Documentation · Updated 13 November 2026
How NewsFacts works
From the moment a story is published to the moment a change appears in the public record, six things happen. This page walks through them in order.
1. We notice a new story
Every outlet we follow publishes a feed or a news sitemap: a machine-readable list of its newest stories, meant for search engines and feed readers. We read each of those lists every 2 minutes. When a link appears that we haven't seen before, it goes into the queue.
Links to live blogs, videos, photo galleries, podcasts, interactives and puzzles are set aside at this point. They either change constantly by design or carry too little text to compare. A page that describes itself as a live blog in its own code is dropped as well, even when its address looks like an ordinary story.
2. We save the first version
Within a few minutes we fetch the article page and pull out the parts that make up the story: the headline, the headline written for social media, the summary, the byline, the paragraphs of the story, and any correction or editor's note. Navigation, adverts, "most read" boxes, subscription prompts and images are left out.
That set of text is saved as version 1, stamped with the time we captured it and a fingerprint of its contents. More on versions and fingerprints.
3. We keep checking
Most edits happen soon after publication, so we check most often at the start and less often as a story ages:
- every 10 minutes during the first hour
- every 30 minutes up to 6 hours
- every 2 hours up to a day
- every 6 hours up to 3 days
- once a day up to a week
- every 3 days until the story is 30 days old, when we stop
When we do find a change, the next check comes within 10 minutes, because edits tend to arrive in clusters.
4. We compare
On each check we fingerprint the story again. If the fingerprint matches the last saved version, nothing has changed and nothing is stored. If it differs, we save a new version and compare the two: first paragraph by paragraph, to find what was added, removed or reworded, then word by word inside each reworded paragraph.
Differences that aren't editorial are ignored before the comparison: spacing, curly versus straight quotation marks, dash styles, "Updated at 14:32" lines, and promotional text that sits inside article pages.
One safeguard applies here. If a long story suddenly comes back at less than 40% of its previous length, the likeliest cause is a page that didn't load fully, not an editor deleting most of the story. We don't record it; we look again 15 minutes later, and accept the shorter text only if three checks in a row agree.
If a story we've saved returns "not found" twice in a row, or starts redirecting to a section front, that is recorded as an event too.
5. We sort the change
Each change gets one label describing what kind of change it is, such as "Headline rewritten" or "Typo or style fix", plus tags for the details: whether figures changed, whether quotations changed, whether text was removed. The sorting is done by fixed rules that are the same for every outlet. The full list of labels and tags.
6. A person reviews it
Nothing is published automatically. Changes that look meaningful go into a review queue, where a person checks each one against the saved versions before it appears on the Changes page. Small fixes and routine additions are kept in the archive but not listed. How review works.
What you end up with
For every approved change there is a permanent page showing the story's headline, each saved version with its capture time, and the changed passages marked up word by word, with a link to the original article. How to read one.