Earlier quoted context omitted.
I've heard people complain about Calibre's EPUB conversion utility, but I've never seen anyone do a deep-dive to explain exactly why it's bad. Is it comparable to the HTML-soup that Microsoft Word spits out when you export a Word doc to HTML? FWIW, I still heed those warnings, and so I still hold on to the de-DRM'd AZW file even after I convert it to EPUB.
For me the most obvious mangling the converting process is doing is inconsistent paragraphing, some times a single paragraph in original splits into two at an arbitrary place after conversion, and vice versa, seriously disrupting the readability.
There are two reasons I can think of for why you'd get broken paragraph separation. First, the markup in the kindle version (if it's old style mobi, or really badly generated kf8) could be broken, so when calibre tries to split into multiple files to prevent single (x)html files from being too big, it can't tell where paragraphs are so it might split in the middle of a paragraph.
Second, you might be converting a pdf or azw4 (essentially a pdf) to epub, in which case you'll get paragraph breaks across pages. You shouldn't convert those formats to epub without extensive editing.