Top AI research institutions (weekly)

Weekly ranking of the institutions driving AI research, from the official keyless OpenAlex API (https://api.openalex.org, CC0 data). Trailing 30 complete days: works tagged with the Artificial intelligence concept (C154945302) are grouped by institution (top 200), and a second all-works group_by over the same window (batched, with a single-count fallback) supplies each institution's total output so an AI share of its own production can be computed. Institution identity is resolved on OpenAlex's canonical ids (ROR-linked), so one university never appears twice; countries are normalized to ISO alpha-3 via hub.normalize; the OpenAlex institution type is carried as a coarse classification. The impact_score is a documented 0-100 composite = 50% min-max-normalized AI works per day + 50% min-max-normalized AI share of trailing-30-day output, and institution_rank orders by impact_score descending (ties: AI works desc, then OpenAlex id). as-of stamping is day-granular (window end), so re-running inside the same window is a no-op. Columns: ISO week of the window end, as-of date, window start/end, OpenAlex id / institution URL / name / ROR, ISO alpha-3 country code and name, institution type, AI works and total works in the window, AI share (%), AI works per day, impact score (0-100), institution rank. Primary key: (week, openalex_id). Cadence: weekly. Nullability: country_code, ror, homepage_url may be empty when OpenAlex has none; ai_works, total_works, ai_share_pct, impact_score and institution_rank are never null. Caveats: OpenAlex concept tagging is automated — bulk mis-tagging can inflate a few rows (see the facility-type rows near the top); counts are full (not fractional) attributions, so co-authored works credit every institution; ai_share is clipped at 100% if the API's two sweeps ever disagree. Sample use: order by institution_rank for the week's leading AI research producers, or filter country_code = 'CHN'.