AI tool to correct epubcheck and daisy/ACE epub issues

[crdjm]

Forgive me if this is the wrong forum to ask this.

I've been experimenting with a tool to automatically fix epubcheck and Daisy ACE errors in epubs. Also, to map epub2 to epub3 format (as a precursor to fixing issues). Cleary leveraging the rapid rise of AI technology.

This is largely a hobby for me at the moment -- not sure if the technology is quite there yet, but having been involved with ePub generation for ~25 years, across many publishers workflows, it seems to me like we are getting closer.

My tests are working pretty well at the moment, so I'm looking to expand my test-base to stress it more and see if there's any interest out there.

This is a hands free tool -- you provide an ePub, if epub2 it creates an epub3. If an epub3, it runs epubcheck and ACE, then works through the issues found and fixes what it can, creating a new epub and a detailed report of what was fixed, how it was fixed, and if not, why not. It will optionally also re-run epubcheck/ace on the new ePub to allow easy review of the improvements.

Does that sound useful? If so, please see epubs.online for a few more details and/or contact me at crdjm1@gmail.com. As noted, I'm really looking for some test cases, with issues, that I'd run as test cases.

If I've broken some forum rules here, I'm sorry. As noted, not sure as yet where I'm taking this, just trying to move it a long a bit for now.

Many thanks,
David.

[Quoth]

Learning how epub, css and HTML works and fixing manually is best.

AI tech isn't having a rapid rise. That's a lie from people wanting to monetise LLM type systems.

Also many people don't want epub2 changed to epub3 by a checking tool.

[Quoth]

See https://www.theregister.com/2025/09/...ctivity_gains/

Meanwhile, another recent study from nonprofit research group Model Evaluation & Threat Research (METR) found that AI coding tools actually made software developers slower, despite expectations to the contrary, because they had to spend time checking for and correcting errors made by the AI.

[JSWolf]

Originally Posted by Quoth

Learning how epub, css and HTML works and fixing manually is best.

AI tech isn't having a rapid rise. That's a lie from people wanting to monetise LLM type systems.

Also many people don't want epub2 changed to epub3 by a checking tool.

Sigil has a rather good plugin to go from ePub2 to ePub3.

[crdjm]

Originally Posted by Quoth

Learning how epub, css and HTML works and fixing manually is best.

AI tech isn't having a rapid rise. That's a lie from people wanting to monetise LLM type systems.

Also many people don't want epub2 changed to epub3 by a checking tool.

100% agreed on first comment. Having spent many years doing that, can't agree more. But some folks have many epubs and don't have many years to spend and want a helping hand to get them started. But maybe I'm wrong, hence my post, so I value your feedback.

Not completely agreed to 2nd point, it (AI) has its place. Over-hyped, perhaps. A place, certainly.

Agreed on epub2 comment. So, if you don't want it, don't use it. If you do want all that epub3 offers, automation to give you a head start, at least, would seem useful.

[crdjm]

Originally Posted by JSWolf

Sigil has a rather good plugin to go from ePub2 to ePub3.

Agreed. The epub2 mapping piece was a sideline really avoid me having to try to fix issues in epub2 and just focus on epub3. If you convert from 2 to 3 via other, better, existing means: great.

[Quoth]

Originally Posted by crdjm

Not completely agreed to 2nd point, it (AI) has its place. Over-hyped, perhaps. A place, certainly.

Some things labelled AI are very useful.

Originally Posted by crdjm

Agreed on epub2 comment. So, if you don't want it, don't use it. If you do want all that epub3 offers, automation to give you a head start, at least, would seem useful.

Conversion of epub2 to epub3 on it's own is almost pointless. A human has to author the epub3 extra features.
There are different applications and features of epub3:

• A simple conversion to epub3 from epub2. There are no extra semantics, no interactivity.
• Fixed layout that's really just an alternate to PDF.
• Adding more semantic information, such as figure, section etc. Makes little or no difference to most rendering.
• Adding interactive or dynamic content via javascript coding *
• Adding audio media (works on some eink) *
• Adding video or animation (rubbish on eink) *

Only the fixed layout conversion can be done by automated means, once additional conversion settings similar to conversion to PDF are added.

The items with * are akin to multi-media authoring, but using epub3 instead of some other system.

A straight conversion from epub2 to epub3 may no effect on appearance and should still work on an epub2 renderer, as epub3 simply adds an extra different navigation system etc.

So to take any advantage of epub3 a human needs to author. This can be done in Calibre, though it's more suited to conversion or editing or fixing existing content. Sigil is a more flexible authoring tool.

So any tool offered to do checking should be able to check epub2 or epub3 and never convert. Conversion needs human authoring to be worthwhile, so needs Sigil, Calibre or any of the more expensive and often poorer commercial tools. I'd only use InDesign for PDF or Fixed Layout epub3.