Four things: a live map of reported incidents, a chronological feed of the source reports behind them, a permanent page for every tracked incident, and running live-update pages for each theater. Everything on those surfaces is compiled from public reporting, and nothing is published without a linked source.
Feeds are fetched every 20 minutes. Each new article is checked against the theater it belongs to, so a Ukraine story does not land on the Iran map and a global wire item date-lined Tehran does not get pinned to the Iranian capital. Reports that describe the same event are grouped into one incident with a source count. The grouping is automated and can still split one story into several incidents or, more rarely, join two, which is one reason every incident shows its linked sources.
The place named in a report is matched against a gazetteer of cities, bases, ports, islands and provinces for each theater. Where the gazetteer has no match, a language model proposes coordinates, and those are checked against OpenStreetMap place data with a distance guard before the incident is allowed on the map. Locations that cannot be resolved stay off the map and appear only in the feed. Verified locations are locked so later automated passes cannot move them, and wrong pins reported by readers are fixed by hand.
Each incident page carries a short brief in three parts: what happened, why it matters and what is unclear. The briefs are drafted by a large language model working only from the reporting linked on that page, and they are labelled as AI-assisted wherever they appear. They are meant to orient a reader, not to replace the sources; they can be wrong, and the linked reporting always takes precedence. The same applies to the daily brief and the analyst chat under the intel tools.
A report appearing on the map means that a public source made that claim at that time. It is not independent confirmation that the event happened as described. Source count, perspective labels, timestamps and the disputed flag are there so readers can weigh a claim themselves.
Corrections and takedown requests go through the contact page. A wrong location is re-pinned and locked; a mis-grouped or out-of-scope report is removed; a source that objects to being linked is delinked. Incident pages are kept as a permanent archive so that links keep working and the record of what was reported, and when, stays available. Pages that no longer meet the publication standard are withdrawn from search indexes but not deleted.
How often is the map updated? Source feeds are fetched every 20 minutes and new incidents appear as soon as they are processed and placed. The timestamps on each incident show when it was first reported and last updated.
Why is an incident pinned in the wrong place? Automated geolocation can misread a place name or fall back to a provincial capital. Report the incident from its page and it will be re-pinned by hand and locked against further automated changes.
Are the AI briefs fact-checked? No. They are drafted from the linked reporting on each page, labelled as AI-assisted and reviewed when readers flag problems. The linked sources are the record; the brief is an orientation aid.