How is a record of a code check kept?

A record of a code check is kept so that the check can be repeated, and so that the entry can be rechecked after a recoding. A check that is not recorded cannot be repeated, and a check that is recorded without the version and the date cannot be distinguished from a check made on a different version at a different date. The record needs five items: the holder consulted, the unit identifier, the code as shown, the classification version, and the date of the check.

A sixth item, the source page or register consulted, should be added wherever the record will be reused, so that a later reader can open the same record and repeat the check.

What does a complete check record contain?

A complete check record contains the holder, the identifier, the code, the version, and the date, in that order, with each item copied exactly as shown. The holder is the institute or register consulted, named in full: for example, the Central Statistics Office Business Register, the KVK Business Register, or the Sirene directory. The identifier is the register identifier for the unit, such as the SIREN number in France. The code is copied exactly as shown, including any letter, dot, or national digit, and it is not shortened to a higher level unless the record states that it is an aggregate. The version is the classification edition, such as NACE Rev. 2.1, SBI 2025, or NAF 2025. The date is the date the record was consulted, stated as the check date. The table below shows the record shape, with an example row left blank for the reader’s own check, so that the shape, not a fabricated entry, is what is carried forward.

Item	Record
Holder consulted	Institute or register, named in full
Unit identifier	Register identifier, as shown
Code as shown	Copied exactly, with letter, dots and any national digit
Classification version	Edition, such as NACE Rev. 2.1, SBI 2025, or NAF 2025
Check date	Date the record was consulted
Source	Page or register consulted, so the check can be repeated

The table is a template, and it should be copied into the record, not filled with an invented entry. An invented entry in a template would be a fabricated record, and it would mislead any reader who treated it as a real check. The template is useful precisely because it is empty until a real check fills it.

When is a check record rechecked?

A check record is rechecked when the classification changes, when the register carries out a recoding, when the unit reports a change of activity, and when the record is reused after a gap. The French change is a clear case: the new APE code is shown in the Sirene tool before it takes effect from 1 January 2027 (FIXED), and a record made before that date shows the future code, not the code currently in force. That record must be rechecked on or after the effective date, and the new record must state the new check date. The Irish and Dutch changes are gradual: the Central Statistics Office begins publishing on NACE Rev. 2.1 from January 2026 onwards (FIXED, as stated by the Central Statistics Office), and Statistics Netherlands implements SBI 2025 gradually from 2026 onwards (FIXED, as stated by Statistics Netherlands), with both SBI versions in use during the implementation period. A record made during a gradual change must state the version shown at the time, and must be rechecked when the version in use changes. A record that is never rechecked will, in time, describe a version that is no longer in use.

How are two records for the same unit compared?

Two records for the same unit are compared by version, then by date, then by code. If the versions differ, the difference in code may be a classification change, and the correspondence table should be consulted for the mapping type. If the versions are the same and the dates differ, the difference in code may be a correction or a reported change, and the holder should be asked which. If the versions and the dates are the same and the codes differ, one of the records is wrong, or the records describe different units, and the identifiers should be checked first. A comparison that starts with the code, without checking version, date and identifier, will treat a version change as a correction, a correction as a version change, or two different units as one. The order of the comparison matters, and the record shape in the table above enforces that order by putting the holder, identifier, version and date before the code in the reading of the record.

What is not kept in a check record?

A check record does not keep a determination of the correct code, a statement about tax, licensing, funding, collective agreement or social security status, or a claim about the unit’s activities beyond what the register shows. Those statements belong to other records, held by other holders, and adding them to a check record would make the record claim more than the check established. A check record also does not keep a code from a finder tool, a vendor listing, or a blog post as if it were a register entry. A code from such a source can be noted as a lead, with the source named, but it is not a confirmation, and it should never be copied into the code row.

Where is a check record stored, and who can repeat it?

A check record is stored with the work that uses it, so that the work can be audited, and it is written so that another reader can repeat the check from the record alone. That means the holder, the identifier, the version, the check date, and the source must all be present, and the code must be copied exactly. A record that meets that test can be repeated by a reader who has never seen the unit before. A record that fails it can only be trusted on the word of the person who made it, and it should not be reused. The test is simple, and it is the test that separates a check from an assertion.

Keep the holder, identifier, code, version and check date together in every record, recheck after every recoding, and confirm a single entry with the register that holds it.