1. Event discovery
WORFLICT uses GDELT event data to identify potentially relevant reports involving armed conflict, military activity, violence, or related geopolitical developments. GDELT is a discovery layer; it is not presented as the publisher of the linked article.
2. Source selection
Available source records are evaluated for language, recency, and connection to the detected event. The selected publisher domain and original URL remain visible. A source appearing in the feed does not mean WORFLICT endorses its framing or conclusions.
3. Source coverage signals
WORFLICT counts distinct publisher domains attached to the same event record and labels reports as single-source or multi-source. Readers can filter the live map by this signal. Several domains may repeat the same wire story, share ownership, or rely on one original claim, so the count is evidence of coverage breadth—not independent verification, truth, or political neutrality. We do not automatically label a publisher as propaganda without a maintained, transparent ownership and affiliation dataset.
4. Source types, claim comparison, and timelines
A maintained registry identifies broad source types such as news agency, state news agency, international media, investigative outlet, official source, or unclassified publisher. “State news agency” records formal ownership or institutional status; it is not a claim that a report is false, propaganda, or less reliable. A type describes the publisher and is never a trust score. DeepSeek V4 Pro compares available publisher headlines and candidate updates to determine whether they appear to describe the same concrete incident; records judged related but different, clearly different, or too vague are excluded from that incident's publisher count. This is event matching, not fact verification or a publisher reliability score. Artificial intelligence structures only claims explicitly present in the reports. Deterministic rules preserve the attributed figures and sources and identify potential differences in figures, responsibility, and confirmation status. DeepSeek V4 Pro then compares each candidate against the reports grouped under that incident, removes aliases, equivalent wording, and non-comparable claims, and distinguishes a meaningful difference from a later update. Differences are displayed without automatically deciding which claim is true. Automated comparison can still miss differences or classify them incorrectly.
5. Automated summaries and translations
Artificial intelligence may generate a concise title and summary and translate them between English and Turkish. When at least three matched publisher pages are publicly readable, WORFLICT may also create an original multi-source synthesis. The system fetches only public pages that permit access, does not bypass paywalls or authentication, and always keeps links to every report consulted. A report-level evidence-support score measures visible attribution, specificity, uncertainty and overlap with the other supplied reports; it is not a truth score or a general rating of the publisher. The system is instructed to preserve the central claim, avoid invented details, and use restrained language. Automation can still mistranslate names, merge unrelated details, or repeat an inaccurate source claim.
6. Location and grouping
Events are grouped using geographic information associated with the underlying record. A location can refer to where an event was reported, discussed, or geocoded and may not always identify the precise incident site.
7. Severity
The displayed severity score is a prioritization aid derived from event coding and processing rules. It is not a casualty estimate, legal classification, or editorial judgment about human suffering.
8. Updates and retention
The live feed refreshes throughout the day. News translations and feed records are retained for no more than 150 days. Timestamps indicate the source record time when available, not the time WORFLICT independently confirmed an event.
9. Attribution
Event discovery is credited to GDELT, map data to OpenStreetMap contributors, and journalism to the publisher linked on each report. WORFLICT summaries are not copies of full articles.