The shape of the work has changed
Translation used to be one purchase: a linguist produced a file, someone proofread it, and the proofreading was folded into the price.
That is not how content is produced any more. An engine produces the first draft. Software scores each segment for risk. Only the risky segments reach a person. The human contribution has moved from producing text to judging it — which means the human contribution now has to be specified, priced and evidenced separately, because it is no longer automatic.
There are only three things anybody can do with machine output: change it, judge it, or prove that somebody qualified looked at it. The first two services below change it. The ones above them judge it, and hand you evidence rather than a corrected file. The last runs the part of your own quality process you would rather not run. What separates them is what is in the box and how it is priced.
Judging AI output
Working on AI output
How we price
| What you are buying | How it is priced | |—-|—-| | Independent Quality Review | Per hour, or per sample of an agreed size | | Exception Review | Per segment, or per hour | | Engine Fitness Report | Fixed fee per study | | Terminology and translation memory work | Per hour, or per term | | Linguist testing and evaluation | Per test piece | | In-country review management | Per hour, or as a management fee | | Post-editing | By revision band, with a quality floor clause | | Specialist translation | Per word or per hour, depending on the content | | Model output evaluation | Per task or per hour |
We price quality work by time, by sample or by decision rather than by word, because per-word pricing assumes machine output quality is constant across a file — and it is not. A clean segment and a badly mangled one would pay us the same, which penalises exactly the work you are sending us. Happy to discuss.
What they have in common
Two stages, two people. Editing and sign-off are separate passes performed by different linguists. Nobody signs off their own work.
Named reviewers with recorded qualifications. We record qualifications, subject expertise and time in domain for every linguist we assign. In regulated work you should ask for it, and we will produce it.
Structured scoring. MQM error categories and severity levels, so quality is a number that moves over time rather than an opinion that varies by reviewer.
A version trail. What changed, when, by whom, against which source version.
Independence. We do not sell an engine, so we have nothing to defend when we assess one.
Working with us
- A company, not a freelance pool. One supplier, one contract, one point of accountability.
- Priced by the hour or the segment. Per word only on translation from source.
- Reviewers are named, and we build the qualification record as part of the job. Ask for it and you get it.
- Delivered in our format, or staffed into your platform and run under your rubric.
- We hold no ISO certification today. ISO 17100 and ISO 18587 are in progress; ISO/IEC 42001 is under evaluation.
- We commit language pairs individually rather than advertising a number. Ask us about the pair you need.