibility. |
| 4.2 | End of the transition. Only DPM 2.0 and its taxonomy are published. The enhanced glossary applies to all modules, not only those changing — from reference dates of 31 December 2025 onwards, depending on the module. DORA follows in release 4.3. |
| Reference date March 2026 | For submissions to the EBA’s Euclid platform, only xBRL-CSV is accepted. Earlier reference dates may still use xBRL-XML or xBRL-CSV. DORA uses plain CSV. |
For firms reporting to a national authority rather than to the EBA, the national authority sets the format it accepts, so check its own technical notices. The model underneath is the same one.
5. Data point IDs and the conversion file
When the glossary behind a requirement changes, the EBA creates a new version of the data point. Its ID usually — not always — stays the same. The description is updated, but the requirement itself does not change. In release 4.2 the EBA did generate new IDs in a few cases, where earlier modelling had not reflected the business identity correctly.
To document the rewrite, the EBA published a conversion file from the DPM 1.0 glossary to the DPM 2.0 glossary, updated with each release until the project closed in 4.2. It has one sheet per element type and records each decision: splits (one to many), combinations (many to many), removals, relabelling, definitions as compound items, or no change. Resubmissions are not affected: each is made under the module version it refers to, and module versions before 4.0 were not changed.
The EBA itself flags who is exposed: anyone whose systems rely on glossary elements — national validation rules on top of the EBA’s own, analytics, data warehouses keyed on old dimension codes.
6. Validation rules in DPM-XL
Since release 4.0 validation rules are written in DPM-XL. Much of what used to sit in spreadsheet columns now sits inside the rule expression. The EBA describes two key elements: a selection operator in braces, which specifies the record set, the cell, variable or operation the rule applies to, the default value and the arithmetic approach; and a with operator that gives shared context once instead of repeating it.
From release 4.2 the EBA publishes a single document consolidating each rule’s history since 4.0 — applicability by module version and period, severity, active or inactive status, activation dates and changes to the expression. Most rules within a module are still implemented as XBRL assertions in the taxonomy, so a standards-compliant XBRL validator keeps working. The full metadata of every rule sits in the DPM 2.0 database. For how rule failures surface in practice, see Validation rules.
7. Worked case: two rows with the same label
Facts: a payment institution builds its first prudential return. Two templates each carry a row labelled “loans and advances”. The analyst maps both to the same ledger query.
What the model says: the two rows are different data points. One is qualified by counterparty sector and the other is not — or they sit in different accounting portfolios. Each has its own data point ID and its own validation rules.
What the practitioner does: pulls the data point definitions from the DPM database for both cells and lists every qualifier side by side. One row needs a counterparty-sector split that the ledger cannot produce, so it is logged as a data-collection gap and not derived in the reporting layer.
Outcome: the return validates and is right, and the gap log becomes the brief for the source-system change.
8. Worked case: a national rule breaks after the glossary rewrite
Facts: a reporting team runs its own pre-submission checks, keyed on DPM 1.0 dimension and member codes. After moving to the 4.2 taxonomy, a dozen checks stop firing.
What the model says: the requirements did not change, but the glossary did. Some old members were split, merged or relabelled, and the checks look for codes that no longer exist.
What the practitioner does: runs every code used in the in-house checks through the EBA conversion file, re-keys one-to-one changes automatically, and reviews splits and merges by hand. Then compares data point IDs between the old and new versions to catch the few that changed.
Outcome: the checks run again, and the mapping document records which glossary version each check targets.
9. Worked case: the move to xBRL-CSV
Facts: a firm’s reporting tool produces xBRL-XML. Its first return under the 4.2 framework, for a reference date after March 2026, is due.
What the rule says: for submissions reaching the EBA from reference date March 2026, only xBRL-CSV is accepted. Whether the national authority still accepts xBRL-XML from the firm depends on its own technical notices.
What the practitioner does: checks the authority’s notice first, then tests an xBRL-CSV export of the same data and reconciles it cell by cell against the XML version before the deadline.
Outcome: the format change is handled as a tested release, not discovered as a rejection.
10. The mapping document
| Step | What good looks like |
|---|---|
| Inventory | The data points actually in scope for your firm, not the whole framework |
| Map | Each in-scope data point traced to a source field and a transformation, qualifier by qualifier |
| Document the gaps | Where the source cannot produce a qualifier, that is a data-collection change, not a reporting workaround |
| Version the mapping | Each mapping carries the framework release and glossary version it was built for, and data point IDs, not cell positions |
The mapping document is the deliverable that survives staff turnover, vendor changes and framework releases. Without it, the knowledge lives with whoever last built the extract, and every release becomes archaeology. A derived qualifier is not wrong in itself, but it needs the same documentation as any other transformation: a supervisory query about one figure will eventually ask how each of its qualifiers was set.
FAQ
Is the DPM the same as the taxonomy?
No. The DPM is the structured representation of concepts, relations and validation rules. The XBRL taxonomy expresses that model in a technical format for transmission. Since release 4.0, EBA taxonomies are generated from the DPM 2.0 database.
What is a data point ID?
The identifier of one reporting requirement: a main property combined with the items and properties that qualify it. It usually survives glossary changes, though the EBA created new IDs in a few cases in release 4.2.
What changed between DPM 1.0 and DPM 2.0?
A versionable glossary of categories, properties, items and subcategories replacing domains, dimensions, members and hierarchies; versioned template headers; and machine-readable validation rules in DPM-XL.
Does the EBA still publish DPM 1.0?
Only for releases 4.0 and 4.1. From release 4.2 it publishes DPM 2.0 only.
When does xBRL-CSV become mandatory?
For submissions to the EBA’s Euclid platform, from reference date March 2026. Your national authority’s notices set what it accepts from you.
Do we need the DPM if we have the templates?
Yes. The templates show the layout. The model shows what each cell is, qualifier by qualifier, which is what a correct mapping needs.
What to do, today
- Head of regulatory reporting: check which framework release and glossary version each of your mappings was built for.
- Data team: key mappings and in-house checks on data point IDs, and run old codes through the EBA conversion file.
- Technology: confirm your tool produces xBRL-CSV and that your authority’s format notices are tracked.
- Compliance: keep the gap log — qualifiers the source cannot produce — as a standing item with an owner.
Related: XBRL taxonomies · Validation rules · The AnaCredit data model · Building a reporting pipeline · Surviving a framework release


