How we make sure your translation is actually good
You're not a translation expert — and you shouldn't have to be. 43 AI agents work like a full translation team, checking meaning, terminology, fluency, and technical structure, then giving you an objective score.
What gets checked
- •You got a translation back. How do you know if it's actually good? Now you can check with a number.
- •A high score isn't the whole story. The report shows exactly what's strong and what still needs work.
- •43 specialized AI agents check meaning, terminology, audience fit, technical structure, fluency, locale conventions, and tone — the same things a professional translation team would review.
- •Whether you're shipping a game, listing products, or localizing a SaaS app, the quality bar is the same.
- •The scoring is built on MQM, the international translation quality standard — translated into a score and grade anyone can understand.
How the quality score works
The score isn't about whether the translation "sounds okay." It's about finding actual errors, classifying how serious they are, and turning that into one clear number. Under the hood, this uses MQM (Multidimensional Quality Metrics) — the same framework used by professional translation companies and international standards bodies.
Quality Score = 100 − (Weighted Error Penalties / Word Count × 1000)
Each error gets a penalty based on severity: critical errors cost 25 points, major errors cost 5 points, minor errors cost 1 point. The framework follows ISO standards, so scores are consistent across any language pair.
The meaning is wrong, a number is incorrect, or essential information is missing. Think: a payment amount displayed wrong in your game, or a medication dosage that changed. Must be fixed before release. (Penalty: 25 points)
The meaning comes through, but a native speaker would definitely notice something is off — or the same term is translated two different ways in the same document. Like 'Settings' appearing as both '설정' and '환경설정' in your app. Shippable, but it hurts perceived quality. (Penalty: 5 points)
Nothing is wrong per se, but the phrasing could be smoother. Slightly awkward word order, a comma in an odd place. Worth polishing if you have time. (Penalty: 1 point)
7 areas checked by 43 AI agents
Translation quality isn't just "does it sound natural." A professional translation team checks seven core areas — and that's exactly what our AI agents do automatically.
Meaning preservation
Checks that the original meaning came through accurately. Catches missing content, reversed meanings, and subtle nuance shifts.
Mistranslation, omission, nuance distortion, false equivalence
Terminology consistency
Makes sure the same term is translated the same way throughout the document. Inconsistent terminology confuses users.
'Settings' translated as both 'Einstellungen' and 'Konfiguration', product name variations
Audience fit
Checks that the tone and complexity match the actual reader. You wouldn't use legal jargon for gamers or slang for doctors.
Expertise-level mismatch, cultural inappropriateness, wrong register for target audience
Technical structure
Verifies HTML tags, variables (like {name} placeholders), and formatting survived the translation intact. Broken structure means broken UI.
Missing tags, lost placeholders, broken formatting
Fluency
Checks that the text reads naturally in the target language. Grammatically correct but awkward still counts as an error.
Translationese, grammar errors, literal translation artifacts
Locale conventions
Verifies that dates, numbers, currencies, and units follow local conventions. The US writes $100; Korea writes 100달러 — details like this matter.
Date format errors (MM/DD vs DD/MM), missing currency symbols, unconverted units
Style and tone
Checks that the brand voice and document purpose are reflected consistently. Mixing formal and casual in the same document undermines professionalism.
Tone inconsistency, register mixing, brand voice drift
What the score means
| Grade | Score Range | What it means |
|---|---|---|
A+ | ≥ 98 | Ship immediately |
A | 95–97 | Good for publication |
B+ | 90–94 | Needs a second look |
B | 85–89 | Needs partial review |
C+ | 80–84 | Needs careful review |
C | 75–79 | Rework recommended |
D | 70–74 | Partial retranslation needed |
F | < 70 | Should be retranslated |
Ship it vs. Needs another look
Ship it
Ship it: target score met, zero critical errors, terminology is consistent. Good to go.
Needs another look
Needs another look: score below target, critical or major errors unresolved, terminology inconsistencies remain. We recommend fixing flagged issues and re-checking.
Why this matters for your content
"The translation looks fine to me" and "the translation is actually fine" are two different things. Without an objective quality standard, you're guessing — and guessing doesn't scale.
Game developers: verify localization quality yourself
Your Japanese and Chinese translations came back, but you can't read them. The score and error report tell you exactly where the problems are — no language skills required.
Ecommerce sellers: know your listings land
Product descriptions, review replies, CS templates — check translation quality with a number before you publish to international marketplaces.
SaaS teams: keep UI text quality consistent
Every string in your app, checked against the same quality bar. When terminology drifts, users get confused. This catches it.
Content teams: guarantee multilingual quality
Blog posts, help docs, marketing pages — verify that each language version reads naturally before it goes live.