Quality checks
The automatic checks and the quality score every translation goes through.
transglot checks translations in two layers: automatic checks for problems that would break your product, which run on every plan, and an AI review that scores each translation and lists specific findings, on Team and above.
Automatic checks
These look for problems that can be detected with certainty, and they run without spending any of your usage:
| Check | What it catches |
|---|---|
| Placeholders | A placeholder such as {name}, %s or an HTML tag that is missing, added or changed. |
| Plural forms | A plural key missing one of the forms the target language needs. |
| Length | A translation longer than the key's maximum length. With screenshots, a translation likely to overflow the space it renders in. |
| Glossary | A translation that departs from a glossary term, uses a forbidden word, or translates a term marked Do not translate. |
Glossa's drafts go through these checks as they are written. A draft that still breaks a placeholder, a plural form or a forbidden glossary term after a retry is not written: the string stays untranslated and the run reports it, rather than shipping a broken string. A draft that only breaks a softer rule, such as the maximum length, is written as a draft; with Route AI translations by confidence turned on, it is flagged for review. See Translation states.
How placeholders and plurals are read depends on the project's Message syntax setting, and how strictly glossary departures are treated depends on Glossary enforcement. Both are on the project's Checks page.
AI review and the quality score
On Team and above, Run QA in the editor, or Re-run checks on the project's Quality page, has Glossa review the translations against their source.
The quality score is a 0-100 rating that transglot produces for a translation, alongside a list of specific findings, each with a severity. The scorecard groups those findings using the industry-standard MQM categories so they can be compared with other tools' reports.
Findings are either blocking or warning. A translation with a blocking finding, or a low score, is marked needs checking. A run is incremental: it scores what has changed since the last run, so running it again over an unchanged project does no new work. Reviewing translations counts towards your usage; see Words and usage.
The Quality page shows every language's score and its findings, and the Scorecard shows the per-language MQM scores you can print or export. See The per-language scorecard.
Self-heal
On Team and above, Self-heal retranslates machine translations that fail review. The project's Auto-heal mode decides whether a fix is offered as a suggestion for a reviewer to accept (Suggest) or written in directly (Apply). Source drift repair does the same when a source string changes, re-checking its translations and repairing the ones the edit broke. Neither ever overwrites a translation a person has edited or reviewed. See Self-heal a translation.