How are two sources cross checked for the same code?

Two sources are cross checked for the same code by comparing them in a fixed order: version first, then unit, then date, then code. A difference in version explains a difference in code, and it is not a contradiction. A difference in unit explains a difference in code, and it is not an error. A difference in date explains a difference in code, where a recoding happened between the dates.

Only where the version, the unit, and the date are the same, and the codes still differ, is there a genuine discrepancy to resolve, and the resolution is to ask the holder of the record, not to choose between the sources by preference. A cross check that starts with the code, without checking version, unit and date, will treat every version change as an error, and every error as a version change.

What is compared first, and why?

Version is compared first, because it controls the meaning of the code. Eurostat states that statistics for the period 2008 to 2024 use NACE Rev. 2, while NACE Rev. 2.1 is to be used for European statistics from 2025 onwards (FIXED, as stated by Eurostat), and that parallel datasets on both bases will be published for many domains during the transition. Two sources on different versions can therefore show different codes for the same unit, and both can be correct on their own version. The correspondence table between the versions shows the mapping type, and where the mapping is one to many or a split, the difference is expected, and the explanatory notes for the new classes must be read. A cross check that finds a version difference should record it, record the mapping type, and stop, unless the question requires a single code on a single version, in which case the register entry on the required version is the answer, and the other source is cited as the earlier version, not as a contradiction.

What is compared second and third?

Unit is compared second, and date is compared third. Unit is compared by the register identifier, not by the name, because two units can share a name, and one unit can appear under different names in different sources. In France, the identifier is the SIREN number, as Service Public states for the Sirene tool. In the Netherlands, the identifier is the business identifier in the KVK record, with the code assigned by the KVK, as Statistics Netherlands states. Date is compared by the check date and, where the source states it, the reference period. A source consulted before a recoding and a source consulted after it will differ, and the difference is dated, not disputed. The French change, effective from 1 January 2027 (FIXED), is a clear case: a source consulted before that date can show a future code, labelled 2025 APE Code, while a source consulted earlier can show the current code, and both are correct for the date they were consulted. A cross check record should therefore give the check date for each source, so that a dated difference is visible as dated.

How is a genuine discrepancy recorded?

A genuine discrepancy is a difference in code where the version, the unit, and the date are the same, and it is recorded as a discrepancy, with both codes, both sources, and the check dates, and with a note that the holder was asked. The holder is the register that holds the record, or the institute that published the dataset, depending on which source is in question. The holder is asked for the entry as held, the version, and the date of the record, and the holder’s answer is recorded as the resolution, with the date of the answer. A discrepancy should never be resolved by choosing the source that looks more authoritative, or the source that was consulted most recently, without asking the holder, because the more recent source can be a copy of the older one, and the more authoritative looking source can be a listing, not a record. The table below shows the cross check record, with the discrepancy row included, so that a discrepancy is recorded in the same form as an agreement, and is not hidden by omitting one of the sources.

Comparison item	Source one	Source two
Version	Stated version	Stated version
Unit identifier	Identifier used	Identifier used
Check date	Date consulted	Date consulted
Code as shown	Code, copied exactly	Code, copied exactly
Outcome	Agree, differ by version, differ by date, or discrepancy	As recorded, with the holder asked where discrepant

The table forces the version, unit and date into the record before the code is compared, and it forces the outcome to be named, rather than left as an impression. A cross check that fills the table has been done. A cross check that quotes two codes without filling it has not.

What does a cross check not decide?

A cross check does not decide the correct code for a unit, and it does not decide tax, licensing, funding, collective agreement or social security questions. It decides whether two sources agree, and, where they differ, why they differ. The correct code, where a correction is needed, is decided by the holder of the record, on the activity description the holder requires. The cross check is the evidence presented to the holder, not the decision itself. A cross check that is presented as a decision overstates the check, and it should be reframed as a record of agreement or difference, with the holder named as the decider.

Where is a cross check completed?

A cross check is completed in the register that holds the record, for the entry, and in the classification documents, for the meaning and the movement. The meaning of a category is confirmed in the Eurostat explanatory notes, and the move between versions in the correspondence tables. Confirmation for a specific business comes from the national statistical institute or the business register holding its record. A cross check that ends with two sources and no holder has not been completed, and it should be recorded as open, with the holder to be asked, rather than as closed.

Compare version, unit and date first, record the outcome, and ask the holder where the codes genuinely differ.