How we measure translation fidelity.
Every translated document gets a fidelity score from 0 to 100. It is not a grade of writing style. It measures whether the structure of your document survived translation: formatting, numbers, links, and completeness. Here is exactly how it is computed.
Five structural checks, one weighted score.
Each translated segment is checked against its source. The five component scores below are combined with fixed weights into the final 0–100 number. The weights reflect how damaging each failure is to a finished document.
Completeness
How much of the document actually made it through translation. Untranslated segments are the worst fidelity failure, so this carries the largest weight.
Markup preservation
Bold, italics, inline styling, and structural tags should survive the trip. We compare formatting markers between the source and the translation, segment by segment.
Number & link integrity
Figures, dates, URLs, and email addresses must come out exactly as they went in. Any segment where numbers differ or links are dropped counts against this component.
Length sanity
Translations naturally expand or contract, but extreme swings usually mean something went wrong. We flag segments that are unusually short or long relative to their source.
Untranslated detection
A segment that comes back identical to its source in a different-language run probably was not translated. We detect and count these.
What the score means in practice.
The structure survived intact. Spot-check the output, especially names and terminology, and ship.
Most of the document is fine, but some segments tripped a check. The warnings list tells you exactly which kind, so review those areas first.
A meaningful share of segments has structural issues. Open the document review view, work through the flagged segments, and re-export.
The score is computed when a translation finishes and recomputed whenever you edit a segment, so it always reflects the current state of your document. Alongside the number you get a warning list naming each issue type and how many segments it affects, so you know where to look.
Warnings that ride along without moving the score.
PPTX overflow risk
Slides have fixed text boxes. When a noticeable share of translated segments in a presentation come back with length anomalies, we warn you that text boxes may overflow. Check the deck before presenting.
Right-to-left output
Arabic, our right-to-left target language, flips the reading direction. Exports are rebuilt with real RTL markup, and we flag the run so you remember to verify text direction and layout in the exported file.
What the score does not measure.
Fidelity is a structural measure, not a judgment of translation quality or style. A document can score well and still benefit from a human read for tone, terminology, and nuance. The built-in review workflow supports that final read.
We would rather give you a precise, verifiable number for the things a machine can actually check than a vague promise about the things it cannot.
More about the productSee your first fidelity score today.
Upload a DOCX, PDF, PPTX, or XLSX, translate it into any of 13 languages, and get the score with every run.