At Streamrise we publish guides for streamers on an English site and a Russian site. Our rule is blunt: we never ship a placeholder translation. A new guide goes live in English only, and it sits on an explicit English-only list. That list keeps the page off the Russian domain completely: no direct URL, no index entry, no sitemap line. It is there because we once had pages whose "Russian" version was just the English text copied over, and that leaked English onto the Russian site. An honest gap is less confusing for readers than a page that looks translated and isn't.

Our source text moves too, just not ours. Our guides quote platform help centres, and those pages get edited without notice. This month, two of the Twitch help pages we cite were republished days before we used them. So every citation in our recent guides records the date we read it. When a reviewer checks a quote, they compare it against that version, not against whatever they remember.

The example that reduced the most confusion was a translation mistake, not a timing one. A draft quoted a Kick help article "word for word" in English. The article only existed in Arabic. The draft had translated it and presented the translation as the original wording. We caught it before publishing. The rule we took from it: a translation is never placed inside quotation marks as if it were the source. If a source exists only in another language, we describe it and say so, or we don't use it.

For interface and help text specifically, my advice is to split by risk rather than by schedule. Anything a user must act on correctly, such as a setting name, a number or a deadline, gets pushed to every language now, or the page waits. Explanatory text can wait for the next batch. The rework comes from mixing the two.

I sort changes by whether the meaning moved. If an edit changes what the user is told to do, or what a food term points to, it goes out now in every language. A stale translation is a wrong instruction. If it's a wording polish, it waits for the batch. Reviewers hate polish edits landing mid-cycle, because they re-read a string that hasn't really changed.

The rule I hold to is that no string travels without a note on where it appears and what it refers to. Food is the reason. A word like once or merienda looks like a plain label, but the meal it names shifts by country, so a Spanish interface is never one Spanish. A translator who gets a bare string picks the most common reading. A reviewer from another country then flags it. One line of context settles that argument before it starts.

I also keep dish names out of the translation pass. Bandeja paisa and ajiaco stay as they are, and only the words around them get translated. That removes a whole class of review comments about whether a dish name should be translated at all.

When a late change is small, I'd send that one string on its own with its context note and leave the rest of the batch alone.

The question I ask about every mid-cycle change is whether it changes what the reader has to do.

If it changes a step, a number, a warning, or a button name, it goes to translation right away, because a wrong instruction in every language costs more than a late one. If it only polishes wording, fixes a typo, or moves a sentence, it waits for the next batch.

What cuts rework most is the change note. I write one line per change and label it "meaning" or "polish," so translators and reviewers can see at a glance what needs a fresh look. Short strings help too, since a one-word edit then touches one string instead of a whole page.

My background is in technical writing, UX content, and information architecture, and I currently build Pact, an AI-powered contract review tool.

Decide by what the change touches, not by how big it is.

If the source change alters something a user acts on (a price, a limit, a step in a flow, a warning), push it to every language now: machine translation with a same-day native check. A stale translation there is a wrong instruction, and that costs more than an imperfect sentence.

If the change is wording, tone or layout polish, let it wait for the next batch and leave the old translation live. It is still correct, just less elegant.

The rule that tends to cut rework the most: whoever edits a source string tags it "meaning" or "polish" at the moment they change it. Translators stop re-translating the same string three times in one sprint, and reviewers stop receiving batches where half the edits are cosmetic.