The Recall Count and the Recalled Poundage Have Stopped Telling the Same Story

Every few weeks a cluster of food recall headlines arrives together — cheese, leafy greens, eggs, a shelf-stable soup — and the framing writes itself: the food supply is getting less safe. That framing is intuitive and it is testable. Testing it requires separating three series that headlines collapse into one: how many recall events regulators announce, how much product those events cover, and how many people actually get sick.

Through 2025 and into 2026 those three series stopped moving together. One is roughly flat. One has multiplied. The third is measured through instruments that changed so much mid-decade that its own custodians warn against reading it as a trend. That divergence, not the raw headline count, is the part with structural content.

The Surface Reading: A Count That Is Not Actually Rising

Start with the number the headlines implicitly reference. For meat, poultry, and processed egg products, the Food Safety and Inspection Service published 42 recall cases in calendar year 2025 — a figure in the dozens, not the hundreds, and not a runaway one by two-decade standards.

Composition is more informative than the total. Of those 42 cases, 38 were Class I — the classification reserved for a reasonable probability that consumption causes serious health consequences or death. Stated causes break down as 13 foreign material, 9 undeclared allergens, 7 production without inspection, 5 import violation, 4 Listeria monocytogenes, 2 unapproved substance, 1 STEC, and 1 misbranding.

The two largest categories — foreign material and undeclared allergens — are not pathogen events. They are process-control and labeling failures. A recall triggered because an allergen was omitted from a panel enters the count with exactly the same weight as one triggered by a confirmed outbreak strain. Any argument resting on the count alone treats those as equivalent.

On the FDA side the count did rise, but modestly. Quarterly data compiled from agency enforcement reports showed 415 FDA food recall events through the third quarter of 2025 against 363 in the comparable 2024 period — roughly 14 percent. A real increase, nowhere near the scale the coverage implies.

The Structural Cause: Pounds per Event, Not Events per Year

The series that actually moved is volume. Over the same window, FDA recall events covered about 109.74 million units through Q3 2025 against 45.02 million a year earlier — roughly 144 percent, while the event count rose 14 percent. The FSIS figures are starker: 30 events covering 59.99 million pounds through Q3 2025, against 28 events covering 7.91 million pounds a year earlier. Two more events. Roughly seven and a half times the poundage.

The quarterly detail shows how this happens. FSIS recorded 15 recall events in Q2 2025 covering 1.04 million pounds, then 6 events in Q3 2025 covering 58.52 million pounds — the largest quarterly volume in that program in more than thirteen years. Fewer events, an order of magnitude more product. One upstream node failing now drags a volume of finished goods that would once have required dozens of separate failures.

This is a distribution-structure signal, not a hygiene signal. Consolidation in processing and co-manufacturing means one facility, one ingredient lot, or one contract packer sits underneath many brands and many retail banners at once. When that node is implicated, the perimeter is drawn around everything downstream, because drawing it tightly around the affected lots requires lot-level traceability most of the chain does not have. The perimeter is wide because the information is thin, not necessarily because the contamination is broad.

Detection Has Split Into Two Tracks Moving in Opposite Directions

The standard rebuttal to recall alarmism is that detection improved. It is half right, and the wrong half matters more in 2026 than it did five years ago.

The genomic track improved and keeps improving

Whole genome sequencing is now the standard PulseNet method for Salmonella, STEC, Vibrio, and Listeria, with the full transition for routine surveillance completed around 2019. CDC's framing of the effect is concrete: identifying an outbreak once took up to 39 days; it now takes roughly 16 days.

That capability produces not only faster detection but a different kind — clusters that are geographically scattered and temporally stretched. The soft cheese Listeria investigation posted in June 2026 illustrates it. Twelve confirmed illnesses across four states: Illinois, Maryland, New York, Virginia. Ten hospitalizations, one death. Onset dates spanning March 6, 2023 through June 2, 2026. Six product samples and two environmental samples matched the outbreak strain. Recalls followed on June 3, June 5, June 18, and June 26, 2026.

Twelve cases over three years and three months across four states is not something a pre-genomic system detects. Without sequence-level matching, those twelve stay twelve unrelated sporadic illnesses and no recall issues at all. Each such investigation converts previously invisible background illness into a visible recall event — a detection artifact in the strict sense, and simultaneously a real safety gain, because the facility gets cleaned.

The physical inspection track moved the other way

Genomic surveillance tells regulators that something happened. It does not prevent it. Prevention runs through facility inspection, and that series is going the wrong way. FDA foreign food facility inspections peaked at about 1,727 in 2019. In 2024 the figure was roughly 1,410. In 2025 it fell to just under 1,000 — the lowest annual count since 2011.

Set that against the statutory frame. The 2011 Food Safety Modernization Act set an escalating foreign inspection mandate that reached 19,200 facilities per year. FDA's own internal assessment of need is far lower — roughly 4,700 annually — and even that is missed by a factor of nearly five. The gap shows up case by case: the facility examined in a 2026 cyclosporiasis investigation had last been inspected in 2019, and before that in 2013.

So detection is not one dial moving up. It is two dials moving opposite ways. Post-hoc linkage capability is rising; pre-event verification is falling. A recall count generated by that combination is increasingly weighted toward events found after people were already sick, and decreasingly toward events found before product shipped.

The Traceability Layer Moved From January 2026 to July 2028

The mechanism that would narrow recall perimeters already exists on paper. The FSMA Section 204 Food Traceability Rule — docket FDA-2014-N-0053, RIN 0910-ZC21 — requires records for seven critical tracking events: harvesting, cooling, initial packing, first land-based receiver, shipping, receiving, and transformation. Covered entities attach a traceability lot code and must produce records to FDA within 24 hours of request.

That is the capability that lets a firm recall three lots instead of a quarter's output. Its compliance date was January 20, 2026. A proposed 30-month extension published August 7, 2025 moved it to July 20, 2028, and Congress subsequently directed the delay through appropriations legislation. FDA's analysis of the extension estimated annualized net cost savings of roughly $54 million at a 7 percent discount rate, with $16 million to $22 million accruing annually to small entities.

The stated reasons for the delay are themselves the diagnosis. Required data elements are not routinely maintained or shared along supply chains, systems are not interoperable, and distributors cannot reliably obtain lot codes from suppliers. For another two-plus years, then, the default response to an implicated node remains the wide perimeter, and recalled poundage per event is likely to stay structurally elevated through at least 2028 regardless of underlying contamination rates — a forecast about record-keeping, not hygiene.

What the Illness Data Can and Cannot Settle

If recalls are an unreliable safety proxy, the obvious move is to check illness incidence instead. That series has its own instrumentation problem, and CDC says so explicitly. FoodNet covers ten sites across roughly 15 percent of the U.S. population, about 51 million people. The 2024 preliminary crude incidence figures, per 100,000 population, are below.

Pathogen	2024 incidence per 100,000	Versus 2016–2018
Salmonella	18.0	Similar
Shigella	7.9	Increased
STEC	7.1	Increased
Yersinia	3.5	Increased
Vibrio	1.4	Increased
Cyclospora	1.4	Increased
Listeria	0.4	Similar

Taken at face value, that table refutes the comfortable version of the detection argument: reported infections in 2024 increased or were similar relative to the 2016–2018 baseline. Nothing there shows illness falling.

But the measuring instrument changed underneath the series. A 1996–2023 analysis of FoodNet data found that by 2023, 78 percent of bacterial infections in the historic catchment were diagnosed by culture-independent diagnostic tests and 46 percent by CIDT only, with no culture attempted — against 49.5 percent CIDT use in 2016–2018. Because CIDTs detect infections culture-based workflows missed, reported incidence rises without any change in true incidence. Reported unspeciated infections rose by factors of 7.19 for Yersinia, 4.29 for Vibrio, 2.33 for Shigella, and 1.31 for Campylobacter. CDC's conclusion is that apparent lack of progress toward reduction goals may reflect changing diagnostic practice rather than a real increase.

The 2024 data carry a further discontinuity: from 2024, FoodNet collected only culture-confirmed Campylobacter, making that pathogen non-comparable to prior years and excluding it from trend modeling.

The honest summary is uncomfortable and worth stating plainly: neither the recall series nor the illness series is currently clean enough to settle whether the food supply became more dangerous. Both are contaminated by measurement change in the same direction — upward. The correct posture is agnosticism on the level and attention to the composition.

What the Market Misses

Three blind spots follow, and none are about whether food is safe.

First, the count is priced and the poundage is not. Coverage volume tracks announcements and brand recognizability. Financial exposure tracks pounds, retail value, and customer-list width — which, as the Q3 2025 figures show, can move nearly independently of the count. A quarter with six recalls can carry more balance-sheet consequence than one with fifteen.

Second, the trigger type is ignored. A recall from routine environmental monitoring, one from a genomic cluster investigation, and one from a labeling audit are three different events with three different implications. Only the second implies illness already occurred; only the third is largely a documentation failure. The count treats them identically.

Third, the fragmentation of oversight is treated as background noise rather than as a variable. GAO has kept federal food safety oversight on its High-Risk List since 2007, describing at least 30 federal laws administered by 15 federal agencies, with six recommendations and five matters for congressional consideration unaddressed as of January 2025. GAO puts the annual economic cost of foodborne illness at roughly $75 billion in 2023 dollars, against CDC-derived estimates of about 10 million illnesses, 53,300 hospitalizations, and more than 900 deaths per year from six pathogens. Which agency owns a commodity determines inspection cadence, recall classification vocabulary, and disclosure speed — an accident of statutory history, not of risk.

What Would Prove This Wrong

This frame treats the count as a measurement artifact and the poundage as a distribution artifact. Several observable conditions would break it, and they deserve equal weight.

• Poundage per event falls while the count holds. If median pounds per FSIS recall event returns toward the pre-2025 range without a change in traceability capability, the consolidation explanation is weaker than argued here.
• Pathogen-cause recalls rise as a share of the total. The 2025 FSIS mix was dominated by foreign material and undeclared allergens, with only five pathogen-contamination cases across Listeria and STEC. If that share inverts across two consecutive years, a genuine microbiological deterioration becomes the better explanation.
• Illness incidence rises in culture-confirmed subsets. If incidence rises where the diagnostic method did not change, the detection-artifact argument loses its main support. The 2024 Campylobacter switch to culture-only collection creates exactly this test series from 2025 onward.
• Recall perimeters narrow before July 2028. If firms adopt lot-level traceability ahead of the compliance date and recall volumes fall accordingly, the regulatory-timing argument is overstated.
• Inspection counts recover without a change in recall composition. If foreign facility inspections return toward the 1,300–1,700 range and the recall mix does not shift, the inspection channel matters less to observed outcomes than assumed here.

There is also a scenario in which the alarmist reading is simply correct. Falling pre-market verification, delayed traceability, and rising volume per event could compound into genuinely higher realized risk, with measurement noise obscuring a real deterioration rather than manufacturing a fake one. Nothing in the current data rules that out.

Concrete Framework — Practical Monitoring

1. Track two numbers, never one. Each quarter, record recall events and recalled pounds or units separately for FSIS and FDA. The ratio is the working indicator; the count alone is not.
2. Classify every recall by trigger. Routine testing, environmental monitoring, illness-cluster investigation, labeling or allergen audit, import or inspection violation. Only the illness-cluster bucket is a lagging safety indicator.
3. Log the recall class. Class I share is the severity control. FSIS reported 38 of 42 CY2025 cases as Class I; a falling Class I share alongside a rising count means the count is being padded by low-severity events.
4. Watch the inspection series annually. FDA foreign facility inspections against the 19,200 statutory figure and FDA's own 4,700 need estimate. Sustained readings under 1,200 indicate the prevention channel is not being restored.
5. Read FoodNet with the diagnostic footnote attached. Compare only within consistent diagnostic regimes, and treat any pathogen whose collection method changed — Campylobacter from 2024 — as a broken series until three post-change years exist.
6. Calendar the traceability milestones. July 20, 2028 is the operative compliance date, and FDA has committed to quarterly stakeholder engagement in the interim. Each engagement is a checkpoint on whether the date slips again.
7. Separate firm-level from sector-level conclusions. Repeat recalls at one facility carry information about that facility; a rise in aggregate counts does not transfer to any individual producer.

What to Watch Next Week

• Weekly FSIS recall and public health alert postings. Record the poundage line and the recall class, not the headline. One high-poundage Class I case changes the quarterly picture more than five small ones.
• FDA enforcement report updates. Classifications are assigned after the initial announcement; the announcement-to-classification gap measures how fast the system is processing.
• Status changes on active outbreak investigation pages. The soft cheese Listeria investigation was still listed as ongoing after the June 2026 recalls; case counts typically move for weeks after an initial posting.
• Any FDA statement on traceability implementation cadence. The 2026 priority deliverables commit to quarterly engagement on the Food Traceability Rule; the tone of those updates is the earliest signal on whether July 2028 holds.
• Inspection dashboard updates. Monthly foreign facility inspection counts are the cleanest read on whether the prevention channel is recovering or still eroding.

The disciplined conclusion is narrower than either headline position. Recall counts are a detection-and-distribution metric before they are a safety metric, and in 2025 the count barely moved while the volume multiplied. That is a supply chain structure story. Whether the food supply itself became more or less dangerous is a question the current instrumentation cannot answer cleanly in either direction — and saying so beats picking a side the data does not support.

This article is analysis of public regulatory and surveillance data. It is not investment, legal, medical, or food safety compliance advice.