A recall has three dates, and the pipeline had to pick one

Pull a device recall record from openFDA and the record does not answer “when did this happen” with one date. It answers with three, sitting in the same object, and none of them is labeled the date. recall_initiation_date, center_classification_date, and report_date can each be weeks or months apart on the same recall, and a pipeline that reads only the one it happens to sort on is quietly answering a different question than “when did this recall start.”

What each date actually marks

Recall initiation date is the closest thing to real time in the record: the date the recalling firm began notifying the public or its consignees that a correction or removal was underway. That is the firm acting, not FDA. Center classification date is a different event entirely — the date FDA’s center completed the health hazard evaluation required under 21 CFR 7.41 and assigned the recall its Class I, II, or III designation. FDA’s own enforcement-report documentation is direct about the order these usually happen in: classification is typically finished only after the firm has supplied FDA everything needed to judge the hazard, so classification date routinely lands after initiation date, sometimes by a wide margin. Report date is the date the record actually posts in FDA’s weekly Enforcement Report, per 21 CFR 7.50. Three dates, three different questions, and only the first one is close to what a reader means by “this recall happened on.”

Why the gap is not a data-quality bug

None of this is FDA moving slowly on paperwork. A firm can start pulling product before anyone has finished judging how dangerous the defect is, and FDA’s own report definitions describe exactly that case: a recall can post as Not Yet Classified, a real category, carrying a firm notification date and no classification date yet to sit beside it. The hazard evaluation takes the time it takes. The report catches up when it can, sometimes arriving in two pieces rather than one.

What a pipeline has to choose, and what it costs

A monitoring feed built on this data has to decide which of the three dates governs when a recall enters a window, and every choice trades something away. Sort by report date, the field most tools default to, and a recall that has quietly been pulling product for two months looks brand new to a downstream reader the day the classification finally posts. Sort by initiation date instead, and a recall can enter a “this week” query before FDA has said what class it is — unclassified, and therefore not yet actionable by severity. No single date among the three is both current and complete. The choice is which incompleteness a given use is willing to tolerate.

The practice

Before trusting a recall’s date to mean “this is when it happened,” check which of the three fields produced it. A dashboard sorted by report date is answering “what did FDA publish this week,” not “what started this week” — useful for staying current, misleading if read as a timeline of events in the field. A portfolio review asking whether a supplier or predicate had a recall in a given quarter should check initiation date specifically, since that is the field closest to when the firm actually acted, and it is the one most easily lost if a tool only surfaces whichever date it happens to sort by.

FDA Radar’s recall ingest keeps recall_initiation_date as an item’s effective date and report_date — falling back to center_classification_date when report date is absent — as its publication date: two fields kept separate rather than one field asked to answer both questions. Which of the three dates matters most for a specific supplier or predicate assessment is a judgment for whoever owns that risk file, made on the current record. This is a description of what three fields mean, not a determination about any recall in particular — regulatory intelligence and method, never regulatory advice on your own facts.