This page explains how the recall record is kept, so that anyone quoting it knows exactly what they are quoting. If something here is wrong, we want to know: the correction address is at the bottom.
What the record is
We read the official recall registers of three countries, keep a copy of every pet food entry we find, and go on keeping it after the authority takes it down. That last part is the whole point. Official recall pages are working documents: entries roll off, archives get frozen, links rot. A recall that has vanished from a government site has not stopped mattering to somebody whose animal ate the product.
We are a website, not a regulator. We do not investigate, we do not test products, and we do not decide whether a recall was justified. We keep what was published, with the date it was published and the register it came from.
Where each entry comes from
UK FSA pet food alerts. The Food Standards Agency’s pet food alerts, read every six hours. This is the whole register rather than a sample. Published under the Open Government Licence v3.0.
FDA animal and veterinary recalls. The recalls the US Food and Drug Administration currently lists for animal food, read twice a day from the spreadsheet the agency publishes beneath its table. It reaches back to January 2019. The agency’s own archive of older recalls is a frozen web snapshot rather than data, so nobody can read it as a list. US federal work, public domain.
FDA food enforcement reports. The pet food entries inside the FDA’s food enforcement dataset, read daily. This register is thin on purpose: most American pet food recalls are announced on the animal and veterinary pages above, not in this dataset. US federal work, public domain.
Canadian recalls and safety alerts. The pet and animal entries inside Canada’s open recalls file, read daily. Canada publishes every recall of every kind in one file, so we keep the food categories and then the rows whose title or product names an animal. Open Government Licence – Canada.
How the reading works
Each register is read on a schedule. Every entry is stored with a hash of its content, so a change to the wording of an existing recall is recorded as a change rather than silently overwritten. Every record page shows the date we first held it and the date we last saw it on the source.
Nothing is written by hand. Every figure on the recall pages is computed from the record when the page is built, which is why the counts on the search page and the counts on this page always agree.
What we do not claim
A recall is not a verdict on a business. It is what an authority published on a date. Most recalls are the system working: a company or a regulator found a problem and acted on it. Businesses with several entries are often the ones testing most carefully.
Finding nothing here does not mean a product is safe. We hold three countries. A product can be recalled by an authority we do not read, or sold under a name we do not hold. If your animal is unwell, phone your vet rather than trusting an absence on a website.
We do not rank or score businesses. The counts shown are what we hold, not a league table, and they are shaped by which registers we read and how far back each one reaches.
Mistakes, and how to get them fixed
Two kinds of mistake are possible. Either we misread a source, or the source itself has changed. Both are worth telling us about. One entry has already been withdrawn this way: a licorice recall matched our search because one of the recalled brand names contained the word rawhide, and it had nothing to do with pets. It was removed and its address now points back to the record.
If an entry about your business is wrong, write to us and we will correct it and say that we did.
Using the record
The record is readable by machine as well as by eye. There is a JSON copy of every entry, an RSS feed for watching new arrivals, and a sitemap of every record page. Each source keeps its own licence, named on the search page beside the register it belongs to. Attribution to Paw Multiverse is welcome and a link is better than a mention.
Take the whole record
If you would rather have the lot than query it a page at a time, the entire record is mirrored nightly on GitHub at multiverseadmin/pawmultiverse-open-data, as JSON and as CSV, under CC BY 4.0. The mirror runs at 04:20 UTC, checks every file before it publishes, and refuses to overwrite good data with an empty download, so a failed night leaves the previous copy standing. The repository carries a citation file if you need to reference the record formally.
