Detection got faster than tracing, and the produce buyer now owns the recall

Recall counts measure paperwork, not root causes. Jennifer McEntire of Food Safety Strategy makes the accounting point plainly: one contaminated ingredient used across many foods produces a separate recall for each finished product, so the total inflates while the number of underlying failures stays small ^[2]. Run April's dairy case through that logic and the multiplier is at least twelve public recalls per suspect batch ^[17]. Twelve brands did the pulling. One supplier made the mistake.

The finding side has genuinely improved. Public health laboratories have spent the past five to ten years using whole-genome sequencing to match Salmonella and E. coli cases that would once have looked unrelated ^[4]. Melanie Firestone, an epidemiologist at the University of Minnesota who studies surveillance, says some of what reads as more illness is cases that used to be missed ^[5]. The practical effect for a buyer is timing: a scatter of patients becomes a named outbreak attached to a named commodity before anyone has a grower, a lot, or a shipment date.

The biggest event of 2026 is the one that machinery cannot see. Cyclospora will not grow in a lab, so sequencing is largely unavailable for it and investigators fall back on asking sick people what they ate ^[12]. That method returns "iceberg lettuce." It does not return a farm. Across the confirmed cases it produced a 4.2 percent hospitalization rate ^[16], which is a serious public health signal and, at the same time, an advisory a food-service buyer can only act on by dropping the category.

Behind both of those sits the response capacity. Business Insider attributes the summer's appearance to better detection layered on long-standing weaknesses in the food system, compounded by cuts to the public health infrastructure built to catch problems ^[18]. Craig Hedberg's description of modern sourcing explains why a single lapse travels: foods no longer go out of season, buyers simply go to whichever growing region is producing, and long chains with large distribution networks amplify failures ^[6]. So the sensitivity of the alarm rose while the ability to localise what set it off did not. The difference gets settled at the only point in the chain with a consumer-facing package.

It is settled twice, in fact. Genetic linkage does not only identify outbreaks; it converts strangers with similar symptoms into a defined, documented cohort, which is the raw material of a claim. Bill Marler, 33 years into representing people sickened by food [c12a], says of this summer: "It's real, and it looks bad" ^[15]. That verdict and Hedberg's "not all that unusual" can both hold if what changed is the resolution of the picture rather than the safety of the food. Operators do not get to buy on the epidemiology, though. They buy on the advisories, and the advisories now arrive earlier and less specific than the contracts behind them assume.