Most AI translation tools give you a result and move on. You have no indication of quality — whether the translation is excellent, acceptable, or contains an error that could cause you a serious problem.
Translova adds a confidence score to every translation. This is not a marketing feature. It is the output of a real multi-pass validation process, and it is designed to give you the information you need to decide whether a translation is reliable enough for your purpose.
The problem with translation quality
Translation accuracy is not binary. A translation can be broadly correct in meaning while containing terminology errors that matter greatly in a legal or medical context. It can be fluent and natural while missing a specific date or reference number. It can be technically accurate while using informal language where formal administrative language is required.
For official documents — visa letters, employment contracts, medical records, tenancy agreements — these distinctions matter. An immigration caseworker reading a translated document will notice if the terminology does not match what they expect. A landlord reading a translated reference letter will notice if the tone is wrong.
The confidence score reflects how well the translation avoids all of these problems.
How the score is calculated — the five-pass pipeline
Pass 1 — Translation
The source document is translated from the original language to the target language. The AI uses formal administrative language appropriate for the document type and country context selected. For example, a French employment contract translated for UK use will use UK employment law terminology rather than French-English literal translation.
Passes 2 and 3 — Explanation and quality validation (run simultaneously)
Two separate processes run in parallel after Pass 1 completes.
Pass 2 generates the plain-language explanation — the four sections that tell you what the document is, what you must do, any deadlines, and the risk of inaction. This runs simultaneously with quality checking to save time.
Pass 3 is the quality validation. A completely separate AI reviews the original document and the translation side by side. It checks five things:
1. Accuracy — Does every sentence in the translation accurately reflect the source? (40% of score)
2. Terminology — Are legal, medical, or administrative terms correct for this document type in the target language and country? (30% of score)
3. Completeness — Are all dates, names, reference numbers, amounts, and addresses preserved exactly? (20% of score)
4. Fluency — Does the translation read naturally in the target language? (10% of score)
The validator uses a strict problem-counting rubric and is explicitly instructed not to assign a default score. It counts problems:
Pass 4 — Automatic refinement if below 95%
If Pass 3 returns a score below 95%, Pass 4 runs automatically. A refinement AI receives three inputs: the original document, the first translation, and the specific list of problems identified in Pass 3. It produces a corrected translation that addresses each problem.
Pass 5 — Re-validation
Pass 5 re-scores the refined translation using the same methodology as Pass 3. The higher of the two scores is used. The result is what you see.
What each score range means in practice
95–98% — Professional quality. The validation found no meaningful issues, or found only extremely minor phrasing variations that do not affect meaning. This is the minimum standard for all paid Translova translations. Suitable for any purpose that does not legally require certified human translation.
90–94% — Very good quality. The validation found one or two minor issues — perhaps a slightly sub-optimal translation of a formal term, or a sentence that is accurate but reads slightly awkwardly. Suitable for most purposes. For critical legal decisions, consider reviewing with a professional.
85–89% — Good quality. Some terminology or structure issues were found. The meaning is conveyed but there are imperfections that could matter in formal contexts. This range should prompt careful review before acting.
70–84% — Acceptable for basic understanding but not for formal use. Problems were found that affect the translation's reliability for official purposes.
Why free previews show a lower score
The free 300-word preview translates using only Pass 1 — no validation, no refinement. The score shown for a free preview is an initial assessment only, not a validated score.
This is intentional. The free preview is to help you understand what a document is about and decide whether to pay for the full translation. The investment of validation time and additional AI passes is made for paid translations.
The 95% minimum commitment
Translova will not deliver a paid translation below 95%. If the first attempt comes back at 91%, Pass 4 runs automatically and the result is re-scored. If the refined translation reaches 95% or above, that is what you see. If it does not reach 95% despite refinement — which can happen with highly technical or unusual documents — you see the best score achieved and the validation notes explaining what issues remain.
We show you the real score. We do not inflate it.