Every so often someone asks me why a recall they saw all over the news barely exists in FDA’s files, or why a recall they stumbled on in an FDA database never made the news at all. The answer is that FDA keeps two recall lists. They are built differently, they are filled in by different people, and only one of them is complete.

The page most people mean

When people say, “the FDA recall page,” they almost always mean Recalls, Market Withdrawals, & Safety Alerts. FDA describes it as information gathered from press releases and other public notices, and every entry on it carries the same disclaimer: when a company announces a recall, FDA posts the company’s announcement as a public service, and FDA does not endorse the product or the company. FDA also says outright that not all recalls have press releases or are posted on this page. It is a bulletin board. If a company writes nothing, there is nothing to pin up.

The page almost nobody reads

The other list is the weekly Enforcement Report, published under 21 CFR 7.50. This one is FDA’s own, and it includes all recalls the agency monitors — Class I, II, III, or not yet classified. It is searchable back to June 8, 2012, exports to CSV, and has an API. It is also written in lot codes and batch ranges, it appears only after FDA works through classification, and no ordinary consumer has ever read it.

FDA is explicit on one point that matters here: classification and posting in the Enforcement Report should not be seen as an expansion or change to a firm’s voluntary public warning. Deciding a recall is Class I does not send anybody back to warn the public.

Here is what the split looks like in practice, using two recalls from this year.

The recall you saw

On September 6, Made Fresh Salads Inc. of Bay Shore, New York recalled all of its ready-to-eat deli style salads and cream cheese over possible Listeria monocytogenes contamination. Environmental sampling by FDA and the company had turned up positive areas in the facility. The products went to Brooklyn, Queens, the Bronx and New York City by direct delivery to retail stores and distributors, in 5-pound and 30-pound tubs and half-size aluminum pans, with expiration dates running from September 3 to September 18. Stores repack it, so as the company notes, retail packaging and coding vary by location. No illnesses have been reported to date.

That recall got the full treatment: a press release on the recall page, sixty-eight product lines named one by one, dozens of product photos, refund instructions and a consumer phone number. Whatever hazard class FDA eventually assigns it, the public already knows.

The recall you did not

Recall H-0909-2026 was initiated on May 6 by The Coffee Connexion Co., Inc. of Lebanon, Tennessee — 913 cases of alfredo sauce in 3-pound 7-ounce sealed poly bags, twelve bags to a case. The sauce contained dry milk powder caught up in the California Dairies Inc. bulk recall of April 20, which covered bulk powdered milk and buttermilk sold to wholesale distributors and manufacturers, reported at roughly 2.7 million pounds. The alfredo went to forty-one states. On June 4, FDA classified it Class I — the category for a reasonable probability of serious adverse health consequences or death.

The Enforcement Report record has a field for the press release. It reads: Press Release Not Issued For This Recall.

It is not on FDA’s Major Product Recalls hub for the California Dairies event either. That page lists eleven downstream companies — Ghirardelli, Utz, Sanfilippo, Pork King Good and the rest — and every one of them links to a company announcement, because the table is assembled out of announcements. The alfredo also carries its own event number, unlinked to the parent recall, so nothing in the database connects it to the milk powder except a sentence of free text in the reason field.

The difference is not what you would guess

It is not hazard. The alfredo is a federally designated Class I. The salads had not been classified at all.

It is not reach. The alfredo went to forty-one states. The salads went to four boroughs.

It is not whether a consumer could recognize the product. Neither is recognizable. Both are bulk packs that get repacked, plated or ladled onto something else before anyone eats them.

The difference is that one company wrote a press release and the other did not. FDA can issue its own public warning under 21 CFR 7.42(b)(2), but the agency reserves that for urgent situations where other means of preventing use of the product appear inadequate, and it almost never uses it. A firm that can telephone every consignee on its invoice list is treated as having taken care of the problem.

What this does to the numbers

Recall counts get thrown around constantly as proof that the food supply is getting safer or getting worse. Almost nobody says which list the count came from. They should, because the two lists cannot be compared.

They count different things. The press release page counts companies willing to publicize. The Enforcement Report counts recalls.

They count in different units. Made Fresh Salads is one press release covering sixty-eight product lines, and in the Enforcement Report product lines generate their own records. The alfredo sauce is a single product record inside a larger recall event. Depending on which page you use and what you treat as one recall, these same two matters can be counted as two, or as dozens.

They run on different dates. The press release page is dated when the company announced. The Enforcement Report turns on the classification date, which for the alfredo landed twenty-nine days after the recall started. A recall initiated in December gets classified in January and shows up in the next year’s tally.

And the lag always flatters the present. Any count of the current year is missing every recall still sitting in the queue unclassified. The recent past will always look calmer than it later turns out to have been.

When someone tells you recalls are at a multi-year low or an all-time high, three questions settle it: which database, counted in what unit, keyed to which date. Most of the time the person making the claim cannot answer any of the three.

What ought to change

A Class I recall means FDA believes the product can kill someone. There is no defensible reason for one of those to pass without public notice. Make the announcement mandatory for Class I or let FDA issue it when the company will not. Give downstream sub-recalls the parent event number so an ingredient recall can be followed to the end of the chain. And make press-release status a searchable field, so anyone can pull the list of Class I recalls the public was never told about.

I do not know whether anybody got sick from that alfredo sauce. Neither does anyone else, which is the point. Somebody who ate it in a cafeteria in May and spent a week very sick has no way to connect that week to a record that surfaced in June, in a database they have never heard of, on a page FDA does not expect them to read.