Media Text Localization

1. A translated string is not a localization contract

A Persian, German or Arabic string only proves that some text exists. Publication-safe localization needs the exact source_content_version, stable source_unit_id, requested locale, constraints and review state.

2. CON-011 ownership boundary

This lesson owns media-text localization for captions, overlays, titles, transcripts, labels and descriptions. CON-012 owns generic version/derivative lineage, CON-013 campaign readiness, and C04 media transform/QC.

3. Give each source unit stable identity

Do not rely on positions such as “third subtitle line.” Record a stable source_unit_id such as CTA-01 or CAPTION-042; reordering must not silently create a new identity.

4. Bind targets to an exact source version

Every target unit binds to the exact source content version. If the English CTA advances from v8 to v9, a target still bound to v8 is STALE even when its visible text still looks plausible.

5. BCP 47 identifies language precisely

RFC 5646 builds language tags from subtags such as language, script, region and variant. fa, de-DE and ar-Arab are language metadata; they do not prove translation correctness or approval.

6. Keep Translate policy explicit

ITS 2.0 includes a Translate data category. If a brand token or code is translate=no, creating a different target string is a policy violation, not successful localization.

7. Localization notes and terminology are context

A localization note can state tone, length, legal wording or terminology constraints. Removing that note may leave a string present while making its review evidence incomplete.

8. Record directionality explicitly

RTL locales such as Arabic and Persian require directionality to be part of the contract. Arabic text rendered under an LTR binding can be BLOCKED even when the words themselves are correct.

9. Version the target text too

Keep target_unit_id and target_text_version. A punctuation or CTA correction creates a new target version; do not overwrite the previous localization history.

10. Review state is independent from text existence

DRAFT, REVIEW_REQUIRED and APPROVED are distinct states. A target string and a valid language tag do not upgrade a unit to APPROVED.

11. Fallback is product behavior, not localization evidence

A UI may explicitly fall back from de-CH to de-DE or en. The requested locale record must still remain MISSING; fallback does not prove target-locale translation.

12. Staleness should be unit-level

If only the source CTA changes while the transcript remains unchanged, stale only the units bound to that CTA. Do not invalidate an entire locale without evidence.

13. Negative case: valid-looking text, wrong tag

If Persian-looking text is bound to an unrelated target language tag, reject the shortcut. Emit BLOCKED with LANGUAGE_BINDING_MISMATCH.

14. Transfer: CTA v9 + Arabic RTL + removed note

Advance the CTA to v9, keep the transcript unchanged, add ar-Arab with RTL, and remove one localization note. Rebuild only affected bindings and classify CURRENT, STALE and REVIEW_REQUIRED units.

15. Auditable output

The Locale Unit Matrix records Source Content/Version/Unit, source language, target tag, translate flag, notes, terminology, directionality, Target Unit/Version, review, derivative reference, staleness, fallback, publication eligibility, decision and reason. The strongest claim remains bounded to this exact evidence.