A Nature Metabolism Study Matched Two Meals on Every Nutrient and Still Got Different Insulin Spikes. Mr Bite's Barcode Scanner Never Asks Open Food Facts the One Field That Would Tell It Why.

Virginia Tech’s Fralin Biomedical Research Institute published a trial in Nature Metabolism on October 5th that was built to settle one specific argument: when two meals have the same nutrients on paper, does processing still change what happens in your body? Lead author Zach Hutelin and his team, led by Alexandra DiFeliceantonio, built two meals matched within 1 percent of each other on carbs, fat, protein, calories, water and salt. One was ultraprocessed: a PB&J sandwich, deli turkey, veggie chips, a cookie, cereal, instant mashed potatoes. The other was minimally processed: banana, dried cranberries, cheese, egg. Same macros, same calories, same everything a nutrition label would show you. Thirty-two healthy adults ate both, on separate days, while researchers drew blood for three hours afterward and scanned their brains looking at food images.

Every metabolic measure came back different. The ultraprocessed meal produced a bigger insulin surge and kept blood sugar elevated longer, even though it delivered the identical carbohydrate load. The brain scans showed differential activity in the ventral striatum and nucleus accumbens, the reward circuitry, while people evaluated food cues after each meal. DiFeliceantonio is careful about what this does and doesn’t prove: nobody yet knows whether these short-term shifts are what drives the long-run health outcomes tied to ultraprocessed diets. But the headline finding stands on its own: a gram of carbohydrate is not a gram of carbohydrate once you ask the body to actually handle it. Processing is doing something that a macro total can’t see.

That’s an uncomfortable sentence for an app whose entire job is totaling macros, so I went and checked what Mr Bite actually knows about how processed a food is. The honest answer is nothing, and not because we looked and decided it didn’t matter. We never built a way to ask.

Barcode scans go through openFoodFactsService.ts, which calls Open Food Facts’ API and requests a specific, named list of fields back. I read both lists end to end: OFF_SEARCH_V2_FIELDS and OFF_SEARCH_CGI_FIELDS, defined a few lines apart. Each one spells out exactly what we want: product name in six languages, brand, three image URLs, serving size, product quantity, and nutriments, the macro and calorie block. Open Food Facts has carried a nova_group field for years, a 1-to-4 score for how processed a product is, right alongside a nutriscore_grade. Neither name appears anywhere in our codebase. I grepped for both, and for “nova” generally, and the only hit was a variable called noValidIdx in an unrelated test, matched on the substring. There’s no feature flag turning this off, no code that fetches it and drops it before it reaches the UI. The field is sitting on every single product record Open Food Facts already returns to us, and our request simply never asks for it, the same shape of gap this newsletter keeps finding: not a deleted feature, a feature nobody ever reached for because the rest of the pipeline was built around macros and nothing else.

I’m not going to pretend adding nova_group to a field-list string turns Mr Bite into a tool that explains insulin curves. It wouldn’t. A NOVA score is a classification, not a metabolic simulator, and this week’s study is exactly the kind of result that should make anyone cautious about overclaiming what one number can tell a user about their own body. But there’s a real distance between “we can’t model this” and “we don’t even surface the one signal that’s already sitting in the response body.” Right now a user can scan the exact PB&J-and-veggie-chips combination from this study and the cookie from the control arm, log identical calorie and macro totals for both, and get zero indication from Mr Bite that one of those two things is a different category of food than the other. The study’s whole point was that identical macros can hide a real physiological difference. Our own data model is built to make sure nobody ever sees that hidden part.

BITE of the Week

Bite bumped to version 1.17.0 on October 13th and got EAS-assigned build numbers (iOS 120, Android 150) the next day, but the store baseline is still 1.16.0, live since September 28th, so none of this is in a user’s hands yet. The release itself is a straight JS delta: nothing under ios/, android/ or our dependencies changed, so it’s OTA-eligible once it ships. The one piece of actual engineering in the window is a fix to scripts/supabase-identity-health.js, the script that audits whether every production profile can still write data. Since September’s cutover to dual auth, a profile can authenticate through either a Clerk ID or a native Supabase auth_user_id, but the health check had only ever been taught to check the first one. That meant every native-auth user with no Clerk link was getting counted as write-dead, a false alarm on every single release. The fix adds auth_user_id to the query and changes the write-dead condition to “neither id is present,” and it now throws loudly if anything tries to feed it the old, retired key instead of quietly reading it as zero. Production currently reads 317 profiles, 58 without a Clerk ID, 0 without both ids, 0 critical. It’s not a feature a user will ever see, but it’s the difference between a health check that tells you the truth and one that’s been crying wolf since the auth cutover.

Three Quick Bites

1. Leader Formulas recalled one lot of its RAPTURE Preworkout on October 6th after FDA testing found it contains DMAA (1,3-dimethylamylamine), a stimulant the agency links to serious cardiovascular risk including heart attack and difficulty breathing. The recalled lot, 260327-4, covers all three flavors (Rainbow Twist, Electric Lemonade, Berry Mojito) and was sold nationwide online and in retail between January 2025 and May 2026. No adverse events had been reported as of the announcement, but DMAA has no business being in a supplement at all; it’s banned in the US as a dietary ingredient and keeps turning up in pre-workout products anyway. FDA

2. Gias Foods recalled two lots of its bettergoods Authentic Italian Lemon Alfredo Fettuccine, sold nationwide at Walmart, over possible Listeria monocytogenes contamination found through routine state sampling rather than any reported illness. The affected 22 oz bags carry lot numbers L6079C or L6080C with expiration dates of 09/19/2027 or 09/20/2027 printed on the back. It’s a reminder that a recall can start from a lab swab months before anyone gets sick, which is the system working, not failing. FDA

3. A small UCSF pilot published September 23rd in JAMA Psychiatry found that 20 people with depression who cut their ultraprocessed food intake by more than 85 percent over four weeks saw their depression scores improve, with a smaller effect on anxiety. The lead author’s working theory points at gut inflammation rather than any single nutrient, and she’s upfront that a 20-person trial can’t establish cause and effect; UCSF is already planning a larger follow-up for 2028. Paired with this week’s main story, it’s two independent labs converging on the same idea from different directions: processing itself, not just the nutrient panel, seems to be doing real work on the body. UCSF

Tool of the Week: Open Food Facts

Since this issue is about a field we could pull but don’t, the pick this week is the place you can pull it yourself. Open Food Facts is the free, open, crowd-maintained product database Mr Bite’s own barcode scanner already talks to, and its website and app will show you a product’s NOVA group and Nutri-Score the moment you scan it, no account needed. It’s the exact data our pipeline leaves on the table today. Scan the same granola bar in both places and you’ll see everything we show you, plus the one number we don’t. Open Food Facts

Thanks for reading. If you want a coach that’s honest about the field it isn’t asking for yet, Mr BITE is free on iOS and Android.

Michael, building Mr BITE