OSINT Swarm · עברית

How scores work

Every number on this page is read from the scoring code that runs the site, so it always matches the scores you see.

1. What the score is

Each report gets a score from 0 to 100 when it is collected. It is a ranking, not a probability: it says how strongly the headline points to a security event that is happening or about to happen at a place we watch, how trustworthy the source is and how many outlets carry it. The site treats stories scoring 15 or more as Important, and 25 or more as hot: shown in red and eligible to be sent as an alert. A story's score is the score of its strongest report.

2. The steps, in order

The score starts at 0. Each step either adds points or multiplies the running total, in this order. The words are read from the headline and the first 500 characters of the article; reports in other languages are translated to English first.

Threat categories

Words match as written in the report, in English. A few of each category's words:

Explosions reported is not a word list: it is a report of explosions heard at a watched country (“Explosions heard in Sirik”). It gets the category weight but not the critical bonus, because a single report is a claim; independent confirmation (section 4) can take it over the alert line.

Timing cues

Watched places

This server watches: Israel, Iran, Lebanon, Gaza, Judea and Samaria, Syria, Iraq, Yemen, Jordan, Saudi Arabia, UAE, Qatar, Egypt, Turkey, Strait of Hormuz, Red Sea / Bab el-Mandeb. Whole regions watched as well: Europe.

Critical events

Anywhere: a hijacked aircraft or ship, squawk 7500, an aircraft shot down, missiles or drones hitting an airport, airliner, embassy or nuclear plant, ships attacked according to UKMTO, attacks on oil and gas sites, ballistic missiles fired at a target, a nuclear test, a declaration of war, a chemical or dirty-bomb attack, attacks toward Mecca or Medina. Only at the watched places listed above, not across whole regions, because elsewhere the same words are routine: ships struck, fired on or seized, interceptions of drones or missiles, strikes on bases and ports, sirens sounding, named nuclear sites hit. The headline must describe the event itself: anniversaries, archive pieces, films and commentary are excluded.

3. Worked examples

These two headlines are scored live by the same code, as reports from a source rated 0.8, carried by one outlet. The “What counted” column is what the scoring records, word for word; the same table appears under “Why this score?” on every story.

“Tanker struck by projectile off Bukha”

“US considers strikes on Iran, report says”

The first reports an event at a watched place and gets the critical bonus. The second names a watched place and strike words, but reports a plan, not an event, so it is halved and stays far below Important.

4. Corroboration and confirmation

Outlets are the distinct news outlets carrying a story. The more there are when a report is scored, the higher it scores (step “Corroboration”).

Independent origins are what actually confirms a story. Ten outlets re-running one wire agency's report are one origin, and an outlet whose headline shares 80% or more of its words with a report already counted is a copy. Wire agencies recognised by their credit line: AFP, Anadolu, ANI, AP, Bloomberg, dpa, Fars, IRNA, Kyodo, PTI, QNA, Reuters, SPA, Tasnim, TASS, WAFA, WAM, Xinhua, Yonhap. “Why this score?” on each story shows both numbers: “Sources: 9 outlets, 3 independent”.

Confirmation. A report is scored once, when it arrives. When a story reaches 3 independent origins (counting its reports from the last 48 hours), its 5 strongest reports are scored again with the number of origins as the outlet count, and may now cross the alert line. This applies only to events at watched places (Aviation, Civil-defence warning, Critical event, Explosions reported, Hijacking & hostages, Military air activity, Nuclear, Shipping & maritime, Strikes & weapons), not to rhetoric, protests, talks or economic news. A score is only ever raised this way, and the critical bonus is not given a second time.

5. What lowers a score

• Analysis, explainers, opinion and live pages (×0.6). Headlines that start with How, Why or What, end in a question mark, or are live blogs that retell the day's events. A breaking report phrased as a question (“Explosions reported near Sirik: has Iran come under attack?”) still counts as news.
• Plans with a distant or conditional horizon (×0.5). “Considers”, “weighs”, “could”, “after the midterms”, “next year”. Such headlines are scored on the headline alone and never count as near-term. Preparation is different: deploying, readying, raising alert levels stays at full weight, because a build-up shows before a strike.
• Rhetoric (×0.6). A vow, threat or warning with no event in the headline. Official warnings (UKMTO, embassies, the Home Front Command, travel advisories) are news and are not lowered.
• Investigations, prosecutions and trials (×0.2). They look back at something that already happened, so they get no critical bonus and never count as about to happen, unless the headline also has near-term timing.
• Politics (×0.3). Elections, coalitions, succession and appointments, unless the same headline names a military subject or a hostile party.
• Anniversary pieces (×0.6), places outside the watch list (×0.75), no place at all (×0.6) and old news published more than 2 days before collection (×0.5).

6. The tension index

The tension index (0–100) says how much is happening at a place right now. It adds up the scores of the stories that name the place, giving the last few hours the most weight.

• One story counts once: its strongest report, however many outlets carry it (its score already rewards them).
• Decay: a story's weight halves every 3 hours. A story scoring 40 counts:

Scale. A few places carry far more news than the rest, so the sum is scaled logarithmically: index = 100 × log(1 + sum) / log(1 + 1000), and a sum of 1000 or more is 100.

Levels

Trend. The sum for the time window you chose is compared with the window just before it: rising when it is more than 1.3× the earlier one, falling when the earlier one was more than 1.3× it, “new” when the earlier window had nothing. Each place lists its 3 strongest stories.

3-day comparison. The last 3 days are split into rolling 24-hour windows, each summing its stories' strongest scores. The last 24 hours are compared with the average of the 2 days before: 1.3× or more is rising, 0.77× or less is falling.

◆ About to happen. A story counts as about to happen when it has a near-term timing cue (tonight, within hours, next week, by Friday, “imminent”, “preparing to”) and a category from Civil-defence warning, Military air activity, Military buildup, Strikes & weapons, Threats & warnings. The cue must be in the headline, unless the story scores 25 or more and 2 or more outlets report it. Politics never counts, and plans and rhetoric carry no near-term cue.

7. Limits

OSINT Swarm is an automated reading of public reports: news feeds, Telegram channels and open data, read by fixed word rules and partly machine-translated. It does not verify events on the ground. It can miss events, score a story too high or too low, and raise false alarms. A score ranks reports against each other; it is not the probability that something will happen.

It is not an official warning system. In Israel, follow the Home Front Command (Pikud HaOref) and official channels for civil-defence instructions.

Scores are automated estimates, not advice. Any use of them is at your own responsibility: see the terms of use.

Questions or a score that looks wrong: [email protected].