How is a code checked in a dataset?

A code in a dataset is checked through the dataset metadata, not through the code value alone. The metadata states the classification version, the reference period, the statistical unit, and any note about a break in series, a backcast, or double reporting. A code value without that metadata can be read on the wrong basis, compared with a series on a different basis, or treated as a register entry when it is an aggregate breakdown. The metadata is part of the figure, and a citation that drops it is incomplete.

What does dataset metadata state?

Dataset metadata states the classification used, the version of that classification, the reference period, the unit, and the source. Eurostat states on its transition page that parallel datasets presenting current statistics broken down by NACE Rev. 2.1 will be published for many domains, and that backcasting methods will be applied so that previous statistics can be presented by NACE Rev. 2.1. A dataset in transition may therefore exist in two versions for the same reference period, and the metadata is the only record that distinguishes them. The Central Statistics Office states that its outputs will begin to be published using NACE Rev. 2.1 from January 2026 onwards (FIXED, as stated by the Central Statistics Office), starting with Business Statistics and Labour Force Statistics, amongst others, on a gradual basis. A dataset from that office must therefore be checked output by output, not assumed from the office name.

How are version and reference period checked together?

Version and reference period are checked together because neither controls the other on its own. A dataset published in a given year can describe an earlier reference period compiled on the older classification, and a dataset for a recent reference period can be restated on the new basis through backcasting. The backcasting manual, published by Eurostat in its 2025 edition, gives the methods for producing consistent historical series, and it assumes double coding of units in the Statistical Business Register under both versions for at least one point in time. A reader should record the reference period, the version, and the method note together, and should not compare a restated series with an unrestated one without noting the difference. The general range, from 1 January 2025 to 1 January 2031 (FIXED range), frames the transition, but the version for a single dataset is read from its metadata, not from the range.

What does an aggregate code show, and what does it hide?

A dataset table often shows codes at section or division level, as aggregates of the classes below them. National Accounts groupings based on NACE, labelled A10, A21, A64 and A88, are noted by the Central Statistics Office for its publishing of data. An aggregate code shows the total for the grouping, and it hides the movement inside the grouping. Where a class splits, the aggregate may stay similar while the classes inside it move, and where a class moves between divisions, the aggregate for one division will change while the aggregate for another moves in the opposite direction. A reader who compares aggregates across the changeover should check the correspondence table for the classes inside the aggregate, and should check whether the aggregate definition itself changed. An aggregate is a summary, and a summary should never be quoted as if it were a class entry.

Metadata item	Question it answers
Classification version	NACE Rev. 2 or NACE Rev. 2.1, or the named national version
Reference period	Which period does the figure describe?
Statistical unit	Enterprise, establishment, legal entity, or other unit?
Break or backcast note	Was the series restated, and by which method?
Source	Which institute published the dataset, and when?

The table is a citation checklist. A citation that answers all five rows can be repeated by another reader. A citation that answers only the code value cannot.

What does a dataset code not prove about a business?

A dataset code is a breakdown in an aggregate. It does not prove that a named business carries on the activity shown, it does not prove the size or the output of that business, and it does not prove any question about licensing, tax or funding. Datasets are compiled from many units, and a published cell may combine many units, or may be suppressed where a cell would disclose a single unit. A code in a dataset should never be quoted as a register entry for a business. The register entry is held separately, and it is confirmed with the register that holds the record for that unit.

Where is a dataset code confirmed?

The meaning of the category is confirmed in the Eurostat explanatory notes. The move between versions is confirmed in the correspondence tables. The version, reference period and method for the dataset are confirmed in the dataset metadata published by the institute that produced it. The entry for a specific business is confirmed with the national statistical institute or the business register that holds the record, not with the dataset.

Record the version, reference period, unit and method note with every dataset figure, and confirm a business entry with the register that holds it.