9 agents across 3 teams. 34 ACLED incidents reviewed. Alternative sources cross-referenced against UNAMA, WFP, HRW, and OCHA humanitarian feeds. The gold standard for global conflict data has a problem. This report documents it.
ACLED — the Armed Conflict Location & Event Data project — is the database the UN, NATO, NGOs, and governments cite when they need to understand where in the world armed conflict is happening. It claims to track battles, explosions, remote violence, attacks on civilians, protests, riots, and strategic developments across every country on earth.
When the FA CONFLICT swarm analysed the May–June 2026 ACLED dataset against alternative sources — UNAMA sitreps, WFP incident reports, HRW field documentation, NRF communiqués, ISPR press releases, and regional Pashto/Dari/Urdu media — what emerged was not a picture of global conflict. It was a picture of one conflict, reported through one language, via one media ecosystem.
This is not an accusation of bad faith. It is a structural finding. The methodology determines what gets counted. And the methodology has a geography.
The May–June 2026 ACLED dataset contains 34 incidents. This is what they look like by geography.
Africa has more than 15 active armed conflicts as of 2026. The continent returns 1 incident in a 34-incident global dataset. This is not a data quality problem. It is a methodology problem — and the methodology was never designed for global parity.
Afghanistan returned zero entries in the May–June 2026 ACLED dataset. These are the events that were happening.
Primary reporting in Pashto/Dari — not in ACLED's English RSS feed. UNAMA protection data: 6–8 week lag before ACLED entry. Event structurally invisible to English-language scrape.
Zero journalist access to Panjshir under Taliban governance. NRF communiqués distributed in Dari via Telegram channels not monitored by ACLED. RFE/RL Dari reported it; English service did not.
Taliban reclassified as "criminal" — deliberately depoliticised, reducing media pickup. Single-source, single-fatality event below ACLED verification threshold for low-media regions.
ISPR release in Urdu. Cross-border jurisdictional gap — incident sits between ACLED's South Asia and Central Asia regional desks. Neither desk owns it.
HRW access restricted during scrape window — returned empty results. VOA Dari not in English aggregation. Zero-fatality protests in Afghanistan are systematically excluded by verification thresholds.
AP wire published after the May ACLED snapshot cutoff. BBC Persian in Farsi. UNAMA flash updates: 4–6 week verification lag. Will likely appear in July 2026 dataset — absent from current window.
WFP incident reports circulate via ReliefWeb/OCHA humanitarian channels — not public media. Ghor Province has zero resident journalists. This event is structurally invisible to any media-scrape methodology. It does not exist in any database except WFP's internal logs.
The problems are structural, not incidental. They require structural responses.
Partner directly with Pashto, Dari, Urdu, Amharic, and Tigrinya media monitors. → Events in these languages enter the pipeline on the same timeline as English sources.
Publish a flagged "preliminary" layer alongside verified data. → Events near the snapshot cutoff are visible immediately, marked unconfirmed, updated as verification completes.
Fund local incident reporters in zero-journalist provinces — modelled on ACAPS's local information management system. → Events in Ghor, Nuristan, Panjshir, and equivalent zones enter the dataset.
Cross-border incidents get a dedicated shared desk — not split responsibility between two regional teams. → Pakistan-Afghanistan, Ethiopia-Somalia, Sudan-South Sudan border events stop falling through.
ACLED's methodology was designed for a world where English-language Western media provides adequate coverage of global conflict. That world doesn't exist. The 88% Ukraine figure isn't a data error — it's a structural feature of a system optimised for the media landscape of the 1990s. The fix isn't auditing more carefully. It's rebuilding the collection layer.