Moving a site or a company's documents to the new alphabet

This page is for anyone who works with text: a newsroom, a publisher, a company, a state office. The order is cheap and useful first, expensive last.

1. Text is the easy part

Articles, documents, subtitles. txt, md, csv, srt, vtt and docx files can be dropped onto the Files page several at a time and taken back converted. It all happens in the browser: a working document never reaches a server.

Inside a Word document the text is edited in place — the file is not rebuilt. Styles, tables, images, headers and numbering stay exactly as they were.

2. Search — where most people break it

The database holds Toshkent, the user types Toşkent, and there are no matches. There is one fix: normalise both the query and the index to one form before comparing. Which form matters far less than that it is the same one.

Do this before converting the database, or the site will look like it works while finding nothing, and nobody will understand why.

3. Collation

With default sorting ö and ğ end up after z, because their code points are higher. JavaScript needs Intl.Collator("uz-Latn"), PostgreSQL needs COLLATE "uz-Latn-x-icu". Forget it and alphabetical lists quietly come out in the wrong order.

4. Old links

If your addresses contain words — /maqola/toshkent-shahri — leave them alone. Change the address and the old links die: your place in search, links from other sites, people's bookmarks.

If you convert them anyway, add a permanent redirect from old to new and keep both alive.

5. Text in the database

Converting the database once is the most dangerous step, because it is hard to undo. The order:

1. A backup. Not later, now.
2. Fix search and collation first — points 2 and 3.
3. Run incoming text through normalize("NFC"): ö arrives both as one code point and as o plus a combining diaeresis.
4. Start with a small slice and look at the result with your own eyes.
5. Check proper nouns separately: Chelsea must not become Çelsea.

6. Fonts — nothing to do here

All four letters come from the Turkish and Azerbaijani alphabets and have been in every system font for thirty years. No fonts to buy, no software to replace.

If your site ships its own web font, check that it includes the latin-ext subset: otherwise the letters fall back to another font and the text looks ragged.

7. A button for the reader

One line of code on your site gives an article a "show it the new way" button: a reader sees the text in the new alphabet in one tap and puts it back in one. That is far cheaper than converting a database and undoes instantly. See the Widget page.

About deadlines

The law takes effect once the president signs it, and the sector deadlines will be in the implementation programme, which is still being written. So there is no rush — but search and collation are cheaper to fix now, before a converted database lands on top of them.