Editorial Standards at Rosmerta Publishing

Rosmerta Publishing LLC is responsible for everything we publish. This page says how the work is made: where we use AI, how we restore old books, where the sources come from, and what we do when we get it wrong.

Who is responsible, and who writes here?

Rosmerta Publishing LLC is responsible for every page, book and entry we publish. Nothing is published without human editorial control. Articles are read in full before they go out.

Longer work is handled differently, because it has to be. A book passes through staged editorial checks, each one recorded, before it is cleared. Reference works running to thousands of entries are checked by reviewing the method that built them and reading a recorded sample of the result. That procedure is published, with a worked example.

We write under house pen names. Eoin Rosmerta, Lucy McDonnell and Janice Mathers each have their own voice and subjects. None of them is a real person. None has a biography, and we give them no credentials or qualifications. They are ways of writing, not people. The company is answerable for whatever they sign.

Do you use AI?

Yes. We use it for three things:

drafting articles
helping to read scans of old books
building our software

All of it is published under the human editorial control described above.

The middle one is worth separating out. We use AI to build tools, and the tools are what touch a restored book. A language model predicts words, and asked to repair a damaged page it produces something plausible, which is the wrong thing when a reader is trying to learn from it. So we use a model to help write the software instead: find-and-replace rules, parsers, checkers. Those programs are deterministic. Same input, same output, and they cannot invent a word that was not on the page. The procedure page shows this on a real book.

We keep a provenance record for each piece of work: a file saying how it was made, what it came from, who checked it and when. It is written while the work happens. Where one was written up afterwards, it says so on its face.

Three limits do not move:

AI does not decide what is true. The sources do.
AI does not write the historical texts we restore.
Nothing AI has drafted is published outside that human editorial control.

That record is what makes this checkable. Anything new carries one, and we check it against this page before publishing. If the sources are incomplete, the work waits.

Work published before we adopted this does not all have one yet, and we are not backfilling records from memory. A record written afterwards from what somebody recalls is worse than no record, because it looks like evidence.

How do you restore old books?

We photograph the pages of the original. Software then reads the photograph and guesses at each letter, a process called optical character recognition, or OCR. It was built for modern print, so on a Victorian book it makes mistakes, and it makes the same ones repeatedly.

We repair those with rules: precise find-and-replace instructions, one per repeated error. A rule cannot guess or invent, so we can say exactly what was changed. The full method is published, with a worked example.

Books restored under our current method log every change under one of four headings:

taken straight from the source
worked out from a bad scan
rebuilt under a stated rule
lightly copy-edited

Books restored under that method also carry a note on the text: a short section at the front telling the reader what we standardised, what we left alone, and why.

That method dates from May 2026. Earlier titles have shorter records, and our first two carry no note on the text at all. Every restored book has a record of some kind, and you can see which standard it was made to.

We keep dialect spellings, old grammar and the original paragraphing. We are reprinting these books. We do not modernise them, cut them, or rewrite them.

Where do the dictionary entries come from?

Four rules govern every entry:

It names its source, and the year that source recorded it.
Where sources disagree, we show the disagreement instead of quietly picking one.
Modern revivals are marked apart from historical forms.
We check a text is out of copyright before using it, and record why.

An entry without a source is not published. Some notes alongside the entries, like craft advice for writers, are opinion rather than fact. Those are marked as opinion.

Where does the artwork come from?

Art you look at as a subject, like a portrait or a scene, comes from museum collections, old editions out of copyright, or a human illustrator. Decoration like borders, textures and article headers may be AI-assisted. We log each AI-assisted image with its tool, licence and date.

Our mythology app has no AI-made character, god or scene art. Some of its borders and background textures are AI-assisted, and the store listing says so. We corrected that listing ourselves when we found the earlier wording claimed more than was true.

What about the ads?

This site carries advertising through Google AdSense. We choose where ads may appear and we exclude a great deal, including our sensitive entries, our reference tables and our app pages. We do not choose what any individual ad says.

Ads are not our writing, and the standards on this page do not extend to them. Two things follow from that, and both are worth saying plainly rather than leaving a reader to work out.

Some ad content is AI-generated. From July 2026 Google may show an AI label on an advertiser’s creative. That label describes the advertiser’s work, not ours. If you see one here, it is telling you about the ad it sits on.

Some ad formats also carry text generated by Google’s own model. Where that appears it has not been through our editorial process, because it is not ours to review. Our statement that nothing is published without human editorial control covers the articles, books and entries we make, not the adverts beside them.

If an ad here is misleading or offensive, tell us through the contact page. We can block individual advertisers and whole categories, and we will.

Where our interests lie

We sell things. Here is every way that could bend what we publish, set out plainly so you can weigh what we say.

We write about our own books. When a post recommends a Rosmerta title, we published it, we earn on it, and we chose what it says about itself. What we can do is keep the claims checkable. Every restored book states what we changed, so you can test the post against the record.

We earn commission on some links. Book links to Bookshop.org pay us a commission at no extra cost to you, and those links are marked. We link to books we would recommend without the commission, including books from other publishers and books that compete with ours.

We carry advertising. We choose where ads may appear and we exclude a great deal. We do not choose what any individual ad says. No advertiser has seen a piece of ours before publication or asked for a change.

What if you get something wrong?

We fix it, and the fix stays visible. Each restored book carries its note on the text, so a reader can see what was changed and why. Behind that we keep a fuller restoration record for every title, and we will share it on request. If you report one error, we check everything made alongside it, not just the one entry. Tell us through the contact page. Corrections are welcome.

References

For reference. This is not a statement of our legal position.

Responsible: Rosmerta Publishing LLC. Effective 11 August 2026. Reviewed yearly, and whenever our practice changes. Questions through the contact page.