Skip to content
ARC Warning

How ARC Warning verifies records

This page explains the rules behind the public pages, the open feed, and the labels on each record.

Is ARC Warning an emergency service?

No. If you need help now, call your local emergency number.

ARC Warning publishes situational awareness from public feeds. It does not dispatch responders.

What do the verification labels mean?

OFFICIAL — One official source published the record.

CONFIRMED — At least two distinct sources, and either an official source or two sources that are not unverified.

REPORTED — A single non-official report a reviewer approved.

UNVERIFIED — Stays off the public page until a reviewer approves it. Approval requires the reviewer to confirm the source, the date, and the place. The record must already have a citation URL. A single key does not approve it.

DISPUTED — A reviewer accepted a correction or marked a conflict. Disputed records stay visible and never page anyone.

A channel or official feed tagged as state media or a party to the conflict is labeled Claim by that party. That post does not count as corroboration. Official headlines stay in the feed.

Pins

  • A pin is a coordinate the item published, or a city or region named as the object of in, near, at, or over next to a physical-event verb.
  • On an official channel the same rule also runs in Ukrainian, Hebrew, and French, including an oblast name and an Israeli alert zone.
  • A Pikud HaOref alert is an active warning. It leaves the map when the all-clear arrives or 30 minutes after it was issued. It is not a confirmed incident.
  • A country centroid, and a funding, sanctions, or statement story, stays in the feed with no pin.
  • State media and a party to the conflict are not placed until a reviewer approves the place. The claim ring stays.

How are duplicate reports combined?

Records of the same category, close in time and distance, with similar titles, collapse into one incident.

  • The public URL is the oldest record in that group.
  • Each extra distinct source can raise confidence by 0.08, up to 0.25.
  • The page lists every source.

Corroboration counts distinct independent sources. A second copy of the same report does not count as a second source.

Why are crime locations shown at block level?

Chicago, Los Angeles, and NYPD crime records are stored and shown at about a 100-meter block — not a rooftop or apartment. Private names are removed from public text.

Why do some positions wait before they appear?

Some positions wait before they appear or page anyone:

  • A report tagged as a live military unit position waits 6 hours before it is shown or paged.
  • Military aircraft wait 10 minutes outside conflict zones and 6 hours inside.
  • A missing or uncertain position waits 6 hours.
  • If the aircraft is inside a conflict zone at any point during that wait, the wait is 6 hours.

What can trigger an alert?

Push, SMS, and email go out only for official or confirmed records.

  • A personal rule cannot drop below confirmed.
  • Unverified and disputed records never page anyone.
  • Watchpoints and alert rules are private to the signed-in account. They are not indexed and they are not in the public feed.

How are photos and video handled?

  • Wikimedia Commons files stay on the file page. The briefing uses the Commons thumbnail.
  • Social posts stay linked on an ARC Warning record. They are not shown inside another site's frame.
  • Licensed and public-domain uploads can be hosted after approval or automatic approval. A hosted copy is marked with “ARC Warning · arcwarning.com”, a VERIFIED date, and the source with the capture date.
  • A conflict briefing shows an item after it is approved.
  • Graphic images stay blurred until the viewer asks to see them.

The mark means the copy was checked. It does not mean ARC Warning created or owns the media. The unmarked original stays in storage. A takedown request removes the public embed or the watermarked copy.

How are non-English reports shown?

A caption is shown in the reader’s language and labeled with the source language — for example, “Machine-translated from Ukrainian.” Show original brings the caption back.

If the caption is already in that language, it stays as written.

If the machine translation service is busy or unavailable, the original caption stays on screen. Successful translations are cached per post. Requests are rate-limited.

Safari’s own translator still works: only the machine-translated sentence is marked so it is not translated twice. The rest of the page is left alone.

How do corrections work?

The corrections log is public. When a reviewer accepts a correction, the record is marked disputed.

What are the weapons-system reference cards?

When an approved public record names a known system — for example Shahed-136, Kh-101, Kalibr, Iskander, ATACMS, Patriot, or a FAB glide bomb — the event page links to a reference card at /systems.

Each card lists published name and aliases, type and role, manufacturer and country, operators, launch method, guidance, propulsion, speed, range, warhead, blast or fragmentation radius, interceptability, reported cost, and service history. Every field cites an open source and a confidence note. If a source does not state a fact, the card says unknown/disputed.

The cards are encyclopedic. They are not targeting, manufacturing, or acquisition guidance.

A “Seen in ARC Warning” timeline lists approved records in this database that name the system. Photos and video are official embeds or public-domain government files, with the verification note outside an embed frame. Graphic casualty footage is not shown.

What is Overwatch?

Overwatch at /sandbox/overwatch is a 3D view inside the conflict sandbox. The globe loads only after you choose Enter 3D. Imagery is NASA GIBS Blue Marble (public domain). The surface is the WGS84 ellipsoid.

Each drawn item is labeled Observed, Reported, Computed, or MODELED. Modeled items stay hidden until the toggle is on. This view does not invent any.

Military air, civil air, sea, space, and air-defense alerts use the same feeds as the map.

GNSS interference stays off until the layer is turned on. It is computed from adsb.lol NIC and NACp over a rolling hour and labeled “Computed · from ADS-B integrity (adsb.lol).”

  • NIC 0 and NACp 0 are missing.
  • Only airborne aircraft at or above 3,000 m are counted. An ADS-B version below 2 is left out when the report includes one.
  • A 1° cell is drawn only when at least 10 aircraft contributed, at least 25% are degraded (NIC below 7 or NACp below 8), and that share is anomalous against the cell’s own 7-day baseline.
  • Until that baseline has three days, the layer says it is calibrating and draws zero cells.
  • The card says “likely interference, not confirmed jamming.”

Theater revisit windows are computed from CelesTrak stations and military element sets for 72 hours. A window is the time elevation stays at or above 10° at the theater box, labeled “Computed · from CelesTrak elements.” It is geometric visibility, not a sensor footprint.

An official alert is drawn as a polygon only when that feed published one. Issued and cleared times are shown when the feed published them. A place name is not turned into a region.

The time cursor reads stored air and ship fixes from the last 6 hours, one sample every 5 minutes, inside the aircraft and ship listen areas. A track with no sample in that interval is omitted. No position is filled in. A cursor older than 6 hours stays empty. The card includes the element epoch age.

The status line shows UTC, how old the events are, the sources, and the view center. It does not show a classification banner, a resolution readout, or a thermal mode. Desktop visual filters are labeled VISUAL FILTER and are not sensor data.

Data saver, or a browser without WebGL2, stays on the 2D map. The shell and the last Overwatch snapshot can load again while offline. A missing snapshot stays empty. No position is invented.

What is the conflict sandbox?

The conflict sandbox at /conflicts reads public datasets for six theaters: Russia–Ukraine, the Middle East, Sudan, Myanmar, the Democratic Republic of the Congo, and the Sahel.

Middle East is one theater from 7 October 2023: Israel, the Gaza Strip, the West Bank, Lebanon, Iran, Syria, Iraq, Yemen, Jordan, Saudi Arabia, and Qatar. The page shows a combined regional total and a per-country breakdown.

How totals are built

  • Country totals are every UCDP organized-violence event located there, including local wars — not only cross-border strikes.
  • Cross-border arcs use UCDP’s dyad names and schematic country anchors. They are not flight paths or launch sites.
  • The United Kingdom and United States campaign coded against the Government of Yemen (North Yemen) is listed and not drawn.
  • Countries checked with no events in the extract, and Egypt’s domestic events, are labeled and kept out of the total.
  • The Democratic Republic of the Congo total covers the country. A multi-country total lists each place.

Fatalities and other figures

  • Fatality charts use UCDP’s own low, best, and high estimates, including a civilian count when UCDP recorded one. Injuries are not in that source.
  • OHCHR’s Ukraine update is read on its own schedule. A killed or injured count is shown only when that page publishes it, with the page and the publication date. Those figures are not added to the UCDP bars. OHCHR’s “at least” wording stays on the figure.
  • A recorded-deaths total that includes events located outside the subject place names those places and the count inside the subject.
  • A missing day stays blank. A year is labeled with the months the files cover. The yearly total stays hidden when a month in that span is missing. It is not scaled.
  • UCDP figures from 2026 link to the monthly candidate file they came from.
  • When two sources publish the same count and disagree, both are shown as a range.
  • Displacement rows are UNHCR year-end stocks and are not added to deaths.
  • Official military claims are labeled “claimed by” and left blank when no figure was copied from the government page.
  • Equipment counts, where shown, are Oryx visually documented vehicles and equipment — not people.

Map The map draws one-degree event density and waits 72 hours after a month ends. It is not a front line and not a unit position. Control maps that cannot be redistributed are linked, not copied. Nothing on the page is a forecast.

Where do live ship positions come from?

Live ship positions are not shown. AISStream is turned off until written permission exists. The map does not draw those rows, and the public API does not return them.

A later ship feed is drawn only when its catalog commercial use is allowed on the free map. Positions for the same MMSI would be one track. They are not public incidents and they do not send alerts.

Where do live aircraft positions come from?

Air positions come from adsb.lol, which collects ADS-B broadcasts (the position signals aircraft transmit). The map credits read adsb.lol (ODbL). Positions for the same ICAO24 address are one track. The fresher position is kept. They are drawn on the public map and the signed-in map. They are not public incidents and they do not send alerts.

The worker asks for a 250 nautical mile radius around the Black Sea, the Eastern Mediterranean, the Red Sea, the Persian Gulf, the Taiwan Strait, the English Channel, Khartoum, El Fasher, Bamako, Niamey, and Goma. A radius is a listen area.

Military holds (same as elsewhere):

  • 10 minutes outside conflict zones
  • 6 hours inside conflict zones
  • 6 hours for a missing or uncertain position
  • If a track enters a conflict zone during the wait, the hold stays 6 hours

OpenSky is paused, pending a written license. It is not polled, it does not request a token, and it is not shown as a live aircraft layer. airplanes.live and adsb.lol are unchanged.

airplanes.live may supply aircraft that feed tags as military, when that source is enabled. Credit airplanes.live. Their terms ask commercial users to contact them for permission.

What does a fire pin show?

A fire pin is a published record. ARC Warning does not add a risk score.

  • A named US wildland fire comes from NIFC WFIGS. Acres, percent contained, cause, start time, and personnel are shown only when NIFC published them. The perimeter is drawn on the map when NIFC published one.
  • A NASA FIRMS marker is a satellite hotspot. Fire radiative power is the heat in that pixel. It is not the size of the fire.
  • When at least three hotspots fall in the same 10 km and 24 hours, a footprint may be shown with the label "Computed · satellite hotspot footprint, not official size." That number is not an official size.
  • Distance to the nearest town is computed from the coordinate and the public place list. It is labeled Computed.
  • A Red Flag Warning, Fire Weather Watch, or evacuation or civil emergency message is shown when a National Weather Service alert covers the point. Wind is shown only when that alert published a wind speed or gust.
  • Air quality is shown when a reading was returned for a point near the fire. When no reading was requested, the card says so.
  • Burned area stays blank when no burned-area figure is attached.
  • A point inside a mapped conflict region is labeled as geographic context. That label is not a cause.

The pin size follows official acreage when NIFC published it. Under 100 acres is the small pin, 100 to 999 acres is the medium pin, and 1,000 acres or more is the large pin. That size is a map scale. A hotspot with no official acreage uses fire radiative power only to choose the pin, and that choice is not a size.

NIFC incidents that are still on the current-incidents list stay on the map for about 36 hours after the last check, even when NIFC's own update is older than the time window. The time on the pin is that NIFC update. The start time is the discovery time NIFC published.

What is in the public data feed?

Approved records are available as JSON, CSV, and GeoJSON at /api/public/events and at the same paths under /api/v1.

  • Anonymous use: 30 requests a minute.
  • With an API key: 300 requests a minute.
  • When you hit the limit, the response tells you when to retry.

Pending, rejected, and not-yet-displayed rows are omitted. Unverified reports stay off this feed until review. Watchpoints and alerts are not in the feed.

Full header names and developer examples live on the API page.

Plain-text summary for crawlers