METHODOLOGY

How the map reads attention.

The scored-day attention score, live flares, tone shifts, latest brief and map card answer related but different questions. Scores compare a country’s latest complete day with its own history; live flares compare the current UTC day so far with the same hours on recent days.

WHAT WE MEASURE: Country mentions in global news from GDELT. The backend reads GDELT’s machine-readable news feed: article metadata, locations, themes, source domains and tone. It also includes GDELT’s translingual companion feed, so non-English coverage is included rather than treating the world as English-only. The main score compares each country’s latest complete scored day with its own recent history, so a country is never judged against a larger or louder country.
LANGUAGE: GDELT’s translingual feed converts monitored non-English coverage into English-coded article metadata. Evidence terms are built from those coded themes, with local cleanup for noisy taxonomy codes, generic phrases and raw name lists. Displayed headlines come from GDELT article records or source URL slugs, not from a separate AttentionFlare translation step, so they may still appear in the publisher’s language.
ATTENTION SCORE: Each country is scored against a 90-day, weekday-conditioned baseline. The score is the empirical percentile of coverage within that country’s own history, mapped to plain-language bands from quiet to intense.
LIVE NOW: Live Now is separate from the scored-day attention score. It compares the current UTC day so far with the same elapsed hours on recent days, catching surges before the next complete-day score lands.
TONE SHIFT: The tone card ranks the largest moves in average newsroom language between the latest 24-hour window and the previous 24 hours. Tone reflects newsroom language, not public opinion or sentiment.
DAILY BRIEF: The front page includes the latest brief below the shortened leaderboard. It is synthesised once a day from the latest scored-day country ranking, live flare counts, tone shifts, evidence terms and matched headlines already shown elsewhere in the app.
MAP CARDS: The map is the main exploration surface. Country cards combine the attention score, live mention count, 14-day coverage shape, the per-country brief, evidence terms and matched headlines.
EVIDENCE: Evidence terms and headlines are stricter than the score. Terms require recurring country-specific GDELT support across documents and sources. Headlines are shown only when the URL or source domain carries a visible country signal, so a questionable geocode does not become a public article example.
PER-COUNTRY BRIEF: Every map card carries a short brief explaining what the coverage for that country looks like in the latest scored cycle. Countries scoring elevated or higher (70 +) get an AI-written brief when there are enough country-specific evidence terms and matched headlines to summarize; otherwise they get the rule-based brief too. The badge on the card (AI or RULE-BASED) tells you which one you’re reading. Either way the underlying data is identical; the AI version is just better at putting it into words.
WHAT IT ISN'T: Not a poll. Not a national-mood index. Not comparable between countries in absolute terms: each is judged only against its own baseline.