Nine Sick People Named the Same Cheese From Memory. Mr Bite's Own Quick-Log Screen Keeps That Exact Amount of Detail.
A raw milk cheese outbreak that has sickened 13 people, most of them small children, was only traced because investigators asked sick patients to recall what they'd eaten by name. Checking Mr Bite's own logging paths found that barcode scans and packaged-food search both capture a real brand and barcode, but the generic quick-log screen behind most small-producer foods like this one keeps only a typed name, no more durable than what the outbreak investigators had to work with.
Sierra Nevada Cheese Company issued a voluntary recall on September 24th of every variety of its Graziers grass-fed raw milk cheese, medium cheddar, sharp cheddar, jalapeño jack, and Monterey jack, after the FDA linked the cheese to a multistate E. coli outbreak. As of the most recent count, 13 people across nine states have gotten sick, 8 were hospitalized, and 3 developed hemolytic uremic syndrome, a kidney complication serious enough that more than half the patients are children five or younger. The company has no lot or date code to narrow the recall to, so it’s pulling everything currently on the market. What struck me reading the coverage wasn’t the case count. It’s how the outbreak got solved: state and local health officials called nine of the sick patients and asked them what they’d eaten in the week before symptoms started. All nine, independently, named the same cheese. That’s the entire mechanism that closed the case, nine separate human memories converging on one brand name, because there was no faster, more structured record anyone could pull instead.
Bite logs what people eat for a living, so I wanted to know what our own version of that record would have looked like for someone who’d actually logged this cheese. I went in expecting to find the same kind of gap I found two issues ago with barcode scanning, data captured and never checked against anything. What I found instead was more specific, and in a strange way reassuring and worse at the same time.
Packaged-food search turns out to be a real identity match, not a guess. SearchFoodScreen.tsx doesn’t query a generic nutrition lookup when you type a food name, it queries Open Food Facts and Bite’s own crowdsourced community_products table in parallel, and both return a real barcode and brand on every hit. Pick a result, and it flows into the same PlanBarcodeProductModal.tsx that handles an actual camera scan, which writes that brand and barcode onto a user_products row and links the logged meal to it with linkedProductId. A text search for a national brand is, structurally, exactly as identifiable as scanning its box. I’ll admit I was bracing to write about the barcode pipeline again. I didn’t get to.
Graziers raw milk cheese is not that kind of product. It’s a small creamery’s cheese, sold through farm stores, co-ops, its own website, and specialty grocers, which is close to the textbook case a crowdsourced global barcode database hasn’t indexed and a community table only grows once someone’s already logged it there first. So I traced what actually happens when a search for “raw milk cheddar” comes back empty in both places. The user lands on AddItemScreen.tsx, Bite’s generic quick-log screen, the same one behind manual entries, AI-photo estimates, and voice logging. I read the whole function. The object it builds and saves has a name, portions, a serving quantity and unit, calories, protein, carbs, fats, an ingredients list, an image, a day, and a meal slot. There’s no barcode field. No brand field. No sourceType, no linkedProductId, nothing that points back to a specific producer at all. Whatever string the user typed, “raw milk cheddar,” “farm cheese,” whatever they called it, is the only trace of identity that survives into the log.
That’s the part worth sitting with. Someone who logged this exact cheese in Bite through the quick-log screen, which is the only screen available to them since the product isn’t in either searchable database, would have exactly what the outbreak investigators were asking sick people to produce over the phone three weeks later: a name, recalled, with nothing structured behind it. Not a worse record than human memory. The same record as human memory. I even found a smaller tell of how blurred this is internally: the analytics call that logs how a meal was added fires method: 'barcode' for a text-search pick too, the same label as an actual camera scan, because both paths share that one modal. The system that’s supposed to tell us which entries are identity-backed can’t even reliably tell itself.
I don’t think the fix here is “add a brand field to quick-log,” because a user who can’t find their cheese in a database usually can’t type its lot code either. What this actually shows is where the barcode-matching pipeline I wrote about last time would and wouldn’t help, even if we built it today. It would catch a national brand instantly. It would do nothing for a small creamery’s raw milk cheese, which is exactly the kind of product behind this outbreak and, structurally, the kind every food-safety investigator still has to trace one phone call at a time.
Three Quick Bites
-
The FDA’s CORE outbreak table, updated October 2nd, lists 17 active multistate investigations, with a Salmonella Bovismorbificans outbreak tied to Evergreen Fresh and Calco broccoli sprouts now at 32 confirmed cases, most of them in Washington state, and the contaminated sprouts already recalled. The same update flags a brand-new, unsolved Listeria outbreak with at least 11 cases and traceback just getting underway, a reminder that this list rarely stays still for long. Food Poisoning Bulletin
-
Wakefern Food Corp. recalled its Wholesome Pantry Organic Unsweetened Apple Sauce, in 23-ounce jars with a best-by date of June 16, 2027, after routine FDA testing found patulin, a mold toxin, above the safety limit. The jars shipped through ShopRite, Price Rite Marketplace, The Fresh Grocer, Gourmet Garage, and Fairway Market, and no illnesses have been reported. It’s a useful reminder that “organic” and “contaminant-free” are answering two completely different questions. Food Safety News
-
Ohio’s USDA-approved waiver barring SNAP benefits from being used to buy sugar-sweetened sodas and similar drinks took effect October 1st, making it the fourth state this year to get such a waiver approved and the 22nd overall. Compliance falls on retailers, not shoppers, with a six-month grace period built in. It’s a blunt policy instrument next to anything an app can do, but it’s targeting the exact beverage category most nutrition software, Bite included, still just logs as calories. Local 12
Tool of the Week: USDA FoodData Central
Since this issue is about what a search comes back empty on, this week’s pick is the other big free branded-food database Bite’s own search isn’t currently checking. The USDA’s FoodData Central API indexes Foundation Foods, SR Legacy, and the Global Branded Foods Database, all free, all queryable from a plain REST endpoint once you grab a no-cost key from api.data.gov. It’s government-maintained rather than crowdsourced, which makes it a genuinely different source than Open Food Facts, not just a mirror of it. To be honest about its limits here: a small raw-milk creamery selling through farm stores is exactly the kind of product that’s unlikely to be in this one either. But for the much larger set of regional and mid-size brands that never make it into a crowdsourced app database, this is free, well-documented, and sitting there unused. USDA FoodData Central
Thanks for reading. If you want a coach that’s honest about what a search comes back empty on, Mr BITE is free on iOS and Android.
Collins, building Mr BITE