Click and drag rotates and tilts the globe; scroll zooms. The attitude indicator in the bottom right is also draggable for fine pitch and heading control, with a reset button to snap back to top-down. Country panels stay on the right rail; live info cards (aircraft, ships, satellites, fleet, points of interest) live on the left, and the two surfaces can be open at the same time so a country brief and a tracked target are visible together.

Aircraft, ship, and satellite info cards each carry a Follow button. Engaging follow on a ship or aircraft pins the camera directly above the target at a regional altitude that keeps the surrounding area in view; engaging follow on a satellite pins the camera radially above the satellite so the orbit rotates the camera through space alongside it. Any direct interaction with the globe (drag, zoom, touch) cancels the follow.

The layers panel has a “Rotate in background” toggle in the Visual section. When on, the globe slowly spins about its polar axis after roughly fifteen seconds of inactivity, with the speed adjustable via a slider underneath the toggle (default 2 minutes per revolution). Live data layers continue updating during rotation. Any pointer, wheel, or keyboard interaction stops the spin and resets the idle clock.

Click the TV icon at the top-left of the globe (or open Settings then Cinematic mode) to flip the surface into a self-running broadcast view. The camera autonomously cycles through whatever is most interesting right now: air-defense alerts as they fire, strategic airlift waves in motion, tanker activity, refinery thermal anomalies, hurricanes and tsunamis, volcanic activity, GDACS humanitarian alerts, US carrier movements, GPS-spoof events, and breaking news from UN, DoD, and NATO press feeds. Live rocket launches and the White House live broadcast also hold the camera while they are active, with the webcast playing in a small window on screen.

A red pulse badge at the top-left opens a Noteworthy panel that lists active and upcoming priority events with countdown timers and the same embedded video, so a launch or a presidential briefing is one click away even when cinematic mode is off.

Every scene shows a lower-third caption with both UTC and your local time, plus a side detail card that mirrors the same surface a normal click would open. Voice narration describes each event in newscaster cadence. When an air alert fires anywhere in Ukraine or Israel, a curated multi-cam livestream pops in at the top-left and stays for five minutes; a fresh alert in the same region resets the timer. A right-side rail auto-scrolls through the latest X and Telegram signals (under thirty minutes old, with any attached images or short videos rendered inline). When nothing newsworthy is firing, the globe slowly orbits at continent altitude.

When the camera flies to an event, cinematic mode temporarily switches on whatever globe layer that scene needs (FIRMS for fire scenes, fleet markers for carrier scenes, GDACS polygons for disaster scenes, oblast outlines for Ukraine air alerts, and so on) and restores your prior layer state when it cycles to the next event. You always see the thing the camera is pointed at, without leaving every layer toggled on the rest of the time.

Cinematic mode is desktop-only and pauses automatically when any modal opens or when the browser tab is hidden. Press Escape to pause, click the EXIT pill in the lower- third to leave. The mute toggle next to the EXIT pill silences voice cues while keeping the visual loop running. Append ?cinematic=1 to the URL to boot directly into cinematic for newsroom or iframe display.

Voice narration uses your browser's built-in text-to-speech by default. For higher-quality newscaster audio you can paste your own OpenAI API key into the cinematic settings panel; the key is stored only in your browser and never sent to our servers (the audio request goes directly to OpenAI). Get a key at platform.openai.com. Cost is roughly a cent per minute of speech.

Each tracked country has a hand-maintained intelligence-style situation brief, a single dense paragraph naming officials, events, casualty figures, and the immediate forward outlook. Briefs are updated when the on-the-ground picture shifts; the file is plain JSON so changes go live without a rebuild.

Briefs are written from public reporting only. We don’t originate facts; we condense them.

Heads of state and heads of government come from a daily Wikidata scrape (P35 and P6 properties), merged with a hand-curated override file for cases the scrape gets wrong - e.g. recent successions, contested claims, or de-facto leaders Wikidata still lists by an outdated record. Leader photos resolve through the Wikidata image property when available.

Next-election dates are hand-curated from public election commission announcements; the panel shows the ISO country code, date, and election type plus a days-until pill.

Sanctioned individuals, organisations, and vessels are aggregated from the OpenSanctions consolidated dataset, which merges OFAC, EU, UK, UN, and other sanctions lists into one queryable corpus. We stream-parse the consolidated CSV server-side, derive a tiny per-country index, and serve it back to the panel. The country panel breaks the count down by category (rose pink for individuals, violet for organisations, cyan for vessels) and links out to the OpenSanctions search for the country.

The traffic chart shows Cloudflare Radar’s 48-hour netflow timeseries for the country, with annotations when Cloudflare detects a national-scale outage. It’s a useful tell during civil disturbances or government-imposed shutdowns , the curve drops when the country goes dark.

The Overview tab leads with a Country dossier card showing region, subregion, capital, population, land area, GDP and government debt with the year of the underlying World Bank reading, the official currency code, and the official language(s). GDP and debt come from the World Bank Open Data API (NY.GDP.MKTP.CD and GC.DOD.TOTL.GD.ZS) refreshed daily. The local currency exchange rate to USD comes from open.er-api.com, refreshed hourly, so the dossier always shows a current value.

The threat score is the sum of weighted channels. A structural baseline reflects long-running risk (active conflicts, sanctions, recent incident history). On top of it sit live conflict channels: sustained activity, which tracks war-grade reporting held across days and is smoothed so it rises within hours and decays over days rather than flickering; a last twenty-four hours channel that weights fresh items more heavily than day-old ones; detector corroboration, which adds weight when thermal or pattern detections fire above the country’s own normal rate; and a de-escalation channel that subtracts on ceasefire, withdrawal, and truce reporting. All channels read the same curated term lexicon, including non-English vocabulary, so a Russian rocket-strike report or an Arabic ceasefire mention counts the same as its English equivalent. A gate keeps tension talk alone from reaching the war band without direct-attack reporting behind it.

Score composition rows show each channel’s point value and its share of the score, with the underlying signal evidence (source counts and matched vocabulary) listed below. Score breakdown adds the channels to the threat score line, with a range-clamp row when the total is held to the 5 to 100 band, so the math is always reconcilable end to end.

The world tension meter in the top bar is a single 0 to 100 read on how close the world is to wider conflict, paired with a five-level readiness condition. It compounds the active wars, weighting each by the power of the states involved, then adds separate terms for great-power escalation, strategic movement (airlift waves and tanker surges above normal), and nuclear rhetoric across the feed. Wars compound, but only great-power belligerence reaches the top of the scale, so a single regional war cannot saturate the meter. Click it for the breakdown and the top contributing countries. It updates every fifteen minutes.

The Government tab shows the country’s system of government, the head of state and head of government with their official portraits, parliament chamber composition with a hemicycle seat-dot visualisation when party seat counts are catalogued, and the next election date per chamber when Wikidata records one. Leaders, system labels, chambers, and next-election dates are auto-sourced from Wikidata and refreshed daily. Portraits resolve through the Wikidata image property.

The Relations tab surfaces the country’s membership in twenty international organisations across global, defense, economic, and regional categories: UN, NATO, BRICS, OPEC, EU, OECD, G7, G20, Five Eyes, ASEAN, Mercosur, USMCA, AU, Arab League, OAS, ECOWAS, SCO, CSTO, Commonwealth, and OPEC. Each row notes the year the country joined when Wikidata records it. UN Security Council status is called out separately, with permanent five and current rotating membership both surfaced. All data refreshed daily via Wikidata.

The Military tab pulls active duty and reserve personnel, paramilitary strength, annual defence expenditure and percentage of GDP, conscription status, and a list of service branches (army, navy, air, marines, space, coast guard, paramilitary). Numbers come from Wikidata structured records where they exist, plus the open-licensed CIA World Factbook mirror for the narrative summary and personnel breakdowns. Refreshed daily server-side.

The Live tab embeds the country’s catalogued news broadcasts (state and major commercial broadcasters where they stream live) directly inside the panel as auto-mounting iframes with thumbnail posters from the YouTube channel. Streams that go offline switch to an OFFLINE pill; live streams pulse a red indicator. Pop-out opens a floating window that survives panel close and country switches.

Aircraft trails are scored continuously for two shapes - tight orbits (multiple full revolutions inside a small diameter) and racetrack patterns (sustained parallel-leg tracks with reversals). Both are signatures of ISR / AEW / refueling stations. Trainer aircraft are excluded categorically because routine syllabus flights produce these shapes by design.

A vessel is flagged for cable loiter when its recent track sits tightly clustered, mostly stationary, near the published route of an undersea cable, with no other vessels nearby and no registered anchorage in range. The detector layers proximity, stationary fraction, cluster diameter, and crowd-density gates; it weights confidence by how far above each gate’s floor the candidate clears. Time-of-day shading mildly discounts the dawn and dusk windows because that’s when fishing fleets concentrate.

FIRMS thermal detections are joined against the energy-infra catalogue. Detections within proximity of an oil / gas / LNG / nuclear / coal / hydro asset surface as fire_at_infra; higher fire radiative power readings escalate to high_frp_at_infra.

Routine flare-stack burnoff at refineries, gas plants, and LNG terminals is classified separately as infra_flare_routineso the live patterns rail stays focused on novel fire incidents rather than expected industrial activity. Routine detections still render on the globe under a dedicated Routine FIRMS layer (default off) so they remain auditable. An anomaly escape promotes a routine detection back to a live fire alert when the intensity spikes well above the site’s baseline or when several zones at the same facility light up at once.

We watch for sustained transits of strategic airlift airframes , C-17, C-130, C-5, KC-46, KC-135, A330 MRTT, A400M, Il-76, Y-20, and similar, between major theatres (North America, Europe, MENA, Indo-Pacific). When several distinct airframes traverse the same theatre pair within a rolling window, the pattern fires once as a bidirectional cluster, a NA↔EU wave fires the same whether some airframes went outbound, others returned, or both.

The expanded entry shows a per-direction breakdown so you can see which way the bulk of traffic is moving, plus the per- aircraft list with type and last position. A “Show flight paths” button overlays every cluster member’s trail on the globe; each polyline and the matching panel callsign share the same color as the live aircraft icon (tankers pink, cargo turquoise, etc.) so the three are unambiguously linked. Coverage gaps mid-Atlantic that were stitched together by trail-bridging render as dashed segments so the line stays honest about what was actually tracked. Clicking a callsign in the panel auto-loads the path overlay (if it isn’t already on) and flies the camera to that aircraft’s latest position.

Origin resolution is multi-source: a flight’s explicit first-sighting field, a backwards walk through the trail buffer for the earliest point in a different region than current, and a corridor-and-heading heuristic for flights that drop off ADS-B coverage on the eastern Mediterranean approach (where coverage thins). The detector won’t fire on intra-theatre traffic, Tampa to Virginia is both North America, and never qualifies.

Two related signals on aerial-refueling activity. A tanker cluster fires when two or more refueling platforms (KC-135, KC-46, KC-10, A330 MRTT, A310 MRTT, Il-78) are airborne within roughly a typical AAR holding-anchor diameter of each other; that’s the signature of a single anchor being worked, common ahead of regional ops. Connected components by transitive proximity, so three tankers spread across a wider stretch still resolve to one cluster.

A tanker swarm is the same idea at theatre scale: three or more tankers concurrently flying racetrack or orbit patterns across an entire macro-region (the broader Atlantic arc, or the Pacific) at the same time. Distinct from a cluster (proximity = one operation) because a swarm is dispersed and represents coordinated multi-anchor activity. A separate tanker theatre surge fires when six or more tankers are airborne in a single macro-theatre regardless of whether they have settled into orbits yet, which catches mass refueling about to start before the orbits are visible.

Tanker cluster, tanker swarm, and tanker theatre surge details are subscriber-only. Free tier sees the count, the theatre or region label, and the aircraft inventory; per- aircraft callsigns, positions, and the flight paths overlay are gated. Strategic airlift uses the same gate.

We walk each tracked aircraft’s recent ADS-B trail looking for consecutive position reports whose implied ground speed exceeds physical aircraft limits. Modern airframes cruise well below Mach 1; anything that purports to have covered tens or hundreds of kilometers in seconds is either a bad receiver or, more often, a deliberate position spoof, which is routine in known electronic-warfare zones (Eastern Europe, the Levant, the Strait of Hormuz, the Black Sea). The pattern surfaces a critical-severity entry on the rail with the implied speed, displacement, and the two coordinates between which the impossible jump occurred.

The detector trusts the upstream trail-bridging gate, so it skips pairs separated by the long ADS-B silences common on transoceanic routes. A flight that reappears far downrange after a long over-water gap won’t false-positive: by the time the next ping is appended, the gap evaluation has already accepted or rejected it on plausibility grounds, and only short-cadence implausible jumps reach the spoof check.

Every signal post carries a small EN button next to the timestamp. Click it once and the post text swaps to English; click again to revert. Telegram posts in non-Latin scripts (Russian, Ukrainian, Persian, Arabic, Hebrew, Chinese) are pre-translated when they enter the cache, so the button serves them instantly. Other posts translate on demand and cache the result for an hour, so a second click on the same post is free.

Translations are produced by a public translation service. The original text is what gets scored for the threat composition, the English render is just for reading; nothing you read in English changes the underlying numbers.

Click the strategic asset markers for the Suez Canal or the Panama Canal and the side card now shows the latest reported daily transit count along with a vessel-type breakdown (containers, tankers, dry bulk, general cargo, ro-ro). Numbers come from the IMF PortWatch dataset, which derives daily chokepoint transits from automatic identification system traffic and publishes them as a public ArcGIS feature service. The card links back to the PortWatch landing page as the source.

A Weather group in the layer filter exposes three live overlays. Active weather alerts are pulled from the National Weather Service Common Alerting Protocol feed (United States coverage) and rendered as severity-shaded polygons. Active hurricane and tropical-storm cones come from the National Hurricane Center in the Atlantic, East Pacific, and Central Pacific basins, with the cone of uncertainty as a polygon and the forecast track as a dashed line. Volcanic ash advisories from the nine regional Volcanic Ash Advisory Centres render as ash-colored polygons over their published extents.

All three are off by default; toggle them on from the Weather section of the layer filter. The weather radar overlay (RainViewer, separate live precipitation feed) and the space-weather chip in the toolbar are unchanged.

A once-per-day summary delivered at the local hour you pick, in the timezone you pick. Each issue leads with a short prose read of the day, then steps through the operational picture: strategic airlift activity by corridor, tanker movements, repeat thermal anomalies at infrastructure with coordinates, active tropical cyclones and tsunami warnings, severe US weather alerts, high-severity disaster events, US carrier locations and exercises, and a small commodities snapshot covering the oil pack and gold.

The world-report block surfaces curated headlines from UN, US Department of War, and NATO press feeds with publisher thumbnails alongside top-scored signals from X and Telegram OSINT channels. Every headline and signal links back to the source.

Subscribe from the email icon in the toolbar (visible to everyone, configurable for Pro). One-click unsubscribe is in every email. If your subscription lapses, the digest pauses automatically; re-enabling requires an explicit click in the same panel after you renew.

Real-time push: when a pattern fires that you’ve subscribed to, SENTINEL POSTs an HMAC-signed JSON body to the URL you provide. Useful for routing specific pattern kinds into Slack / Discord / PagerDuty, your own dashboard, or a downstream automation.

Pick which kinds to subscribe to (compound pre-strike, strategic airlift, tanker activity, tactical-flight events, infrastructure thermal anomalies, maritime loiter), the minimum severity floor, and a per-pattern cooldown that prevents the same pattern from firing more than once within the window. Each target gets its own signing secret, displayed once at creation; rotate by deleting and recreating.

Dedup is automatic. A pattern that keeps emitting on every detector cycle (a refinery thermal anomaly that stays active for hours, a flight in racetrack pattern for an hour) fires the webhook once, then again only after the dormancy window resets and the recurrence counter increments. Severity escalation always fires even within cooldown. Up to 60 fires per hour per target; surplus is dropped and surfaced to your receiver via an X-Sentinel-RateLimit-Droppedheader on the next successful fire. After 3 consecutive receiver failures the target backs off to retry once per hour; after 10, the target auto-disables and you re-enable from the account page.

Signature: X-Sentinel-Signature: sha256=<hex>computed over the raw body. Same shape as Stripe and GitHub webhooks, so existing receiver libraries verify without translation. Verify on the receiver, never trust the body without it.

Manage at /account/webhooksor the webhook icon in the toolbar.

Every catalogued country has a dossier at /<iso> (lowercase ISO-3, e.g. /usa, /rus, /irn). Each page leads with the country flag, a close-cropped country shape, and a world map with that country highlighted. The body carries a generated history article, a preview of the daily briefing, and the in-app country panel embedded as tabs (Overview, Intel, Live, Government, Relations, Military, Cyber, Wanted). The full Pro briefing is not exposed publicly.

/countries lists every catalogued country grouped by region. The same directory is rendered inline at the bottom of the splash page, so every country dossier is one click from the homepage.

/today renders the current daily SITREP narrative; per-day archives live at /today/<yyyy-mm-dd>. The narrative is the same prose lede generated each morning for the email digest; the public version surfaces it without the email chrome.

/topics is a directory of long-form explainers, one per major globe-layer category: aviation, maritime, conflict, energy and infrastructure, weather and disasters, signals and prose, cyber and spectrum. Each explains where the data comes from, how the live classifier reads it, and how it cross-links to the country dossiers.

Several toolbar tabs on the live globe also have a public article-style page that reads from the same in-process caches and renders the data as text and tables:

• /patterns — current rolling-window list of auto-tagged tactical patterns (free tier; per-aircraft and per-vessel detail stays Pro on the live rail).
• /posture — flat world map with US fleet positions, plus paginated UN / DoD / NATO / OPEC / BRICS / G7-G20 release feeds with article thumbnails.
• /cyber — OONI censorship leaderboard, Cloudflare Radar BGP route-leak and hijack events, MISP threat-actor presence by country.
• /current-events — Wikipedia Portal:Current_events scrape, day-by-day, with article thumbnails.
• /launches — upcoming orbital launch schedule with rocket imagery.

/glossary is a single anchored page covering naval and maritime acronyms, aviation terminology, conflict and hazard taxonomies, cyber terms, pattern category names, and the international organisations tracked on the country foreign-relations panel.

/sitemap.xml enumerates every page above plus the daily-brief archive entries. /robots.txt allows everything except the API surface and deep-link query parameter URLs that bypass the splash.

GET /api/ais?bbox=-7.0,49.0,2.0,52.0
{
  "fetchedAt": "2026-05-22T18:04:31.214Z",
  "ships":     [ … ],                       // legacy top-level
  "data":      { "ships": [ … ] },
  "meta":      {
    "kind":      "raw",
    "sources":   ["aisstream.io"],
    "asOf":      "2026-05-22T18:04:31.214Z",
    "fetchedAt": "2026-05-22T18:04:31.214Z"
  }
}

GET /api/patterns
{
  "patterns": [ … ],                        // legacy top-level
  "cached":   true,
  "data":     { "patterns": [ … ] },
  "meta": {
    "kind":      "inferred",
    "sources":   ["flights","ais","fires","energy-infra","submarine-cables"],
    "asOf":      "2026-05-22T18:03:00.000Z",
    "fetchedAt": "2026-05-22T18:04:31.214Z"
  }
}

Individual Pattern records inside data.patterns carry their own per-entry confidence band when the detector emits one; the envelope-level confidence is reserved for aggregate scores and is omitted on multi-detector responses.

Loading the site on a phone shows a static globe placeholder with a Launch Globe button. The 3D engine does not initialize until you tap that button. If you are just passing through, you spend nothing.

The launch portal also has a Reset settings button. It clears your saved layer toggles, graphics preset, and photoreal state on this device. Use it if a previous session left the globe in a state that crashes or stalls. After reset, tap Launch and you are back to safe defaults.

Reset only affects this device and only the SENTINEL settings listed above. It does not log you out, clear any data outside the site, or affect another device you have signed in on.

Phones start at the Low graphics preset for the safest first paint. You can bump it to Medium or High from the Graphics panel inside the layers menu. Higher presets render more terrain detail and a wider visibility ring for distant contacts; lower presets keep the globe responsive on hot or older devices.

If a session ever ends abnormally, the next visit may automatically fall back to safer defaults so you are not stuck in a loop. If that happens, your manual settings are restored whenever you toggle them again.

Some layers are heavier than others. Submarine cables, photorealistic 3D buildings, satellite catalogs, and high density imagery are tuned to render lighter on mobile, including hiding when zoomed out far enough that the detail would not be readable anyway. If you want the full desktop experience, open the site on a laptop or desktop.

The globe is tuned to move less data and use less graphics power than a naive build would. Layers you have turned off do not fetch their backing data in the background. Heavy lists ship a compact form and load full details only when you click an item, so the live aircraft, ship, and satellite feeds stay light. Region scoped feeds send only what is in view, and slow moving data uses caching so a repeat fetch returns a short "not changed" reply instead of the whole payload.

On the graphics side, idle rendering pauses entirely when nothing is moving on the globe so laptops stop spinning their fans, dense icon layers such as cameras draw in a single batched pass, layers that are off release their graphics memory, and several internal render settings scale with the chosen graphics preset to fit integrated GPU budgets. None of this changes what data you can see.