Industries / Retail, hospitality and personal services / Wholesale trade / Foodservice distribution
Food safety and recalls
Tracing a contaminated food back to where it came from and forward to every kitchen that received it, and getting it out of those kitchens when it is recalled.
4
Milestones achieved, from 2005 to 2012.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2005
One supplier, 44 schools
In September 2005 E. coli O157 spread from cold cooked meat supplied by John Tudor and Son, a butcher in Bridgend, Wales, to schools in four council areas. More than 150 people fell ill, most of them children in 44 schools, 31 were admitted to hospital and a five-year-old boy died, by the Scottish Craft Butchers' summary of the public inquiry; Food Manufacture gave 118 confirmed cases. Professor Hugh Pennington's inquiry reported in March 2009 that the councils' process for awarding the school supply contracts in 1998 and 2002 "was seriously flawed in relation to food safety", and made 24 recommendations; the owner, William Tudor, was jailed for 12 months. From November 2014 Wales extended its food hygiene ratings to businesses that supply other businesses, so that public buyers could look up a supplier's rating.
LessonWhen one contract supplies the schools of a whole area, one supplier's failure reaches all of them.
Credit
Professor Hugh Pennington; the Welsh Assembly Government; the Food Standards Agency
2008
Recalled beef reached 7,000 districts
In January 2008 an animal protection group released a video of workers forcing cows that could not walk to stand at the Westland/Hallmark plant in Chino, California, a supplier of beef to US school meals. The US Department of Agriculture put the plant's commodity beef on hold on 30 January and on 17 February announced a recall of more than 143 million pounds (65,000 tonnes) of beef produced over two years. About 50 million pounds had gone to more than 7,000 school districts in 46 states and the District of Columbia, about 30 million pounds had been eaten before the recall, and no illnesses were reported. The Government Accountability Office (GAO) found that states learned only on 26 February, almost four weeks after the hold, which processed products, such as sloppy joe mix and frozen beef patties, contained the beef.
LessonA recall of a raw ingredient has to name every product made from it, or kitchens keep serving those products.
Credit
GAO, in its August 2009 report
2011
Sprouts traced through a distributor
In May 2011 Germany had a large outbreak of the hemolytic-uremic syndrome, a form of kidney failure, caused by E. coli O104:H4. In a case-control study 88% of patients recalled eating cucumbers and 25% sprouts, but a study of 10 groups of diners at one restaurant, built from its recipes, found that sprouts explained all of the cases there. Tracing the sprouts back from that restaurant's distributor led to one producer, which could explain all 41 case clusters with known trading links. The tracing team built a database and network maps of suppliers, distributors and producers during the outbreak, which led to a farm in Lower Saxony and a lot of fenugreek seeds imported from Egypt; Germany's Federal Institute for Risk Assessment (BfR) later released the tools as FoodChain-Lab, open-source tracing software, and used them again in a European hepatitis A outbreak in 2013 and 2014.
LessonTracing tools built during one outbreak can be kept and reused in the next.
Credit
Udo Buchholz and colleagues, Robert Koch Institute; Armin A. Weiser and colleagues, BfR
2012
One frozen lot, 11,000 ill
From 20 September to 5 October 2012 nearly 11,000 people fell ill with norovirus in 390 institutions in five German states, mostly schools and childcare centers, in what the Robert Koch Institute called the largest recorded food-borne outbreak in Germany. Almost all of the institutions were supplied by one large catering company, whose regional kitchens had made dishes such as compote from frozen strawberries, and every affected institution had received strawberries from one lot imported frozen from China. The lot had been delivered to commercial kitchens and did not reach shops. The investigators identified the food within a week, which led to a recall that kept more than half of the lot from being eaten. Germany's Federal Institute for Risk Assessment (BfR) then advised that heating large amounts of frozen berries evenly to over 90 °C appears to inactivate the virus, while stirring them into boiling water or bringing a large batch briefly to the boil may not.
LessonFood sold only to kitchens can still reach thousands of people through one supplier, and a fast trace to the lot is what limits how many eat it.
Credit
Helen Bernard, Mirko Faber, Hendrik Wilking and colleagues, Robert Koch Institute; BfR
Watch for
Lot codes calculated, not scanned
Under the FDA's traceability rule, distributors must pass the lot code of each listed food to the restaurant that receives it, by website, electronic data exchange or invoice. The International Fresh Produce Association (IFPA) said in December 2025 that capturing lot codes in distribution centers is one of the biggest barriers, that extra scanning steps would cost 8 million hours and $250 million a year (its estimate), and proposed instead that warehouse software calculate which lots went to each restaurant from receiving records and pallet moves, with pilots the FDA would be invited to observe. iFoodDS, a software firm, described a product that does this, Lot Code Assist, in March 2026, and wrote that a rigorous evaluation against tracked cases is still needed to measure its accuracy.
Achieved whena test in working foodservice distribution centers, with every outbound case also scanned as a check, publishes how often the calculated lots match the real ones, and is run or reviewed by someone outside the vendors.
Claimed by
IFPA, December 2025; iFoodDS, March 2026
Open
01
Case labels through the freezer
A public bench test of whether the GS1-128 barcodes that carry lot codes on foodservice cases still scan after the trip a frozen case makes, from freezer to cold truck to warm, humid dock to walk-in freezer, again and again. Twenty combinations of label stock, adhesive and printing method, 50 labels each, go on corrugated cases filled with water. The cases cycle between a −18 °C chest freezer and a room at about 20 °C and 60% humidity, and every label is scanned after every cycle with handheld scanners and a phone.
The briefCredit · unclaimed02
Allergen data, catalog against case
A field check of whether the allergen information in foodservice distributors' online catalogs, the pages kitchens order from, matches the allergen statement printed on the case or pack that arrives. Start in the UK, where food law names 14 allergens that must be declared: 300 products from at least five distributors' catalogs, each case photographed at delivery or at a cash-and-carry, each coded for the 14 allergens and any "may contain" statement as given on the pack and online, with every mismatch classed as missing online, extra online or different.
The briefCredit · unclaimed
Open 01 · the brief
Case labels through the freezer
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public bench test of whether the GS1-128 barcodes that carry lot codes on foodservice cases still scan after the trip a frozen case makes, from freezer to cold truck to warm, humid dock to walk-in freezer, again and again. Twenty combinations of label stock, adhesive and printing method, 50 labels each, go on corrugated cases filled with water. The cases cycle between a −18 °C chest freezer and a room at about 20 °C and 60% humidity, and every label is scanned after every cycle with handheld scanners and a phone.
Why it is still open
Tracing depends on reading these codes. The IFDA asks suppliers to print lot-code barcodes on two adjacent sides of every package, and iFoodDS lists unlabeled, mislabeled and partly labeled cases among the problems its method cannot yet handle. GS1's own robotic scanning tests, published in July 2022, measured 2D codes at retail checkouts, not codes on frosted or wet cases. Searches in November 2026 found label makers' advice to test labels in cold, wet conditions, but no published read rates.
Built with what exists
A chest freezer and a room with a humidifier; thermal-transfer and direct-thermal label printers; freezer-grade and standard label stocks; two or three handheld 1D and 2D scanners and a phone running the open-source ZXing-C++ decoder; a barcode verifier, borrowed or hired, for ISO/IEC 15416 print grades before and after.
First step
Print 100 GS1-128 lot labels on two stocks, apply them to frozen water-filled cases and record read rates after three cycles.
Done looks like
First-pass read rates for each combination after 1, 5 and 10 freezer-to-dock cycles, the share of labels that lifted or smeared, and print grades before and after, published with photos as a ranked table a distributor could use to set label specifications for its suppliers.
Open 02 · the brief
Allergen data, catalog against case
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A field check of whether the allergen information in foodservice distributors' online catalogs, the pages kitchens order from, matches the allergen statement printed on the case or pack that arrives. Start in the UK, where food law names 14 allergens that must be declared: 300 products from at least five distributors' catalogs, each case photographed at delivery or at a cash-and-carry, each coded for the 14 allergens and any "may contain" statement as given on the pack and online, with every mismatch classed as missing online, extra online or different.
Why it is still open
The UK Food Standards Agency tells caterers to know what is in the food they serve by recording allergen ingredients on product specification sheets and keeping them up to date, so kitchens depend on what suppliers give them. The nearest check, a study funded by Campden BRI and published in Food Control in 2024, bought 100 products sold on UK retailers' websites and found 15 whose online allergen information did not match the pack; it covered shoppers' grocery sites, not distributors' trade catalogs. A web search in November 2026 found no such check of foodservice distributors' catalogs.
Built with what exists
Phone photos of case labels; an AI model that reads images, such as Claude or Gemini, to transcribe each ingredient and allergen statement, with every allergen code checked by a person; catalog pages saved with a headless browser such as Playwright; a spreadsheet; Python for the tallies.
First step
Photograph 30 cases from one delivery or one cash-and-carry visit and compare each with the same item's page in the distributor's online catalog.
Done looks like
300 products coded by two people who agree on at least 95% of the allergen codes, with the mismatch rate published by distributor and by kind of mismatch, and the photos and saved pages released.