How often do publishers change/tweak articles to achieve visually appealing results?

Question

When reading this question it seems the theoretical answer is that LaTeX is not perfect and manual/human intervention is required. In the TeX Book Knuth says:

But the problem of page make-up is considerably more difficult than the problem of line breaking that we considered in the previous chapter, because pages often have much less flexibility than lines do. If the vertical glue on a page has little or no ability to stretch or to shrink, TeX usually has no choice about where to start a new page; conversely, if there is too much variability in the glue, the result will look bad because different pages will be too irregular. Therefore if you are fussy about the appearance of pages, you can expect to do some rewriting of the manuscript until you achieve an appropriate balance, or you might need to fiddle with the looseness as described in Chapter 14; no automated system will be able to do this as well as you.

In the LaTeX Companion Mittelbach writes:

For this and other reasons, getting the final layout of the book was fairly labor intensive and even required minor rewriting (on maybe 10% of the pages) in order to avoid bad line breaks or page breaks.

Ignoring the fact that LaTeX is not the only software that publishers use, I am curious if the theoretical answer is actually the practical answer. Specifically, I am interested in:

1. How often do journals make manual tweaks to their templates/layouts when typesetting articles?

2. How often wording of the text is changed to increase the visual appeal of the article?

Answer 1

Since some journals provided me not only the final typesetted result but also additional documents that captured the editing (for example a scanned document with the remarks or the copy editor or even a diff of the "source code" (actually a mix of text and source code for the formula)) I can say, based on this sample size of about a dozen: Always.

To be more precise: The text has been changed always, but I can't really tell if this was for better visual appearance or because of the text. Also the figures have been shifted around in all cases (and I guess that this was for the visual appearance). I can't say whether the journal tweaked their template (because I don't care too much about the templates in the first place).

Maybe I should add that I have been more or less always pleased with the editing. In most cases I did not notice the changes from the typesetted version, but only after a closer look at the diff. At a few places the editing introduce a wrong wording (most cases there was a note to check this passage) and in some cases the typesetting of the formula was worse.

Answer 2

A partial answer because I'm not a typesetter (i.e. I can only base off my memory of the un-typeset article vs. the completed one), and it's been a while since I personally handled production of journal articles.

The answer to question 1 is very often, perhaps always. Completed articles always looked different from the original; however I also handled a lot of non-CRC articles so it's expected that the completed one looks different. Still even with CRC papers they look different. Figures move, referencing style changes, a license is added to the first page. I also remember a typesetter telling me they run checks. I didn't ask for details, but I know that one result is that if there's a referencing error (e.g. "see Ref. (??)" which is moderately common) it's caught and fixed.

The answer to question 2 however is almost never. It's dangerous to heavily modify the text without the author's approval, plus my experience is that typesetters are almost always able to fix issues without changing the text. How they do it I'm not sure, but I'm regularly able to say things like "hey, the left-most word of these three lines are the same, improve it" and they'll manage without changing any word. I also remember two relevant incidents, both for books not journals:

1. One was an equation overrun. The text nearby was already too tight and the only way to fix it was to add a new line, which would however move every later page. If we'd noticed this earlier in the production process it would've been fixable; as it is the book went to print with the flaw (damn).
2. The other was a case where the author wanted to typeset the entire book himself. I remember I spotted a serious issue in the completed PDF (don't remember exactly what - could've been a figure caption that split across two pages) and the author rewrote the text to eliminate the issue. We eventually persuaded him to give us the source files, and the typesetters fixed the issue without rewriting. I don't know how they did it.

You might also be interested in the production timeline. If I gave a normal-sized manuscript (up to 20+ pages) to the typesetters, they'd usually give it back in 3-4 working days. If I said it was urgent, they'd usually return it in one day. This is imperfect data however: it's possible they pushed down the time taken by e.g. having two typesetters to the manuscript. Not having worked in the typesetting department, I don't know the internals.