Seven government recall feeds, and what it takes to make them agree

Short summary

A detailed engineering deep-dive into aggregating seven government product recall feeds (openFDA, EU Safety Gate, RappelConso, NHTSA, CPSC) into a unified schema. The hard parts aren't schema mapping but semantic disagreements about what constitutes a recall, identifier extraction from free-text fields, API pagination quirks, and a production outage caused by shared Cloudflare Worker budget limits across independent jobs.

• •Seven government recall sources have fundamentally different record models (per-product vs per-event vs per-campaign)
• •Identifiers must be regex-mined from free-text fields and validated via check digits; invalid IDs must be rejected not answered
• •Production outage lesson: independent ingestion jobs sharing one Worker budget can silently fail with no exception or finally block

Generated with AI, which can make mistakes.