Key takeaway
A language-tag match describes metadata under a rule. It does not verify the content’s language, its creator’s location or fitness for a buyer’s task.
A language column is the start of the evidence
A buyer asks for English records suitable for a particular locale. An export owner counts every tag beginning with the letters en and calls the result verified coverage. That skips two decisions: which matching scheme expresses the request, and what evidence supports the tags assigned to the content. A clean code column cannot answer both.
RFC 5646 separates language, script, region and other subtags. Its region subtag describes linguistic variation associated with or appropriate to a region, such as spelling or usage. It is not a geographic provenance field for the author or event. Keep a location claim tied to separate source evidence rather than reading it out of an en-US label.
Four invented tags produce two explicit selections
This hypothetical metadata fixture contains four distinct records, tagged en-US, en-GB, en and de-DE. Apply RFC 4647 basic filtering, with case-insensitive exact matching or a match at a hyphen-delimited prefix. Under range en, the first three tags match. Under range en-GB, only the second tag matches. The bare en tag does not establish the narrower regional variation.
These are counts of declared metadata matches: three and one. The fixture supplies no content review, language-identification validation or source-person location. Those observations remain unknown rather than becoming zero verified English records or a claim that the labels are correct. The resulting selection can guide a review request without pretending that a text evaluation already happened.
| Synthetic record tag | Basic range en | Basic range en-GB | Content/location evidence supplied |
|---|---|---|---|
| en-US | Match | No match | None in this fixture |
| en-GB | Match | Match | None in this fixture |
| en | Match | No match | None in this fixture |
| de-DE | No match | No match | None in this fixture |
Name the matching scheme before broadening the request
RFC 4647 distinguishes basic filtering, extended filtering and lookup. Basic filtering uses an exact tag or an exact prefix followed by a hyphen; it is not an arbitrary substring search. A buyer can intentionally choose a broader range, but that choice changes the selection and belongs in the evidence record. Do not quietly substitute a fallback scheme to increase the count.
Retain the original tag, any approved canonical or mapped representation, registry/version basis where relevant, and the requested ranges. An extra script subtag can affect a basic-prefix comparison even when a human regards the language descriptions as related. Test the precise method on representative tags rather than assuming a shared language name implies identical matching behaviour.
Verify content through a separate, bounded procedure
Ask how tags were assigned: a source-system setting, a human annotation, a document-level detector or a rule inherited from a template. These are possible origins to investigate, not evidence that any particular archive used them. Preserve the assignment unit and method so a document containing several languages is not silently treated as a single verified language on every page.
A content-review plan should define its eligible units, intended language varieties, mixed/undetermined cases and the evidence needed for acceptance. Report the selection population and the reviewed subset separately. A tagged record that was never examined should remain tagged-but-unverified in that account, even if its code passes syntax and matching checks.
Return a coverage statement with its units intact
For the four invented records, the reproducible statement is three matches for en and one for en-GB under basic filtering. It makes no claim that three documents were read, that a speaker is from a particular country, or that the text meets a downstream training requirement. A receiving note can retain those matching results while holding the content claim open.
Keep requested ranges, matching algorithm, original/mapped tags, matching counts, unknown labels and actual content-review results in separate fields. That lets a buyer decide whether a targeted language review is worth requesting. Correct Unicode preservation remains a prerequisite for reading text accurately; it does not supply the missing language or provenance evidence.