US consumer product recall intelligence (monthly)

Every U.S. Consumer Product Safety Commission product recall announced in the trailing 365 complete days, ranked by a documented consumer-risk composite. The official keyless CPSC Recall API returns the full recall index in one call; recalls are filtered to the trailing 365-day window ending on the trailing complete UTC day. Free-text hazard narratives are classified into a closed 18-category hazard taxonomy by deterministic ordered keyword rules (first match wins); unit counts are parsed from strings like 'About 324' and 'About 1.2 million'; the recalling firm resolves through a Manufacturers -> Importers -> Distributors -> Retailers chain with suffix-stripped canonical firm keys; remedies normalize from the structured RemedyOptions field; manufacturer countries map to ISO alpha-3 via hub.normalize. The risk_score is a documented 0-100 composite: 45% hazard-severity weight (1-5 scale from the hazard category, pinned to 5 when the narrative mentions death) + 30% min-max-normalized log10(1 + units recalled) + 15% injuries-reported flag + 10% children's-product flag; risk_rank orders by risk_score desc (ties: units_recalled desc, recall_id asc). ConsumerContact phone numbers are excluded from the schema; rows whose remaining text trips the phone-PII pattern are dropped. Primary key: recall_id. Cadence: monthly. Nullability: units_recalled is null when the unit string is unparseable; mfr_country_code is null when unresolvable. Caveats: the hazard and product-family taxonomies are first-match heuristics over free text, not authoritative classifications; incident counts are digit-extracted from narrative text and are approximate; CPSC recall announcements are US federal public-domain works (17 U.S.C. 105).