A Pasta Recall and a Sprout Outbreak Both Named Exact Products This Week. Mr Bite Already Has Their Barcodes and Has Never Checked One.

Two unrelated recalls landed this week, and both of them are built entirely around one piece of information: the exact product someone bought. On October 2nd, Gias Foods expanded its recall of bettergoods Authentic Italian Lemon Alfredo Fettuccine, sold frozen at Walmart, to cover every lot code of the 22-ounce pouch after state lab testing in Florida and Washington turned up possible Listeria monocytogenes, widening a recall the company first issued September 15th for just two lots. The same week, the CDC’s running count of a multistate outbreak tied to Everything Sprouts and its Calco-brand alfalfa sprouts climbed to 76 confirmed cases across 16 states, six of them hospitalized, driven by three different E. coli strains plus Salmonella Agona. Neither notice is a vague “processed food, be careful” warning. Each one names a specific product, a specific lot or date range, a specific UPC. That specificity is the entire mechanism: a recall only protects you if you can match it against what you actually bought.

Mr Bite logs what people actually eat, so I went looking for whether we could ever make that match. The obvious place to start is the barcode scanner, since a barcode is the UPC a recall notice is written around. I traced what happens after a scan, and the identity doesn’t get lost the way sodium or fiber did in earlier issues of this newsletter. When you scan a product and add it to a plan, PlanBarcodeProductModal.tsx inserts a row into user_products with the real barcode column populated from the scan, then writes linkedProductId: savedProduct.id onto the logged meal itself. AddProductScreen.tsx does the same thing on the direct add-a-product path. The logged meal in your history doesn’t carry the barcode directly, but it carries a stable pointer to a database row that does. Scan a recalled product today, and three weeks from now the exact UPC is still sitting there, tied to the exact day you ate it. This is, structurally, exactly the data a recall check would need.

So I went looking for the check itself. I grepped the entire codebase, client and Supabase functions both, for “recall.” The only hits are a chat tool named recall_recipes_from_history, which is about an AI model remembering what it suggested you cook last month, and a couple of comments about write-receipt “precision over recall” that have nothing to do with food safety. There is no recall table, no scheduled job that pulls an enforcement feed, no function that takes a UPC and asks whether anyone, anywhere, has flagged it. We built the plumbing that keeps a barcode attached to a logged meal. We never built anything downstream of it that does a single useful thing with that fact once the product turns out to be dangerous.

I want to be fair to what this isn’t. This isn’t a case of a field quietly getting dropped on the way into a database, the shape of gap I’ve written about before. The barcode is intact, joinable, and has been the whole time we’ve had a barcode scanner. The gap here is simpler and in some ways more embarrassing: nobody ever asked the next question. Recall data for exactly this purpose is free and public, which I’ll get to below, and checking a UPC against it is not a hard engineering problem compared to most things already built into this app. It just isn’t anything we’ve ever pointed our own stored data at.

What would it actually look like if we did? Probably not a push notification that panics someone over a recall three tiers removed from what they ate. Probably something closer to what Abbott and Berry Street are promising with glucose data: a quiet, specific surfacing, “this product you logged on September 20th was recalled on October 2nd,” shown once, in context, backed by a real match on a real identifier we already have sitting in our own database. We’re not there. But for once, the missing piece isn’t the data. It’s that we never asked the question.

Three Quick Bites

1. The CDC’s October 1st update on the Everything Sprouts outbreak puts the case count at 76 across 16 states, with six hospitalizations and illness onset dates stretching from May 31st to September 7th. Three separate E. coli strains, O26:H11, O103:H25 and O168:H8, plus Salmonella Agona, are all turning up in the same Minnesota-grown alfalfa sprouts, sold under the Calco and Everything Sprouts brands to restaurants and grocery stores. One contaminated seed lot producing four distinct outbreak strains is an unusually messy signature for a single source. Food Poisoning Bulletin

2. Gias Foods’ recall of bettergoods Lemon Alfredo Fettuccine, sold nationwide at Walmart in 22-ounce pouches, expanded on October 2nd from two flagged lots to every lot code the company shipped, after routine state testing in Florida and Washington found possible Listeria monocytogenes. No illnesses have been reported yet, which is exactly the window a recall is supposed to close before any show up. FOX 10 Phoenix

3. Novo Nordisk’s OCTANE real-world study, presented September 30th at the EASD meeting in Milan, followed 194 adults who switched from an injectable GLP-1 to the oral Wegovy pill and stayed on it for three months, finding an average weight loss of 4.1 percent and a drop in obesity-range BMI prevalence from 87.1 to 65.5 percent of the group. It’s a real-world switching cohort, not a placebo-controlled trial, but it’s one of the first looks at what actually happens when people move off the needle. Patient Care Online

Tool of the Week: openFDA’s Food Enforcement API

Since this issue is about a check we’ve never built, the pick this week is the free, public data that check would actually run against. The FDA’s openFDA platform exposes its Recall Enterprise System as a plain JSON API at api.fda.gov/food/enforcement.json, covering publicly releasable recall records from 2004 to today, updated weekly, released under CC0 with no license fee and a free key that lifts the daily request quota to 120,000. You can query it by product description, recalling firm, classification or date range, right now, from a browser or a single curl command, no app required. It’s not a UPC-indexed lookup table, so matching it to a barcode still takes some text matching on our end. But the data a recall check would need already exists, for free, maintained by the federal government, and has for years. openFDA Food Enforcement

Thanks for reading. If you want a coach that’s honest about the question it’s never asked, Mr BITE is free on iOS and Android.

Michael, building Mr BITE