How this archive is collected, and what it cannot tell you

The request

Once a day we make exactly one call: https://api.fda.gov/food/enforcement.json?limit=100&sort=report_date:desc. We store one row per recall per day, and we never edit a sealed row. Nothing in this pipeline uses a language model.

What we store, and what we deliberately do not

Each sealed row keeps only these fields: status, classification, product_description, recalling_firm, recall_initiation_date, report_date, state. The recalling firm is a business named on a public federal enforcement record, so it is published. Consignee-level detail that openFDA also returns — street addresses, cities, postal codes, distribution patterns, lot and code information, and the free-text reason for the recall — is excluded at the reader, not hidden in a template, so no page on this site can reach it. That exclusion is why you will find no ‘recalls by hazard’ page here: the hazard lives in the free-text reason field we do not publish.

How we identify a recall

Not by recall number alone. openFDA seals the literal string “N/A” for records with no number assigned, and several unrelated products share it, so a recall is keyed on its number and its product description when the number is missing. Grouping on the bare number would show unrelated products as one recall.

How we count a change

11 status changes and 0 classification changes, both out of 260 recalls sealed more than once — not out of 260. A change is only witnessable in a record we read twice, and because the window slides, a recall can leave and re-enter it. Every change is reported as a pair of seal dates rather than a single date, because our cadence cannot resolve the moment inside that bracket.

What this archive is not

It is not a complete record of US food recalls, it is not a regulatory record, and it is not a substitute for the FDA enforcement report. It is a dated log of what one public API served us each day, published so that a claim about how the record moved can be checked against the seals that back it.

Every fetch behind this archive

Every seal below is one call to the exact request we make. The digest is the first characters of the SHA-256 of the response body, so a reader can see for themselves which days returned the same file.