Protect tokens early
Apply translate="no" to braces, codes and URLs. Keep id values identical, because they are keys rather than prose.
Preflight for localized pages
Mango Translations is a new independent QA guide for translators and localization engineers. It provides checks for HTML markup, placeholders, storage sizes and quality issues. It does not sell translation services.
01
Translate="no" and the HTML translate attribute protect tokens; ITS Id Value protects identifiers. Nothing downstream is trustworthy if this stage is skipped.
02
The lang attribute is the counterpart of ITS Language Information, and direction is checked separately. A correct translation under the wrong lang is still a defect.
03
Compare placeholder multisets, then measure characters for expansion and bytes for storage. Log issues by type, severity and location before computing an overall rating.
Use ITS Translate="no" or the HTML translate attribute on every token, identifier and brand string that must survive translation unchanged.
W3C ITS 2.0 defines Translate values yes and no to express whether selected content is intended for translation. In HTML5 the native translate attribute is the counterpart of the ITS Translate category, so the same intent can be carried in markup without a separate ITS namespace.
Identifiers need a second guard. The HTML id attribute is the counterpart of ITS Id Value, which means an id used as a lookup key must be preserved exactly; translating or normalizing it silently breaks scripts, anchors and form labels.
Marking is only half the job. The check is whether the target page still carries the same set of protected tokens as the source. A token that disappears in translation is a defect even when the surrounding sentence reads well.
Apply translate="no" to braces, codes and URLs. Keep id values identical, because they are keys rather than prose.
Verify the html element declares the target language and the correct direction, including inline opposite-direction runs.
Six source placeholders against five target placeholders is a mismatch count of one, and one is enough to break runtime substitution.
Measure expansion in characters and storage in encoded bytes. Normalize to NFC before counting to avoid phantom differences.
Log error type, severity and location per issue, then derive the overall rating from those entries instead of typing it in.
Repeat the preflight after every target change and compare issue counts before and after publication.
A translated page can read perfectly and still ship broken: one dropped placeholder, one stale lang attribute, one string over its byte limit.
Six source placeholders against five target placeholders is a mismatch count of one, and one is enough to break substitution at runtime.
| Check | What is measured | Typical failure | Example figure |
|---|---|---|---|
| Markup | Protected tokens and ids preserved | Token translated or id changed | 1 missing token |
| Language markup | lang and dir values | Source language left on target page | 1 wrong lang value |
| Placeholder parity | Source versus target placeholder counts | One placeholder dropped | 6 vs 5, mismatch 1 |
| Expansion | Characters against source length | String grows past field cap | 1,200 extra characters |
| Expansion rate | Growth as a percentage | Underestimated budget for layout | 15% |
| Storage size | Encoded bytes against limit | Character count fits, bytes do not | NFC normalization required |
| Quality issues | Type, severity, location per issue | Type and severity collapsed into one field | 3 fields per issue |
References are limited to the W3C ITS 2.0 specification, its implementation guidance and the HTML5 translate attribute.