Quoting a job you haven't read

Every translation job begins with a piece of work nobody pays for: reading enough of the file to know what you are agreeing to.

It is a real half hour. What kind of document is this — a contract, a set of accounts, a patent, a medical report? How much of it is boilerplate you have translated forty times and how much is genuinely new? Is it a clean digital file or a scan of a fax? Does it reference other documents you are not being sent? Is there a defined-terms section that will govern every choice you make for the next three days?

Get it right and the quote holds. Get it wrong — usually by underestimating how much of a scanned document is unusable, or how much specialist terminology sits in the back half — and you have priced three days of work at one day's rate, and you will still deliver it, because your reputation is the business.

The scoping questions are always the same

Which is exactly what makes them worth writing down once:

• What kind of document is this, in the sender's own words rather than the file name's
• What is the subject domain — the thing that decides whether you take it at all
• Are there defined terms, and where are they
• Which languages actually appear, including the ones nobody mentioned in the email
• What does it reference that has not been sent — schedules, annexes, prior agreements
• Is any of it handwritten or stamped, which is the single biggest source of a bad estimate

Ask those of one document and you have done the half hour. Ask them of a set of nine documents that arrived together, and the half hour is most of an afternoon — which is the situation that actually costs agencies money, because the client wants the quote today.

One row per file

This is the shape that suits it: the documents are the rows, the scoping questions are the columns, and the answers arrive as a table you can read down rather than nine files you read in sequence.

Ragextract has no built-in table for translation work, so the columns are yours to write. They are plain-language prompts rather than a configuration format, which matters here more than in most fields — you know how to ask "does this document define terms, and where?" far better than any preset would, and you can sharpen the wording once you have seen it get one answer wrong.

Two features of the output are worth knowing before you try it.

Every answer carries its source. A cell saying the document contains a defined-terms section tells you which page it is on, so scoping stops being a claim you have to accept and becomes a pointer to the part of the file you should look at yourself. For a job you are about to price, that is the difference between a summary and a shortcut.

It reads what is there and nothing else. The engine will not search the web, will not look up a term, and will not consult a glossary it was not given. So it cannot tell you what a term of art should be rendered as — it can only tell you that the document uses it, and where. That is a real limit and also the reason the citations mean anything.

What this is not

It is not a CAT tool. No translation memory, no segment alignment, no glossary management, no bilingual export. It does not touch the tools you work in and does not want to.

It does not translate. Worth saying plainly. It reads documents and answers questions about them in a table; producing the target text is your job and stays your job.

It does not count words for billing. Do not price off it. Use it to decide what the job is, then count with whatever you already trust.

It has no opinion about quality. It will not review a translation, will not compare a target against a source, and will not flag an inconsistency between two of your own deliverables.

If that list reads as narrow, it is meant to. The claim here is a small one: the unpaid half hour at the start of a job is repetitive, and repetitive reading against a fixed set of questions is the one thing this does well.

Where to start

Next time a client sends a batch and wants a number by end of day, put the whole batch in one table with your six scoping questions as columns. Do the half hour properly on one file to check the answers, and see whether the other eight rows saved you the afternoon.

Then keep the table. The questions do not change between clients, which means the second time costs nothing.

• Writing columns as questions
• What a citation is
• What a run costs