QA checks¶
Every new translation is checked before it’s stored: engine results, translation memory hits,
repeats within a job, and fixes made in the review queue. Each finding has a code, a
severity and a message.
Errors hold the translation back. Its state becomes
attention, and it waits in the review queue.Warnings travel with the result.
GET /resultsincludes them inqa, and connectors show them to editors.
Checks¶
|
Checks |
Severity |
|---|---|---|
|
For |
error |
|
An image’s |
error |
|
The same URLs, the same number of times: |
error |
|
Placeholders survive exactly: |
error |
|
The translation is empty although the source isn’t (for |
error |
|
The translation is longer than the segment’s |
error |
|
The translation is less than 0.3 or more than 2.5 times as long as the source (only for sources of at least 20 visible characters). |
warning |
|
|
warning |
|
A do not translate term from the source doesn’t appear unchanged. |
error |
|
A glossary term’s translation doesn’t appear. Matched as a substring, ignoring case, so inflected forms count. |
warning |
|
The translation uses a forbidden word from the brand voice (whole word, ignoring case). |
error |
“Visible characters” means the text without markup, for html and rich_text.
Limits of the checks¶
The HTML check compares which tags occur how often, not the tree structure. A tag moved to another parent passes.
Glossary and forbidden-word checks compare strings. An inflected form can hide a forbidden word, or satisfy a glossary term loosely.
Texts that editors approve in the CMS or admin aren’t checked: editors are in charge there.
Example¶
{"code": "seo", "severity": "warning", "message": "171 characters; search engines show about 160."}