We've received numerous complaints, across all browsers and platforms, of ePub files for works download as zip files, or simply opening in the browser as gibberish:
It appears that if the ePub is not the first file generated (e.g. the HTML file exists), the content type is incorrectly set to text/plain:
It should look like:
This is caused by the file cache. Because we have other types of caching, we'd like to remove this one.
The problem isn't reproducible outside production, so just make sure downloads still work on staging, and then we'll see if this resolves the issue on production.