Product recalls live on four different authority sites, in three languages, in formats that don’t line up. Pulling “what got recalled, why, and what to do about it” into one comparable shape means reading CPSC, FDA, the EU Safety Gate, and Japan’s Consumer Affairs Agency separately. This dataset does that reading and normalizes the result.
It covers a rolling three-month window of recalls across cosmetics, baby products, and food — 30 records in v0.1 — with each authority’s notice mapped onto the same fields.
What’s in it
- 30 recalls across cosmetics, baby products, and food.
- Four authorities — US CPSC and FDA, the EU Safety Gate, and Japan’s Consumer Affairs Agency.
- Per record: official recall ID, hazard type, manufacturer, remedy, and a
current_stateflag.
How it’s sourced and kept current
Each record cites the official authority page it came from, with a verification date. The history is append-only: a recall’s timeline grows as the authority updates it, rather than being overwritten, so you can see how a case moved. Updates come from weekly scans with a human review before anything is committed.
When to use it
- Compliance and trust-and-safety — a normalized cross-authority view instead of four separate watch tabs.
- Marketplace risk tooling — match recall IDs and hazard types against listings.
- Cross-border retail ops — one timeline that spans US, EU, and Japan notices.
Get it
- Free on Hugging Face (CC BY 4.0): dataset page.
A sourced, rolling snapshot for reference — not legal advice and not an exhaustive recall registry. Confirm any specific recall against the issuing authority.