Data

Loudest reads 428 news outlets in 99 countries every ten minutes, groups the articles into stories, and ranks each story by how many countries and outlets are covering it. Since 22 September 2026 each story is also judged hard news or not, and the map shows only those judged within the rule; the files keep both, with the judgement. Every completed day is published as a snapshot you can download.

What a day contains

One file per UTC day: the stories that spread furthest that day, ranked by the highest point each reached that day, each with the outlets and countries covering it. Days archived before 4 October 2026 are ranked as they stood at midnight UTC. Stories are clusters of articles, not articles. We link to the lead article at its publisher and never reproduce article text, so the dataset is coverage measurement (who covered what, and how widely) rather than a news corpus. The outlets read are listed on the sources page, with each feed's country, language, ownership and class; every story in a day file has its own page at /story/ followed by its id.

How often it updates, and how far back it goes

The reader runs every ten minutes, all day. A day's snapshot is captured shortly after that day ends at midnight UTC, and its JSON and CSV files are written once, at the same moment. From then on the day is fixed: every request for a file returns the bytes written then, and the database refuses any change to a captured day or its files. A day's page may change its design; its files may not. A mistake found later is never corrected in place: it is noted, with its date, on the day's page. The one exception is removal on legal grounds, such as a court order or a valid takedown notice, and every such removal is logged with its reason.

Until 25 September 2026 the files were rebuilt on each request. The files of every day before it were written once on that day, as the code rendered them then, so a copy downloaded earlier may differ. 14 to 18 September 2026 were also edited after capture, before the lock; each carries an erratum, and the changelog records the edits. Days marked partial were not fully covered and should be treated as a sample.

21 days, from 14 September 2026 to 4 October 2026. A new day is added a few minutes after it ends, UTC.

Downloads

Replace the date in either URL with any day in the archive:

https://www.loudest.news/archive/YYYY-MM-DD.json
https://www.loudest.news/archive/YYYY-MM-DD.csv

For example, 2026-10-04.json or 2026-10-04.csv. Both are served straight from the CDN and cached for a year. The CSV is UTF-8 with a header row and CRLF endings, one row per story; the day's own fields repeat on every row, so files for many days concatenate into a single table without losing which day each row came from.

Checking a file

Each file's SHA-256 is on its day page and below, so you can check that a copy you hold is the one published: shasum -a 256 loudest-2026-10-04.csv on macOS, sha256sum on Linux. A citation that gives the SHA-256 of the file used points at exactly those bytes.

Day fields

In JSON these sit at the top level, beside a stories array. In CSV they are the first columns of every row.

Day lists

Also at the top level in JSON. These are the only fields the CSV leaves out: each would repeat on every one of the day's rows, and failedFeeds alone would add hundreds of kilobytes to the file. The day fields above carry the length of each, so a CSV reader still knows what is there.

Story fields

The site’s own terms, first seen and first published among them, are defined in the glossary.

Lists in the CSV

CSV has no way to nest a list inside a cell, so each of the story lists is flattened into a single field, its values joined with a vertical bar: |. No outlet name, language tag or country code in the dataset contains one, so splitting on it is lossless. In Python: row['countries'].split('|').

outlets is the exception worth knowing about. In JSON it holds positions in the day's outletNames list, because writing 215 outlet names out across 6,000-odd mentions would be a fifth of the file. The CSV has no list to point into, so it resolves them and writes the names in full.

An empty list field means the story's coverage lists could not be rebuilt, not that nothing covered it; see unreconstructedStoryCount. Every story that was reconstructed has at least one outlet.

Why some stories have no lists

The lists were added on 18 September 2026 and are captured with each day from then on. The days before it were filled in from the articles they were built from, which survive a week past a story's last article. Two kinds of story could not be filled in: one that was folded into another story after the day ended, whose row no longer exists, and the story it was folded into, whose articles now include the other's and so no longer say what it alone covered that day. Rather than publish a list that disagrees with the outletCount beside it, both are left empty and counted in unreconstructedStoryCount. On a day captured within minutes of its end, their counts, score and ranking are the originals and are unaffected. On a day captured later they may not be, which is the next section.

Days captured late

A snapshot is computed from the articles as they are grouped into stories at the moment of capture. When two stories about one event are merged, the survivor takes the other's articles and the other's row is deleted. A day captured within minutes of its end is taken before any such merge can touch it, and since 19 September 2026 the capture runs before the merge pass in the first run after midnight, so on those days contaminatedStoryCount and omittedStoryCount are zero.

14, 15 and 16 September 2026 were captured together on the morning of the 17th, after six merges, and show both effects. Nine stories across the three days carry the merged story's figures rather than their own at the day's end: outletCount, countryCount, the general-press counts, score, rank, the three times and the lead article, and a headline written after the merge. They are listed in contaminatedStoryIds. The stories folded away are absent from those files, and are listed in omittedStories with the id of the story each joined and the figures the merge log recorded for it, which are as at the merge, not at the day's end. The originals cannot be recovered, because nothing records which articles were whose, so they are flagged rather than estimated. A contaminated story is also counted as unreconstructed, since its lists could not be rebuilt for the same reason.

Licence

The daily data files, the JSON and CSV file for each day, and the JSON record of each story, are published under Creative Commons Attribution 4.0 International (CC BY 4.0). You may use them for anything, including commercially, as long as you credit them. Attribute as: Data from loudest.news, CC BY 4.0., with a link to loudest.news where the medium allows.

The licence covers the measurements in those files: the clusters, counts, scores, rankings and headlines written for each cluster. It does not cover the articles we link to, which belong to the outlets that published them, and it does not cover the rest of the site, which is reserved as the terms describe.

Ready-made citations, plain text and BibTeX, for the dataset, a day, a story and the methodology are on the cite page.