GDELT global news attention & tone (weekly)

Weekly country-level intelligence from the GDELT 2.1 global event database: every country generating >= 5 root events over the trailing 7 complete days, scored for news attention and instability. Deterministic pipeline: keyless GDELT 2.1 fifteen-minute update files (https://data.gdeltproject.org/gdeltv2/, terms allow unlimited academic/commercial/governmental use with citation + link); root-event dedupe (IsRootEvent=1 + GLOBALEVENTID dedupe); FIPS 10-4 country codes resolved to ISO alpha-3 via an embedded crosswalk (unknown codes dropped loudly) with canonical names from hub.normalize; per-country aggregation of event volume, article coverage, mean tone (AvgTone -100..100), Goldstein scale (-10..+10), conflict/cooperation shares from QuadClass, and the dominant CAMEO root event category. attention_score = 100 * min-max(log1p(event_count)) ranked as attention_rank (score desc, event_count desc, country_code asc). instability_score = 100 * (0.50 * conflict_share + 0.30 * negative_tone_share + 0.20 * clip(-mean_goldstein/10, 0, 1)). Week-granular as-of stamping (fetched_at = window-end midnight UTC): a second ingest of the same window is a true no-op. Primary key: (week, country_code). Nullability: no nulls are emitted. Caveats: GDELT coverage expands over time and varies by language, so absolute volume is media attention as much as event count; tone is a dictionary measure of sentiment, not sentiment itself; the >= 5 event gate drops quiet countries. Only event metadata and aggregates are stored — no article bodies, no per-row source URLs. GDELT terms require citation of the GDELT Project and a link to https://www.gdeltproject.org/. Sample use: order by attention_rank for the week's most-covered countries, or filter instability_score >= 50 for elevated-conflict coverage.