Methodology & Sources

Why this page exists

newscrate aggregates content from other organizations rather than reporting it directly. Every figure and event you see here should be traceable back to where it came from — this page is that trace. If something looks wrong, this is also the first place to check whether it's a source issue or a bug on our end (and the contact form reaches a real person either way).

News articles

Pulled from each publisher's own public RSS/Atom feed on a 5-minute cycle, deduplicated by article URL. We store the title, a short excerpt from the feed's own summary, and a link back to the original — never the full article text, and we never rehost or republish anyone's reporting. Currently active feeds:

Source	Category
BBC News	World
NPR	World
The Guardian	World
DW	World
France 24	World
The Wall Street Journal	World
The Telegraph	World
Fox News	World
Al Jazeera	Geopolitical
Foreign Policy	Geopolitical
The Verge	Tech
Ars Technica	Tech
TechCrunch	Tech
NASA	Science
Nature News	Science
The Guardian	Business
CNBC	Business

A feed going offline or changing its URL only reduces coverage from that one source — it doesn't affect ingestion of the others. This list will grow over time.

The "Category" column above is each feed's default, not a fixed label — every article's title and summary is checked against a keyword list per category (tech, science, business, geopolitical), and reclassified when one clearly wins, so a business story from a general World feed lands in Business rather than staying tagged World just because of which feed it arrived on. A feed's default is only kept when no category's keywords clearly apply.

Alongside these, GDELT's DOC 2.0 API — a real-time index of a much larger set of global news sources — is queried for articles matching a geopolitical-crisis keyword filter (coups, insurgency, strikes, ceasefires, sanctions, and similar) and stored under the same "Geopolitical" category as Al Jazeera and Foreign Policy above, since it's a broader net than our hand-picked feed list.

Every incoming article's title and summary is checked against a static list of country names and common aliases; a match tags the article with that country and plots it at that country's rough centroid on the Live Map's News layer, linking straight back to the source — a free, keyword-based, country-level signal.

Source bias labels

Articles from our own curated feed list above carry a small colored label — Left, Lean Left, Center, Lean Right, or Right — next to the source name, plus a "source balance" bar showing the mix of labels among the articles currently on screen. This is a hand-set lookup based on general, widely-cited public consensus about each outlet's overall editorial leaning, not licensed data from a commercial ratings service (AllSides, Ad Fontes, etc.) — we don't have access to and can't redistribute those proprietary ratings, so treat the label as a rough orientation, not a precise score. Trade/science publications (The Verge, Ars Technica, TechCrunch, NASA, Nature News) aren't politically oriented in the way a general news outlet is, so they're left unlabeled rather than forced onto the scale. GDELT's much broader, algorithmically-selected set of sources is also left unlabeled — it spans far too many outlets to rate individually, and a fabricated rating would be worse than none.

Duplicate coverage

When multiple curated-feed outlets cover the same story, only the first one seen shows in the main list, with an "Also covered by N other sources" note rather than 2-3 near-identical headlines back to back. Detection uses a text embedding (Cloudflare Workers AI, run only on our own hand-picked feed articles - not GDELT's much larger volume) compared against other curated articles from the last 48 hours; a high similarity score marks the later one as a duplicate of the first. This is a similarity threshold, not perfect judgment - it's deliberately set conservatively (biased toward missing a real duplicate over wrongly merging two distinct stories), so occasionally the same story may still appear more than once.

Confidence tiers

An article from GDELT's broader net (rather than our own hand-picked feed list above) carries a small "Aggregated" label next to its source name - a human chose every feed on our own list, while GDELT selects sources algorithmically from a much larger pool. Neither is hidden or excluded; the label just says which selection method produced that particular article, so you can weigh it accordingly. Curated-feed articles carry no label at all, on the theory that the default case doesn't need flagging - only the exception does.

Conflict & unrest events (Live Map)

Event data on the Live Map page comes from UCDP (the Uppsala Conflict Data Program)'s Georeferenced Event Dataset, a free, no-key academic dataset of political violence, geolocated to a specific place and date. It classifies events into State-based violence, Non-state violence, and One-sided violence, released on a periodic cadence.

Marker size scales with reported fatalities. Event data on this site is used under UCDP's own terms of use and attribution policy; we are not affiliated with or endorsed by them.

Hazards (earthquakes, cyclones, floods, droughts, volcanoes, wildfires, tsunamis)

Earthquakes come from the USGS public GeoJSON feed, filtered to magnitude 4.5 and above so a global feed isn't dominated by the thousands of minor tremors that occur daily. Other hazard types come from GDACS (Global Disaster Alert and Coordination System), which assigns each event a categorical alert level (Green/Orange/Red) rather than a magnitude. Wildfires additionally draw on NASA FIRMS satellite fire detections, capped to the highest-confidence, highest-intensity detections since the raw feed runs to thousands of points a day globally. All shown on the Live Map, filterable by hazard type and independently of the conflict-event layer. The Live Map also offers a globe projection as an alternative to the default flat map (lazy-loaded only if you switch to it, so it doesn't slow down the default view).

Flight emergencies

From adsb.lol, a free, community-fed ADS-B aggregator: aircraft actively broadcasting one of the three recognized emergency squawk codes (7500 hijack, 7600 radio failure, 7700 general emergency), updated in near real time.

Internet outages

From Cloudflare Radar's outage annotations, plotted at the affected country's centroid (Radar reports by country, not exact location). A national- or regional-scale internet disruption is often a useful crisis signal in its own right — deliberate shutdowns, infrastructure damage, or unrelated technical failures all show up the same way here, so treat this as "something happened to connectivity," not a claim about the cause.

GDELT-flagged events

Beyond the keyword-filtered article search described above, we also pull GDELT's raw Global Knowledge Graph (GKG) export directly — a free, no-key file published every 15 minutes with real per-article geolocation. Records are kept only if they're tagged with themes suggestive of conflict or unrest, then capped to the most negative-toned matches per cycle: a broad automated signal for "something newsworthy and tense happened here."

Reference infrastructure (nuclear plants, military bases)

Static reference layers, not live feeds — nuclear power plant locations from GeoNuclearData (sourced from WNA/IAEA) and military base locations from HIFLD, the U.S. government's Homeland Infrastructure Foundation-Level Data. Fetched once rather than every cycle, since this kind of infrastructure doesn't change day to day.

Prediction markets

Live odds from two sources, kept on the same footing rather than picking one as primary: Polymarket's public Gamma API and Kalshi's public market-data API (a CFTC-regulated exchange). Both are filtered to markets whose titles look geopolitically relevant — both platforms cover everything from sports to entertainment, and this filter keeps the Monitor page focused. A market's probability is the current price traders are willing to pay, not a forecast newscrate is making itself.

A multi-candidate race (a presidential election, say) isn't one market — both platforms structure it as a separate Yes/No contract per candidate. Those are grouped back together into one ranked card showing each candidate's odds, rather than scattering a single race across several unrelated-looking rows in the list.

Cyber advisories

From CISA's Known Exploited Vulnerabilities catalog — vulnerabilities with confirmed evidence of active exploitation, including state-linked campaigns. Treated as a cyber-conflict signal alongside the physical-world data above.

Trending entities & sentiment

From APITube's News API, which attaches named-entity extraction (organizations, people, places) and sentiment scoring to each article it indexes. Refreshed periodically rather than every cycle, and shows the organizations, people, and places mentioned most often across a recent sample of geopolitical coverage on the Monitor page, alongside the average sentiment of the coverage mentioning them — a tone reading on the reporting itself, not a judgment newscrate is making about the entity.

Elevated badges

FX rates, prediction-market prices, and trending-entity mention counts on Monitor are snapshotted every 30 minutes into a rolling history; once a currency, market, or entity has at least 20 snapshots, an "Elevated" badge appears when its current value sits 2 or more standard deviations from its own 30-day average - a plain statistical outlier flag, not a prediction or a judgment about whether the move is good or bad. Nothing shown before 20 snapshots have accumulated (roughly half a day), so a badge is never based on too little history to mean anything. Prediction markets also show their 24-hour probability-point swing directly, flagging moves of 15 points or more.

Country pages use the same statistical approach for conflict-event activity: a country's last-24h event count is compared against its own 30-day baseline, and a status line appears only when it's a genuine outlier (2+ standard deviations, with enough history to mean something). One honest caveat - the baseline is built only from windows where at least one event occurred, not from every 30-minute snapshot including the quiet ones, so it reads as "unusually active even by this country's already-active periods" rather than "active vs. total silence."

Natural-language Archive search

Archive's "describe what you're looking for" box (below the plain keyword search) sends your request to Gemini, which extracts a short keyword query plus an optional category/country - the actual search still runs through the same full-text search as everything else, Gemini never sees or ranks the articles themselves. What it interpreted your request as is always shown alongside the results, so you can see exactly what got searched for rather than a black box. Unlike situation summaries, each request is a one-off, deliberate action and isn't cached.

AI-generated situation summaries

Country pages that show unusually elevated conflict-event activity (see "Elevated badges" above) may include a short AI-generated summary, via Google's Gemini API. The model is given only a small structured object - counts of events, curated vs. aggregated articles, and nearby infrastructure for that country - and asked to describe what those counts show in plain language; it isn't given article text, isn't asked to predict anything, and isn't told to explain causes. Every summary is labeled "AI-generated summary" and paired with the confidence note and caveats the model itself produced about its own inputs. A country with nothing notable, or a failed/unconfigured AI call, simply shows no summary - the structured counts elsewhere on the page are the complete, correct picture either way; the summary is a convenience on top, never a replacement.

Currency stress

Daily ECB reference rates via Frankfurter, covering every currency it tracks (~30) rather than a fixed curated subset, with a search box to find any of them by code or name, and a choice of which currency to measure everything else against. A proxy signal, not a claim of causation — currencies move for many reasons.

RSS feeds

/feed.xml is a personal RSS feed of newscrate's own articles, filterable to a category and/or a source (e.g. /feed.xml?category=geopolitical), so you can subscribe to a slice of the site in your own reader instead of checking back. Archive's filter panel and every country page link to the feed for whatever you're currently viewing.

Source status

The Status page shows the live up/down state of every ingest source, checked on each cron run and overwritten each time (not a historical uptime log). If something on the site looks stale or missing, that's the first place to check whether it's a known source issue rather than a bug.

Refresh cadence & limits

Everything on this page refreshes on a 5-minute cycle via a scheduled job — there's no manual curation deciding what shows up first. We cap how much we pull per cycle to stay within reasonable limits for a small, free site.

Public API

Everything newscrate shows is also available as JSON, read-only, no key or account needed. This is the same fused data the site itself renders — not a raw passthrough of the underlying sources, which you can already reach directly (they're all linked above). Be a reasonable neighbor: this runs on a small free-tier setup, so cache what you can and don't hammer it.

Endpoint	What it returns
/api/articles	News articles. Filters: category, country, source, from/to (date), hours, limit/offset.
/api/search	Full-text article search. q plus the same filters as above.
/api/events	Conflict/unrest events. Filters: country, type, hours, limit.
/api/hazards	Earthquakes, hazards, flight emergencies, outages, GDELT-flagged events.
/api/markets	Prediction markets (Polymarket, Kalshi).
/api/sites	Reference infrastructure (nuclear plants, military bases).
/api/entities	Trending entities and sentiment.
/api/situation?country=	A country's fused situation object, plus an AI summary when notable (see above).
/api/history?type=&days=	Trend history for fx/market/entity/cyber/country_events; add stats=1 for the mean/stddev baseline behind the Elevated badges.
/api/status	Same data behind the Status page.
/api/cyber, /api/fx	Cyber advisories, currency rates.
/feed.xml	RSS, not JSON - filterable the same way as /api/articles.

Questions or a source to suggest

Use the contact form — "Suggest a source" is one of the options there for exactly this.