Skip to main content

Quality assurance

Three complementary QA layers are available; use as many as the job warrants.

1. Automated QA scoring​

Run QA scoring across a file to get per-segment quality scores combining:

  • Metric-based checks — edit distance and similarity metrics against references.
  • LLM-based evaluation — a model assesses adequacy and fluency of each translation.

Scores surface in the editor so linguists can prioritise the weakest segments instead of reading everything.

2. XLIFF QA checks​

Rule-based checks catch mechanical errors that humans skim past:

  • Inconsistent translations of identical sources
  • Number, tag, and placeholder mismatches
  • Terminology violations against the attached termbase
  • Length, punctuation, and whitespace issues

Checks are configured on the QA Settings page through QA Check Profiles — named configurations you can create per content type or client (with a default profile), viewing and editing the individual checks each profile runs.

The same page holds editor behaviour settings that affect QA workflow:

  • Auto-propagate 100% TM matches (and separately 101% context matches)
  • Run QA checks on segment save — instant feedback while editing, instead of waiting for a full QA run
  • Show cross-language matches by default

QA settings

3. LQA (Linguistic Quality Assessment)​

For formal quality programmes, run an LQA pass: errors are categorised and severity-weighted (MQM-style), producing a scorecard per file that you can use for vendor management or client reporting. Linguists log LQA errors directly on segments in the editor.

Where results appear​

All three layers feed the editor: segment rows show scores and flags, and filtered views let reviewers work through only the segments that failed a given check.