Products
The archive is the first product. Every claim has one page: the claim, the newsrooms that checked it, their verdict in their own words and a link to their article. Search, filters, themes and dossiers lead to it.
Two products will follow that do not exist yet: an interface that lets a newsroom or a platform ask whether a claim was already checked, and a browser extension that shows this next to what you read.
For developers
Every claim keeps its number, in every language of the site. The address of a page does not change, so you can link to it. The pages are plain HTML and work without scripts.
Every edition is one file, built from sources we may republish. The sources and licences page names them. A public interface to query the archive is planned; it does not exist yet.
Under the hood
- We take fact-checks from sources that allow it, and name the source and the licence of every record.
- When several newsrooms checked the same claim, their checks share one page.
- A newsroom's verdict is shown in its own words. We sort verdicts into a small set of labels. Factiz adds no verdict of its own.
- A language model translates claims. We read a sample per language first, and the original always stays visible.
- A language model files each claim under a theme. We read a sample of that first as well.
- A language model suggests which fact-checks belong in a dossier. A second model reads every suggestion and opens the fact-check itself when the claim alone is not enough. A person decides every case in doubt and where the dossier ends, and approves it before it goes online.
Documentation
Where every record comes from, and under which licence, is on the sources and licences page. To cite a page, use its address: it stays the same. More documentation will follow with the interface.