How it works & what we log
The pipeline
Your book is split into passages. The split follows paragraph boundaries; markup, images and layout are never touched.
Every passage is translated using a translator's brief written for your book. Then it is read again, independently, against the original. This second reading does not rewrite anything. It reports what it found: text that is missing, text that was added, meaning that is wrong. Every passage it marks is translated again, with the finding attached, and checked again.
When all passages are done, the finished translation gets one whole-book read, end to end in one go, with the original next to it when both fit. This finds the problems that no single passage can show: a name or a term translated differently in different chapters, a repeated line that no longer matches, a change of tone. What this read marks is translated again and goes through the same checks. Books too long to read in one go skip this step. Your report says so, and you are not charged for it.
Your quality report is the difference between the first translation and the one you receive. The score comes from that difference. It is measured, not invented.
There is one thing a translation may add. Where the author uses a
unit that is uncommon in your language (miles, °F), the translator may add the
local equivalent in brackets. The author's figure always comes first and is
never changed, for example 350°F (175 °C). Nothing the author
wrote is ever replaced.
The price is exact, not estimated
We open and split your book before quoting, so the price is simple arithmetic: number of characters → model tokens → the steps above → price. The quote is saved before you pay. If your book ends up costing us more than we quoted, we pay the difference. There are no charges per page and nothing is added later.
File lifecycle
- Your upload and the finished files are kept in private storage. They have no public address, and they are only sent back through your own link, which cannot be guessed.
- Deletion after 30 days is a storage policy. The storage system deletes the files itself. It is not a script we could forget to run.
- Your books are never used to train AI models. Not by us — we train nothing — and not by the company whose models we use, under the terms we work with.
What we log
Job records (language pair, word count, status, quality score), payment records from Stripe, and normal web-server logs. We run no advertising trackers, and the pages work without JavaScript, apart from checking progress.
One cookie of ours records which page of this site you arrived on, and which site sent you here. That tells us which of our own pages are worth writing. It holds a page address and a site name. It holds nothing that identifies you, nobody else can read it, and it expires after 30 days.
What AI translation does badly
We would rather say this than sell around it. Rhymed poetry, dense wordplay and strong dialect lose the most in any machine translation, including ours. Our checks catch wrong meaning; they cannot rebuild a rhyme. Normal prose, both fiction and non-fiction, is where this works best. Your free sample exists so that you can judge this on your own book, before you pay.
Upload rights and takedown are covered in the terms.