$ clstr stats --since 2026-04-10
164 days of news,
measured.
Frozen on September 20, 2026. Methodology.
QUERYABLE
The data behind these numbers, over a free news API and an MCP server for agents. No card needed.
00 / TWO WORDS FIRST
What is a cluster? What is a situation?
A cluster is one real-world news event.
When 40 outlets cover the same earthquake, ruling or match, those 40 articles are gathered and deduplicated into one cluster: one event, not 40 copies of one headline. Each cluster carries how many distinct sources stand behind it.
A situation is the storyline those events form.
Related clusters link into one situation over weeks or months, with a maintained summary and a full timeline: not just what happened, but how it developed. The Ukraine-Russia war is one situation holding more than 1,500 clusters.
So the ladder every number on this page climbs is: articles → clusters → situations.
01 / THE FUNNEL
Roughly 12 articles are published for every thing that actually happened.
Since April 10, 2026. 164 days. Base: the 941,096 clusters with two or more sources. The ratio is every article read, divided by that number of clusters. Articles are counted as they are inserted at ingest, copies included.
ARTICLES READ
11,079,667
about 67,559 a day over 164 days
CLUSTERS
941,096
11.8 articles each
SITUATIONS
144,891
6.5 clusters each
11,079,667 articles collapse into 941,096 clusters, and those clusters gather into 144,891 situations, 67,239 of them still active. That works out at 76 articles read per situation: every article read, divided by the number of situations they formed.
02 / STORY LIFESPAN
A headline burns out in three days. The story underneath does not.
Two populations, two windows. Clusters: the 941,096 with two or more sources. Situations: the 77,652 that have finished.
Only 1.2% of clusters keep growing past day three, and a cluster is built to stop: it only accepts an article published within 2 days of its first one, so past that it can grow only by absorbing another cluster. Follow the same news one level up and the picture inverts: of the 77,652 situations that have closed, 81.7% were still developing after day three and 64.3% after a week. Situations still running are left out of those shares, and they are the long ones.
391,230 of 941,096 clusters took no further article after day one.
The longest of the seven in section 06 has run 161 days.
03 / SIGNIFICANCE
Only 0.08% of news clusters rate as exceptional.
Since April 10, 2026. 164 days. Base: the 935,102 multi-source clusters scored so far.
Every cluster is scored 1 to 10 by a language model reading the multi-source summary, weighing how large the event is, who it affects and how much of it is new. As a rough ladder: a 2 is a league result or a regional appointment, a 5 is a national story with real stakes, an 8 is a mass-casualty disaster or a major escalation between states. 8 or above is exceptional, which almost nothing is.
772 of the 935,102 scored clusters reached 8 or above: 765 at 8, 7 at 9. The 3 clusters that scored 10 are left out of this section: on inspection each one was a grouping artifact, not a news event. The middle of the distribution is where almost all news lives, which is the honest version of what a normal day looks like.
04 / PEOPLE
One person appears in 4,815 distinct clusters. The next five most-mentioned people combined reach 3,860.
Measured over the 57 days to September 20, 2026. Counted as distinct clusters, not article volume. No filter: these are the most-mentioned people of any kind.
- 1. Donald Trump 4,815
- 2. Narendra Modi 899
- 3. Vladimir Putin 896
- 4. Javier Milei 750
- 5. Claudia Sheinbaum 677
- 6. Luiz Inácio Lula da Silva 638
- 7. Benjamin Netanyahu 608
- 8. Elon Musk 604
Donald Trump shows up in roughly 84 separate clusters every day. Nobody applied a filter to this ranking: it happens that ranks 2 through 6 are all heads of state or government, and the five of them together still reach only 3,860 clusters. Elon Musk is the 8th most-mentioned person in world news at 604 clusters, behind Narendra Modi, Vladimir Putin, Javier Milei, Claudia Sheinbaum, Luiz Inácio Lula da Silva.
THE BRIEFING
Nobody reads 67,559 articles a day.
Three times a week, the top situations and what moved in each one, built by the same pipeline every number on this page was measured from.
One click to confirm. Unsubscribe anytime.
05 / WHERE CLSTR'S SOURCES WRITE FROM
Over 58 days, 8.3% of the articles CLSTR read that name a publisher country were published in Germany.
Measured July 24, 2026 to September 19, 2026 from the per-country daily counts, copies included, which are a different counter from the article total in section 01. The map's hourly rhythm is sampled from the week of September 13 to 19, 2026, from publication times. Hours are UTC.
This section is what CLSTR's corpus is made of, not a census of the world's newsrooms. Every share below is a share of the 4,270,044 articles that carry a publisher country, out of the 5,998,631 the daily counts retain for this window. 28.8% of articles carry no publisher country and are left out of these shares. The country on an article is where its publisher sits, so a story about one place written somewhere else counts for the writer's country, not the subject's.
PUBLISHER COUNTRIES
share of the 4,270,044 articles with a publisher country, July 24, 2026 to September 19, 2026
- 1. Germany 354,463 articles · 8.3%
- 2. Italy 333,377 articles · 7.8%
- 3. United States 325,250 articles · 7.6%
- 4. Brazil 319,192 articles · 7.5%
- 5. India 186,275 articles · 4.4%
- 6. Greece 175,497 articles · 4.1%
- 7. Japan 156,885 articles · 3.7%
- 8. Spain 153,029 articles · 3.6%
- 9. Poland 131,785 articles · 3.1%
- 10. Mexico 127,225 articles · 3%
The map below plays one week of that corpus. Each dot is roughly two degrees of land, and it brightens with how much its country is publishing in that hour. Run a full week and the bright band sweeps west with the working day: Asia first, then Europe, then the Americas.
Publishing volume by country, one week, hours in UTC.
TOP PUBLISHERS THIS HOUR
Friday 11:00 UTC, the hour the map opens on
- 1. Germany 382 articles · 9%
- 2. Italy 348 articles · 8.2%
- 3. Brazil 285 articles · 6.7%
- 4. United States 244 articles · 5.7%
- 5. Spain 205 articles · 4.8%
The map is a picture, not a figure: it plays an average week, where how much each country publishes on each weekday is measured over the full window from daily counts, and one sampled week, September 13 to 19, 2026, supplies the hour-by-hour rhythm inside each day. No number in this section comes off it. A country enters the map at 50 articles for the week or more.
06 / SITUATIONS THAT NEVER CLOSE
The longest-running of these has been updated for 161 straight days.
Since April 10, 2026. 164 days. 7 hand-picked situations, not a ranking. Each one links to its live page.
- US-Iran Conflict: Maritime Escalation and Geopolitical Tuss로 1,876 clusters · 5,780 sources · 161 days
- Ukraine-Russia War: Economic Warfare & Black Sea Risks 1,585 clusters · 3,869 sources · 161 days
- European heatwave, drought, and logistics crisis intensifies 1,061 clusters · 4,937 sources · 152 days
- Venezuela earthquake recovery and seismic stabilization 786 clusters · 4,272 sources · 88 days
- Lebanon-Israel conflict: Diplomatic friction and escalation 676 clusters · 2,299 sources · 123 days
- Argentina: Economic volatility and labor market decline 563 clusters · 1,351 sources · 154 days
- DRC Ebola outbreak expands and impacts healthcare workers 407 clusters · 3,152 sources · 124 days
Every one of these is a live page you can open right now and read from the first cluster forward.
07 / MOST COVERED CLUSTERS
The most widely covered single cluster drew 1,540 distinct sources.
Since April 10, 2026. 164 days. Ranked by distinct sources across the 941,096 clusters with two or more sources.
- FIFA drops World Cup stake‑sale plan after UEFA boycott threat 1,540 sources · July 29, 2026
- Ceuta migrant surge prompts Spain to deploy troops and sparks EU border tensions 1,510 sources · July 27, 2026
- Colombia earthquake strikes Chocó, causing mass casualties and state of emergency 1,463 sources · August 9, 2026
- Lionel Messi announces retirement from Argentina national team 1,460 sources · August 30, 2026
- U.S. Imposes New Forced‑Labor Tariffs on 60 Trading Partners 1,374 sources · July 22, 2026
- Solar eclipse and Perseid meteor shower peak across Europe 1,322 sources · August 9, 2026
Left out of this ranking: the 10 clusters that grew past 5,000 articles, 3 clusters on a hand-kept list below that cutoff, and 7 further clusters, 3 of them drawing more sources than the first row above, each of which had welded two events into one or ranked one event twice. Those are grouping artifacts, not news events, so they are dropped from these links and from the per-country lists in section 08. The base population the ranking is drawn from is unchanged.
08 / WHERE THE ATTENTION GOES
United States is attached to 19.4% of all clusters that carry a country.
Since April 10, 2026. 164 days. Base: the 749,323 clusters with at least one country tag. A cluster can carry more than one, so shares total above 100%.
- United States 19.4%
- Italy 8%
- Germany 6.8%
- Brazil 6.8%
- United Kingdom 5.8%
- Spain 5.8%
- India 4.8%
- France 4%
- Mexico 3.8%
- Japan 3.5%
- Türkiye 3.4%
- Argentina 3.1%
- Poland 2.9%
- Greece 2.7%
- Canada 2.5%
COUNTRY EXPLORER
What did the news look like from one country?
Pick a country. The panel below is a second measurement of the same corpus, filtered to that country.
For the countries the latest reporting is about, see the world map.
Two windows in here, and every tile says which one it uses. Cluster figures run since April 10, 2026. Article figures cover July 24, 2026 to September 19, 2026, 58 days, copies included, which is everything the daily coverage rollup still retains: article rows age out of the database, so they cannot reach further back.
The explorer needs scripting switched on. It reads a second measurement of the same corpus, with its own window and its own base, so its article figures are not the cluster shares printed above seen through a filter. No finding on this page is gated behind it.
A share of the articles in this window carries no publisher country at all, and the worldwide panel prints that share. Every article share in here is a share of the articles that do carry one, the same base section 05 quotes.
09 / FOUR MORE CUTS
Four more cuts through the same corpus.
Measured separately from everything above, on September 21, 2026. Four statistics, four windows: each one states its own, and none of them is mixed into a figure elsewhere on this page.
TYPICAL CLUSTER SIZE
3
articles
The typical multi-source cluster collects 3 articles, and 42% collect exactly two.
- 2 articles 42.5%
- 3 articles 17.6%
- 4 to 10 articles 29.3%
- 11 to 50 articles 9.2%
- 51 or more articles 1.5%
Sampled: 159,673 multi-source clusters across six evenly spaced bands of the 164-day window. This is the size of a cluster; the 12-per-cluster figure above divides every article read by the number of clusters.
THE WEEK, AS RECEIVED
10%
Friday over Sunday
Friday is the day most articles reach CLSTR, 10% above Sunday, per weekday occurrence over 164 days. The weekend is 27% of an average week.
ARTICLES RECEIVED PER OCCURRENCE · SHARE OF AN AVERAGE WEEK
- Sunday 64,166 · 13.5%
- Monday 66,821 · 14.1%
- Tuesday 69,494 · 14.7%
- Wednesday 68,297 · 14.4%
- Thursday 68,942 · 14.5%
- Friday 70,875 · 14.9%
- Saturday 65,594 · 13.8%
Exact, every article received over 164 days, counted by the UTC hour it arrived at CLSTR rather than the hour it was published, normalised per weekday occurrence. Counted on arrival, which trails publication by minutes to a few hours; that does not move a day-of-week share, so the week is shown and the hour of day is not.
LANGUAGES
29%
English, at most
At most 29% of what CLSTR reads is English. Spanish 14%, Portuguese 7%, German 7%, Italian 6%, across 40 languages.
- English 29.1%
- Spanish 13.8%
- Portuguese 7.3%
- German 7%
- Italian 6.4%
- French 3.4%
- Japanese 3.4%
- Greek 3.2%
59 days of the source rollup, which records unknown language as English, so the English share is an upper bound.
THE BIGGEST STORY OF THE DAY
2.6%
of that day, on a typical day
On a typical day, the largest new cluster is 2.6% of the articles that formed new clusters that day. The peak was 9.4%.
THE PEAK DAY
Strait of Hormuz tensions escalate as US strikes Larak Island
August 30, 2026 · 9.4% of that day
Sampled over 17 complete days. Both numerator and denominator are multi-source clusters created that day, so this is a share among new clusters, not of everything published.
METHODOLOGY
How these numbers were produced
Cluster. One real-world news event: every article
covering the same happening, gathered and deduplicated into a single item.
Situation. An ongoing storyline that groups related
clusters over time, with a maintained summary and a timeline.
What gets counted. Every cluster figure on this page counts only clusters with two or more distinct sources, the ones a reader can actually open on CLSTR. A cluster that never attracts a second source is working state: it is never displayed, and it is deleted after three days. Counting those would pad every base population with rows nobody can see.
The articles-per-cluster ratio is approximate. It divides every article read by the count of displayable clusters, and some of those articles sat in single-source clusters that were later dropped. Treat 11.8 as an order of magnitude, not a precise ratio.
Windows differ, and each stat says which it uses. The article counter reaches back to April 10, 2026 and counts each article at the UTC hour it arrived, copies included. Article rows age out of the database after about a week, so the coverage and entity rollups keep a shorter window of detail: person mentions run over their own stated window, and every article figure drawn from the daily coverage counts, in section 05 and in the country explorer, runs over July 24, 2026 to September 19, 2026 while the cluster figures beside them run since April 10, 2026. No figure here is built from two windows at once.
Counts are floors. Ingest gaps, provider outages and retries all subtract. Every figure on this page is a count of rows that exist, or an average or a share of such counts. Nothing is extrapolated or modelled. The one construction is the map in section 05, which spreads measured daily volumes across the day using hour shares sampled from one week: it is a picture of the week, no figure is read off it, and the section prints how it is built.
Significance scores are model output. Each cluster is scored 1 to 10 by a language model reading the multi-source summary, weighing the scale of the event, who it affects and how much of it is new. In practice the ladder reads roughly like this: 1 to 2 covers routine local items, a fixture result or a minor appointment; 3 to 5 is the everyday middle, national stories with real stakes; 6 to 7 marks major national or cross-border events; 8 and 9 are reserved for events of global consequence, a mass-casualty disaster or a serious escalation between states; and 10 is reserved for an event that dominates coverage everywhere at once; 3 clusters have scored it so far, and on inspection each one was a grouping artifact rather than a news event, so section 03 leaves them out. It is a judgement, not a measurement, and the individual scores carry noise. The aggregate shape is what this page reports. Scores are written at summarization, so 99.4% of displayable clusters carry one: the remainder is the fresh tail, still waiting for its summary, and it is left out of the base rather than counted as a zero.
Where an article is from. The country on an article is the provider's publisher country, which it resolves for most articles and not all. Section 05 and the country explorer quote every article share among the articles that carry one, and both print the share that carries none. Cluster country tags are a different thing again: they are written by the model that writes the summary, they describe what a cluster is about, and section 08 is the one built on them.
CLSTR reports what sources are saying. It does not check whether they are right. Nothing on this page is a claim about the accuracy of any reporting.
Frozen on September 20, 2026 from the production database. Every figure above is compiled into the page, so a number cited elsewhere still matches what is shown here. The country explorer is the exception. It loads its measurements as a separate file when you scroll to it, and that file can be newer than the page around it. The situation titles in section 06 are read when this page was last rendered, so each one should match the page its link opens.
CLSTR: Monitor the situation.
Everything above is a by-product. The product is the other side of it: 67,239 active situations, each one a page that carries the whole arc, so you never lose the thread. Of the 77,652 situations that have already closed, 64.3% were still developing a week in. Follow them here, or run the same pipeline yourself: it ships as a REST API and an MCP server your own tools can query.