Data limitations

Fog of war

During active conflict the same incident is described differently within hours by different sides, and the descriptions change over the following days. First-day casualty counts are routinely revised upward or downward by 50% or more. Treat any live counter on this site as a snapshot of the reporting environment at that moment, not as ground truth.

Source-availability bias

We can only see what is published. Areas with internet shutdowns, embedded journalists, expelled press corps, or active military operations are systematically under-reported. The absence of events for a region on a given day usually means we could not collect, not that nothing happened.

Language bias

• Our pipeline is heaviest on English-language sources, with secondary coverage in Arabic, Persian, Hebrew, Russian, Chinese, and Turkish via NewsData.io.
• Stories that break first in less-covered languages reach our system later than English-language stories, sometimes by a day or more.
• Translation introduces drift. Where we quote a foreign-language source, the original wording may carry connotations the translation loses.

Platform bias

• X (Twitter) coverage is biased by what X surfaces to its API. Suppressed, shadow-banned, or geo-restricted posts are invisible to us.
• Other platforms (Telegram, VK, WeChat, Facebook, TikTok) are not currently in the ingestion pipeline. Conflict reporting that lives primarily on those platforms is under-represented.
• Algorithmic feed bias affects what we see — virality is not the same as accuracy.

State censorship risk

• State-owned outlets do not publish material that contradicts their government. Their silence is a data point.
• Some governments require domestic press to use specific framings. Stories that originate in those press environments inherit those constraints.
• Independent outlets operating inside authoritarian environments may self-censor.

API and pipeline outage risk

• Our data depends on third-party APIs (CurrentsAPI, NewsData.io, GDELT, ACLED, ScrapingBee, X, oil-price feeds). Any of these can fail.
• When a feed is down, the corresponding section of the site falls behind. We try to surface this honestly rather than show stale data as fresh.
• Pipeline health is monitored, but readers should treat a sudden change in event volume as a possible pipeline issue rather than a real-world signal.

Delayed verification risk

Verifying a claim independently takes time. New events sit at Unverified or Partial by default. The label is not a value judgement about the story — it is a status. Some unverified items will turn out to be true; some will turn out to be wrong.

Historical revision risk

Records on the site are not frozen. If new evidence resolves a contradiction, the verification status of the original event changes and a correction note is added. Citing this site, please:

• Include the URL.
• Include the date you accessed it.
• Re-check if the figure or status is load-bearing for your work.

Casualty figures are uncertain

• Casualty counts come from named sources. They are not the site's independently produced numbers.
• Different sources use different methodologies (confirmed vs. reported, civilian vs. combatant definitions, hospital intake vs. on-scene counts).
• We do not sum incompatible methodologies into a single figure.

Prediction and simulator outputs are not forecasts

The Predictions and Simulator features present scenarios and probability ranges. They are not forecasts and they are not advice — see the disclaimers on those pages. They are most useful as a way of structuring "what would need to happen for this to escalate / de-escalate", not as an answer.

Geolocation precision

• Event locations are taken from source reporting. Where multiple sources disagree on the exact location, we use the most specific common element (city, province, country) rather than picking one.
• Coordinates, where shown, are approximate unless explicitly geolocated against satellite imagery.

What we are working on

• Improved non-English language coverage.
• Tighter event clustering across syndicated wires.
• More structured casualty attribution.
• Better surfacing of pipeline health on the front end.

What this page is for

This is the page we point readers to when a number on a dashboard surprises them. The most truthful thing we can say about any specific figure is: here are the sources we drew on, here are the rules we applied, here is what could still be wrong. That is what these methodology pages exist to provide.