Corporate R&D publication velocity (monthly)

Monthly enrichment signals measuring corporate research-output velocity for 49 US and Canadian companies. For the trailing 24 complete calendar months, monthly work counts are pulled from the official keyless OpenAlex API via authorships.institutions.ror group_by sweeps (batched, with a single-count fallback), plus a parallel sweep restricted to the Artificial intelligence concept (C154945302) for the AI share. Each company is resolved by hand to its canonical ROR record (shared curation with hn_company_sentiment; 31 dictionary companies with no ROR record — Palantir, Cloudflare, Shopify, etc. — are excluded, never guessed), then transformed: works (monthly count, full attribution), works_3ma (3-month moving average), works_12m (trailing-12-month total), yoy_pct = 100*(works[m]-works[m-12])/works[m-12] (null when the prior-year month is zero or outside the window), z_12m = (works[m]-mean(prior 12 months))/pstdev(prior 12 months) (null with <12 prior months or a flat baseline), ai_works_12m / ai_share_12m (trailing-12-month AI-concept works and share, clipped at 100%), rd_score = 100*(0.50*min-max(clip(z_12m,-3,3)) + 0.50*min-max(clip(yoy_pct,-100,300))), min-maxed within each month across companies, rd_rank (score desc; ties: works desc, company_ror asc), spike_flag = 1 when z_12m >= 2.0. Who joins this: an R&D strategy / corporate-benchmarking team joins monthly research-output velocity per ticker to their R&D spend and talent-pipeline tables on ticker + month; a competitive-intelligence analyst joins ai_share_12m to technology-bet trackers on company_name + month. Primary key: (as_of, company_ror, month); join keys: ticker, company_name, as_of, month, country_code (ISO alpha-3). Nullability: ticker null for private companies (OpenAI, Anthropic, Databricks, ...); yoy_pct / z_12m / rd_score / rd_rank null where the trailing history is unavailable; ai metrics null for the first 11 window months. Caveats: rd_score is within-month relative, not comparable across months; OpenAlex concept tagging and institution affiliation are automated and can mis-attribute a few works; works are full (not fractional) attributions, so co-authored works credit every company; only the HQ-country ROR record is queried per company, so output tagged to foreign subsidiaries is not included. Underlying metadata is CC0 1.0 (OpenAlex) and CC0 1.0 (ROR), so commercial_use = yes. Sample use: filter month = '2026-08' and order by rd_rank for the latest R&D velocity leaders, or filter ticker = 'NVDA' for Nvidia's 24-month research-output trajectory.