Our method
No observer is completely impartial: every outlet and analyst has a perspective. We aim to make our approach symmetric, explicit, and open to scrutiny.
How a story is analyzed
Coverage is collected from a registered set of outlets across regions, preferring domestic editions in the local language where available. Articles about the same event are compared using the same criteria: headlines, actors and affected parties, causal explanations, evaluative language, quoted voices, omissions relative to other articles, tone, and prominence. The sources and sampling limits accompany each analysis.
What the sections distinguish
Shared accounts are separated from disputed claims, with attribution retained. Agreement among sources does not establish truth or independence. Coverage, framing, gaps in the sample, and the analyst’s assessment are distinct layers. Assessments identify uncertainty and what could change the conclusion.
Two concepts: divergence and change
Current divergence — a property of one snapshot: how far the accounts differ in this analysis, on a scale of 0 to 10.
Change — an observation over time: what shifted since the previous snapshot, described in words rather than a score.
These concepts are deliberately kept separate. Each divergence score is assessed independently, so differences include variation in the assessment itself. The score history is available on the story page, but it is not a calibrated measure of movement. Actual changes are described in words.
Ownership is context, not a verdict
State and private outlets are examined with the same language and skepticism about the interests of their owners and audiences. For each group, the analysis considers the strongest reasonable explanation for why an informed reader in that society would frame the event that way.
Each analysis receives a separate review
The counter-review checks asymmetry, opinion presented as fact, Western-default assumptions, missing perspectives, overstatement, conflated outlets, translation loss, and false balance. Its score is displayed even when low. The review evaluates the analytical procedure; agreement between models does not prove an event is true.
The attention map and its limits
Regional top-story feeds indicate whether and where a story appeared among their selected headlines. The story page shows recorded appearances only. A missing region does not prove missing coverage. Failed or partial sampling leaves coverage unknown. Even a successful sample with no match describes that sample, not everything the region’s outlets published.
Outlet profiles
Profiles use accumulated observations from analyzed stories, each tied to a quoted article. They describe patterns in the sample rather than ratings or recommendations to trust or distrust. The current article takes precedence over a profile.
What this analysis cannot establish
Blocked or paywalled articles may contribute only headlines or excerpts. Political labels are shorthand and may be outdated. Models can share errors and linguistic blind spots. Original quotations remain evidence in their source language; translated prose is a derived rendition requiring separate review.