A polish that improves twelve sentences in a fifty-paragraph article can still leave the reader worse off than no polish at all. The improved sentences read smoothly. The paragraphs around them read a hair less smoothly. The reader notices the contrast, even if they cannot name it. They sense that two writers worked on this piece, and they are correct.
This is the failure mode we work hardest to avoid. The reviewer’s job is to find sentences that need fixing. The translator’s job is to render the source in fluent Ukrainian. Our job, somewhere between them, is to make our intervention invisible.
The seam problem
When we open a translation flagged for polish, we have a list of specific sentences and phrases the reviewer marked. It is tempting to treat each flag as a self-contained editing problem: read the flagged sentence, find a better version, replace it, move on. After twelve substitutions, the article reads, on average, better than it did. By a sentence-level metric, the polish was successful.
But the reader does not consume sentences. They consume paragraphs, sections, an entire article. What they perceive is texture. If three sentences in a paragraph match a baseline level of fluency and the fourth is markedly more graceful, the contrast registers as something off. The “improved” sentence stands out like a freshly painted board in a weathered fence. The board itself is technically the best-looking board on the fence, and that is exactly the problem.
We have come to think of this as a seam between two writers: the translator’s voice on either side of our patch, and our voice in the middle. The seam is visible because the materials are different, not because either material is bad.
Matching the translator we are editing
The first instinct we trained ourselves out of was rendering each flagged sentence in our preferred Ukrainian. We have favorite sentence rhythms. We have a habitual register. We tend, left to ourselves, to write with shorter sentences than most theological prose calls for, and to favor concrete verbs over abstract ones. None of that is wrong. It is also not the translator’s voice.
When we polish now, we read the surrounding paragraph first and try to identify the texture of the translator’s choices. Are the sentences long or short on average? Is the register more formal or more pastoral? Does the translator favor nominalizations, or do they keep verbs active? Are subordinate clauses common, or rare? We then write the patch to fit that texture, not the texture we would have used.
This means our patches sometimes read like compromises. A sentence we wrote could be tightened. A construction we chose is not the one we would have picked from scratch. We leave it because the goal is not to optimize each sentence; it is to keep the article sounding like one writer wrote it. The patch should be invisible in context, not impressive in isolation.
What the reviewer’s silence tells us
The reviewer’s notes flag what needs to change. What they do not flag, they have implicitly approved. The unflagged ninety percent of an article is not a blank space we are free to improve. It is a tacit endorsement of the translator’s choices for those sentences.
We respect that endorsement. We do not retouch unflagged paragraphs because we noticed something we would have done differently. We do not add a comma here, swap a verb there, smooth out a transition that read fine to the reviewer. Every such change is an opportunity to introduce a new error, and more importantly, every change is a small shift in the article’s voice toward our own.
The discipline is uncomfortable. There is almost always something we could improve. The instinct that makes us useful for the flagged sentences keeps firing on the unflagged ones. Suppressing that instinct is most of the job.
When matching costs us the better sentence
Sometimes the right patch, the one that fits the translator’s texture, is genuinely worse than the patch we would have written if we owned the article. The match is closer to the surrounding prose, but the surrounding prose has limitations we did not impose, and our patch inherits them.
We accept this trade. A slightly weaker sentence that reads as part of one author’s work is better than a stronger sentence that breaks the article in half. The reader is not editing the article. They are reading it. Their experience is shaped by continuity more than by the local quality of any one sentence.
There are exceptions. When the reviewer flagged a sentence because it was outright wrong, the fix is the fix. A mistranslation, an omission, a theological term used incorrectly: these we change without worrying about texture. But for the more common case, where the reviewer flagged a sentence as stiff or unclear rather than wrong, the texture argument tends to dominate.
What this generalizes to
The pattern shows up wherever an editor works inside someone else’s voice. A code reviewer who rewrites a function in their own style, even when the rewrite is cleaner, leaves a codebase that reads as the work of many hands. A designer who tweaks a wireframe to match their own composition habits introduces friction the next person on the project will feel without being able to name it. The local improvement is real. The cumulative cost is also real, and it usually wins.
We think of polish as a service to the original work, not a performance of our own skill. The best polish is the one no one notices. The reader closes the article and remembers what it said, not how it sounded. They never wonder who wrote it, because the question never comes up.
The fingerprints we leave are how we know we have failed. The articles we polish well are the ones no one can tell we touched.