One event. Many recall numbers.

A recall event and a recall number sound like the same thing, counted twice. They are not. FDA’s own definitions treat them as two different objects, openFDA keeps a field for each, and only one of the two is the identifier every downstream count actually uses.

What FDA calls an event

FDA’s own Enforcement Report Information and Definitions page defines it in plain terms: an event is a firm’s recall of one or more products. One root cause, one corrective action, and — in FDA’s own description — potentially more than one product carried under it. A firm correcting a single manufacturing defect that reached three catalog numbers, or two lot ranges packaged under different model numbers, can report all of it as a single event.

Why that doesn’t mean one row

Each product line inside that event gets its own recall_number in openFDA’s device enforcement data — a separate record, its own product description, its own copy of the reason-for-recall prose, usually identical across the group. The event_id field ties them back together, and openFDA publishes it on every record. But recall_number, not event_id, is what FDA Radar’s own ingestion treats as the record’s identity — the value stored as the item’s sourceUid, the value the database’s unique constraint checks before writing anything. FDA Radar’s own facts do keep event_id — it rides along on every enforcement item — but nothing downstream groups by it. A three-product event surfaces as three items in the review queue and three lines in a digest, each one carrying the same event_id in its facts and nothing that says these three are one thing.

Where the miscount happens

Ask how many recalls hit a product code this quarter, and the honest answer depends on which of those two numbers gets counted. Counting recall_numbers answers a real question — how many distinct product lines needed a fix — and it will always be the larger figure. Counting distinct event_ids answers a different real question — how many separate corrective actions did a firm actually take. Conflating them is how a quarter that looks like six recalls turns out to be two events that happened to touch three products apiece, or six genuinely unrelated ones. Both are true numbers. They are not the same number, and nothing in the record makes that obvious — the grouping only shows itself if someone actually collects the event_id values and checks which ones repeat.

The practice

Decide which count is being asked for before running one. If the question is exposure — how many of a subscriber’s own products are implicated — recall_number is the right unit, because that is what maps to a specific product description and a specific set of facts. If the question is enforcement activity — how many times has this firm had to act — group by event_id first, or one corrective action gets counted once for every catalog number it happened to cover. Per what FDA Radar monitors and how often, this is regulatory intelligence about how the record is built, never a determination about what a specific count means for a specific portfolio.