Mango Translations — Translation QA checks | mangotranslations.xyz

Preflight for localized pages

Mango Translations — Translation QA checks | mangotranslations.xyz

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.

How do I mark textthat must not beHow do I checklanguage, directionHow do I compareplaceholders andHow do I measurecharacters, bytesHow do I logquality issuesWhat does acomplete preflight
A map of this guide's sections

01

Markup first

Translate="no" and the HTML translate attribute protect tokens; ITS Id Value protects identifiers. Nothing downstream is trustworthy if this stage is skipped.

02

Metadata second

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

Parity and size last

Compare placeholder multisets, then measure characters for expansion and bytes for storage. Log issues by type, severity and location before computing an overall rating.

How do I mark text that must not be translated?

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 product codes, URLs, email-like tokens and placeholder braces.
  • Keep id values identical between source and target; treat them as keys, not prose.
  • Do not translate the contents of attributes that feed scripts or analytics.
  • Re-check protected tokens after every target update, not only at first delivery.

Checks by stage

Markup

Protect tokens early

Apply translate="no" to braces, codes and URLs. Keep id values identical, because they are keys rather than prose.

Metadata

Check lang and dir

Verify the html element declares the target language and the correct direction, including inline opposite-direction runs.

Parity

Diff placeholder multisets

Six source placeholders against five target placeholders is a mismatch count of one, and one is enough to break runtime substitution.

Size

Characters are not bytes

Measure expansion in characters and storage in encoded bytes. Normalize to NFC before counting to avoid phantom differences.

Issues

Separate type from severity

Log error type, severity and location per issue, then derive the overall rating from those entries instead of typing it in.

Release

Re-run after fixes

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.

Measurements and typical failures

Preflight checks with the measurement each one produces and its typical failure.
CheckWhat is measuredTypical failureExample figure
MarkupProtected tokens and ids preservedToken translated or id changed1 missing token
Language markuplang and dir valuesSource language left on target page1 wrong lang value
Placeholder paritySource versus target placeholder countsOne placeholder dropped6 vs 5, mismatch 1
ExpansionCharacters against source lengthString grows past field cap1,200 extra characters
Expansion rateGrowth as a percentageUnderestimated budget for layout15%
Storage sizeEncoded bytes against limitCharacter count fits, bytes do notNFC normalization required
Quality issuesType, severity, location per issueType and severity collapsed into one field3 fields per issue
Preflight checks with the measurement each one produces and its typical failure.

Common objections

Isn't a visual read-through enough to catch translation defects?
No. A read-through catches wording problems but misses dropped placeholders, stale lang attributes and byte overruns, because those do not look wrong on screen. Compare token multisets and measure characters and bytes explicitly.
Why count characters and bytes separately?
ITS Storage Size sets a maximum storage size while Allowed Characters sets permitted characters, and the two limits differ once text is encoded. A string can pass the character allowance and still exceed the byte limit in UTF-8.
What does 15% expansion actually tell me?
Growth from 8,000 to 9,200 characters is 1,200 extra characters, or 15% expansion. Use it to check whether the target still fits fields with hard caps, such as meta descriptions or database columns.
Why keep error type and severity in separate fields?
ITS Localization Quality Issue records the nature and severity of a language-oriented QA error, and the two answer different questions. Keeping them separate lets you triage a placeholder mismatch in a checkout button differently from the same mismatch in a footer.

Standards referenced

References are limited to the W3C ITS 2.0 specification, its implementation guidance and the HTML5 translate attribute.

How do I mark textthat must not beHow do I checklanguage, directionHow do I compareplaceholders andHow do I measurecharacters, bytesHow do I logquality issuesWhat does acomplete preflight
A map of this guide's sections