The funny thing is that if this worked well, I would just convert the ePubs back to PDFs using
Jellby's script. But I doubt it does. The intermediate steps are a good sign, I must admit. But I'm surprised they don't offer even a full example: couldn't they take a public domain title, and let you download both the input and output to compare? Of course, demo version would be better, but a sample would at least be something.