The Rules of Thumb Baked Into Every Food Score Just Had a Bad Week

Three unrelated research teams published three unrelated trials within about a week of each other in early August, and by accident they ran the same experiment: take a diet rule everyone treats as settled, and actually test it.

The one that caught my eye first came out of the University of Toronto. Harvey Anderson’s team took 74 adults with overweight or obesity, put them on one of three 12-week diets, a low-dairy calorie-restricted diet, or one of two higher-dairy diets built around three daily servings of full-fat milk, cheese and yogurt, and measured what happened to their weight, body composition, blood lipids and insulin resistance. The standard advice, the one printed on the side of a thousand nutrition-label explainers and baked into more than one food-scoring algorithm, is that full-fat dairy is the worse choice: more saturated fat, more calories, worse for your heart. Anderson’s group found no meaningful difference in any of it. “Those that had three servings of dairy didn’t have adverse levels of blood cholesterol or lipids or evidence of insulin resistance,” he told his university’s press office, and the higher-dairy groups picked up more calcium, protein and vitamin D along the way, with better blood pressure besides. Published in The Journal of Nutrition, and it’s not the first study to find this, researchers increasingly call it the dairy matrix effect: dairy’s fat doesn’t act like fat lifted out of dairy and put somewhere else, because the food isn’t a bag of nutrients, it’s a matrix those nutrients sit inside.

That would already be a decent essay on its own, and it’s basically the same argument I made about ultra-processed food back in Issue 3: a single nutrient pulled out and scored in isolation doesn’t predict what a whole food actually does to you. What made this week different is that it happened twice more in six days. A six-month Sweet Tooth Trial out of Wageningen University and Bournemouth University, published in the American Journal of Clinical Nutrition, put 180 adults on low-, regular-, or high-sweetness diets and found no meaningful difference in sweet-taste liking, weight, or metabolic markers between the groups. The idea that cutting sweetness retrains your palate and kills your cravings, another thing plenty of coaching apps will tell you as fact, didn’t survive contact with an actual trial. And a Newcastle University trial published in the British Journal of Nutrition found that adding a daily glass of 100% fruit juice or a smoothie to an otherwise-improving diet lowered depression scores over four weeks, with no adverse metabolic effects measured, complicating the reflexive “it’s just sugar water” flag that a lot of tracking apps, guilty as anyone, slap on juice next to soda.

None of these three rules, low-fat over full-fat, cut sweetness to cut cravings, juice equals soda, are things anyone at Nutriflex sat down and decided to believe. That’s exactly the problem. Nobody has to decide to believe a heuristic for it to end up governing a product. It arrives pre-loaded, in a food database’s default sort order, in a “healthy swap” nudge some generic prompt suggested, in a scoring formula borrowed from somewhere because it was there and looked reasonable. A rule of thumb doesn’t announce itself as a rule of thumb. It just quietly sits underneath a feature until a controlled trial, or a user with a nutrition degree, points at it.

So the actual takeaway from this week isn’t “full-fat dairy is fine, sweetness reduction doesn’t work, juice isn’t so bad,” useful as those three facts are on their own. It’s that we owe Mr BITE the same audit Issue 5 asked of the FDA’s ultra-processed definition: not “does this look right,” but “did anyone ever actually test this, and against what.” Three heuristics failed a real trial in the same week purely by coincidence. The uncomfortable question is how many more are sitting in the product that just haven’t had their week yet.

BITE of the Week: The Tap That Froze for Anyone Who’d Never Said Yes

Version status first, and Android’s is worth being blunt about. iOS moved twice this cycle: 1.8.0 (build 101) went live on the App Store around August 12th, and a same-week follow-up, 1.8.1, shipped two JS-only fixes on top of it. Android’s story is worse. 1.6.0, 1.7.0, and 1.8.0 were each cut as Android builds, and none of the three were ever submitted to Play. The 1.8.0 Android artifact got stuck behind a Play upload-key reset still waiting on Google’s approval, so it was superseded rather than shipped. The last Android release actually confirmed live is still 1.5.0. If you’re on Android, that’s what’s on your phone.

The real story of the cycle started with an affiliate’s flat complaint: “the app keeps freezing on me when I try to click on something.” The cause was a deadlock specific to people who’d never granted Mr Bite’s AI-processing consent yet, which is to say, specific to every brand-new install and every affiliate demoing the app for the first time. iOS won’t present a second modal while one is already up, and the consent sheet was sometimes being requested from inside an already-open modal, the voice-log screen, the photo re-estimate sheet. When that happened, the consent sheet never actually appeared, its promise never resolved, and because that promise was cached across the whole app, one blocked request turned every AI feature, photo scan, voice log, re-estimate, chat attachments, into a silent no-op until the app was force-quit. Three fixes at three layers: a 3-second watchdog on the consent sheet that gives up and lets a retry succeed once nothing is blocking it, moving voice-log consent to before its modal opens instead of after, and resetting a stuck photo-analyzing flag in a finally block so one failure can’t wedge the next attempt. The same sweep found a second freeze path, a chat stream with no time limit that could hang forever on a dead connection (now bounded at 120 seconds), and a third where tapping send while the app was still loading context silently threw the tap away instead of queuing it.

Two more fixes worth a line. Recipe imports used to sit on a spinner for 25 to 45 seconds, and a failed import auto-dismissed in about 3 seconds with generic copy, so a failure looked identical to the recipe just never arriving. Imports now confirm in seconds and finish in the background, a failure stays on screen until you dismiss it, and captions detailed enough to already read like a recipe skip the slow transcription step entirely, cutting the wait from as much as 40 seconds to 6 to 9. And food search picks stopped failing outright when Open Food Facts’ main product site was down but its search index was still up: the app now falls back to the product data search already returned instead of re-fetching from the site that’s unreachable.

Three Quick Bites

1. A 12-week trial found full-fat dairy didn’t raise weight, body fat, or cholesterol in adults with obesity. University of Toronto researchers randomized 74 adults to a low-dairy diet or one of two diets built around three daily servings of full-fat milk, cheese and yogurt and found no meaningful difference in weight, body composition, blood lipids or insulin resistance between groups, with the higher-dairy groups gaining more calcium, protein, vitamin D and better blood pressure. It’s the study behind the essay above. University of Toronto

2. A six-month trial found cutting dietary sweetness doesn’t reduce cravings, sweet-taste liking, or improve metabolic health. The Sweet Tooth Trial, run by Wageningen University and Bournemouth University and published in the American Journal of Clinical Nutrition, put 180 adults on low-, regular-, or high-sweetness diets for six months and found no meaningful differences in sweet-taste preference, body weight, or diabetes and heart disease risk markers between groups. ScienceDaily

3. A daily glass of 100% fruit juice or a smoothie was linked to lower depression scores in a four-week trial. Newcastle University researchers gave adults with low fruit and vegetable intake support to eat more produce, with one group also adding a daily glass of juice or a smoothie, and found that group’s depression scores fell significantly more, with no adverse effect on metabolic health markers measured. Published in the British Journal of Nutrition. ScienceDaily

Tool of the Week: ClinicalTrials.gov

Given an essay built entirely on three trials, the pick this week is the place all three, and pretty much every clinical trial run in the U.S. or registered internationally, live in public. ClinicalTrials.gov is the National Library of Medicine’s free registry of hundreds of thousands of studies, and its listings show you the design questions a headline usually skips: how many people, how long, randomized or not, placebo-controlled or not, what was actually measured as the primary outcome. For readers: before you act on a diet claim, search it here and check whether “trial” means 180 people over six months or 12 people over four days. For builders: the free API returns structured trial metadata, phase, enrollment, status, outcome measures, which is exactly the kind of thing a coach should be able to check before it repeats a claim with more confidence than the underlying trial earned. clinicaltrials.gov. Free, public, U.S. government data.

Thanks for reading. If you’d rather your coach’s defaults got the same scrutiny as the studies it cites, Mr BITE is free on iOS and Android.

Collins, building Mr BITE