The Disagreement Has a Shape
Part 2 of a series on the meta-NPS. Run five expert systems across the whole carbohydrate supply and most foods come back agreed. The contested ones aren’t scattered at random — they pile up in one aisle, and they’re worst exactly where America eats the most.
Part 1 showed one loaf earning five expert verdicts. Part 2 runs that meta-rating over every carb-rich food to ask how often the experts split, and where. The answer has a shape: about a fifth of carb foods are contested, grains most of all, and how much we eat a food decides which of those splits to settle first.
The short version. Part 2 of the Meta-NPS series.
Part 1 showed one loaf earning five expert verdicts. Run the same five systems across every carbohydrate-rich food and the disagreement turns out to have a shape.
- Most foods are not the argument. About a fifth of carb foods come back contested. The rest, the experts broadly agree on.
- The contested ones pile up in one aisle. Grains are worst at roughly 50% uncertain, against about 33% for mixed dishes and 27% for beverages. Every food group has its own fingerprint.
- Consumption decides what to settle first. A contested food nobody eats barely matters. Around 122 foods account for roughly half of US carbohydrate intake — that overlap, contested and widely eaten, is where harmonization would actually change something.
The point. “The experts disagree” is not a shrug. The disagreement is structured, concentrated, and rankable — which makes it a work list rather than an excuse.
Part 3 brings in the public and finds a second gap.
In Part 1, one loaf of whole-wheat bread earned five different expert verdicts, from the 12th percentile to the 99th. The meta-NPS turned that spread into two numbers a person can actually read: a meta-score for how healthy the experts call it on balance, and a consensus measure for how much they agree.
One food is an anecdote. The obvious next question is the one the second webinar set out to answer: run the same five systems across the entire carbohydrate supply — roughly 6,500 foods — and how often do they split like that? And when they do, is it random, or does the disagreement have a shape?
Four foods, four verdicts
Start with the two numbers together. Meta-score puts a food high or low. Consensus says whether the experts meant it. Cross them and you get four honest categories, not one tidy score: certainly healthy, certainly unhealthy, certainly neutral (they agree it’s middling), and the category the meta-NPS exists to surface — uncertain (they don’t agree at all).
Those four categories aren’t abstract. Watch four foods sort into them.
Two of those four are easy. Apple sits near the top, the systems bunched tight: certainly healthy. Croissant sits at the bottom, just as tightly agreed: certainly unhealthy. Potato chips is the third, and it lands in the middle with the systems still bunched, so they agree it’s a so-so food. That’s a real verdict. The fourth only looks like it. White rice lands in the middle too, but the systems fan from the bottom quarter to the top: some treat it almost like a vegetable, others almost like a sweet. “Middle” was hiding two opposite things. One is a verdict. The other is a shrug.
Most foods aren’t the argument
Here’s the part that gets lost when people say experts “can’t agree on anything.” Across all carb-rich foods, they mostly do. Only about a fifth come back uncertain. The disagreement is the exception, not the rule.
But that fifth isn’t spread evenly. Sort the supply into food groups and each one has its own fingerprint.
Fruits and vegetables show the cleanest fingerprint: almost entirely agreed, and agreed healthy, with no certainly-unhealthy foods at all. Desserts flip the other way, mostly unhealthy, though even there a commonly eaten food like honey hides in the uncertain slice. And grains are the contested aisle, with roughly half of them uncertain, more than any other group, followed by mixed dishes and beverages.
Each group’s slice mix is its full composition across the supply; the individual foods named in the walkthrough are a recognizable sample, drawn in proportion to that mix from the 122 most-commonly-eaten carbohydrate foods. The cells illustrate the fingerprint — they are not the whole list.
So the disagreement has a shape after all. It concentrates, and grains are the epicenter. The bread from Part 1 wasn’t a fluke. It was a postcard from the most uncertain aisle in the store.
Which disagreements are worth resolving
A contested food the experts can’t settle barely matters if no one eats it. The lever, as in Part 1, is consumption. The foods worth a panel’s time are the ones that are uncertain and widely eaten, because that’s where harmonizing the science would actually move diets.
This is the tool doing exactly that. The map plots every food by healthfulness (up the screen) and by how much America eats it (across). The 122 foods shown account for about half of all US carbohydrate intake. The walkthrough does what we did in the webinar: filter to the contested middle, shift right to the most-eaten, then order by disagreement until four foods are left.
Filter to the foods the experts call uncertain. Shift right, toward the ones America actually eats. Then order what’s left by how far apart the systems land. A hundred and twenty-two foods become seventeen, and seventeen become four: wheat bread, sweetened iced tea, white rice, and white bread. The most contested, most-eaten carbohydrate foods in the American diet.
Four foods, no easy answer
Pull those four back onto the boxplot and the problem is stark. Every one of them has a rank range above 69 — the systems disagree so widely that each food is essentially healthy or unhealthy depending on who you ask. Not a rounding error between the 60th and 70th percentile. The full reach of the scale, four times over, on foods tens of millions of people eat every day.
So we went looking for what drives these splits — some factor that could explain the disagreement across all four. The likeliest suspect is processing: the two systems that score it sink both breads. But it isn’t a clean driver. The same systems rank white rice and iced tea near the top, so processing cuts opposite ways across the foods instead of explaining them. One loaf did resolve that way back in Part 1: setting aside the processing systems collapsed the disagreement for whole-wheat bread, a different loaf from the wheat bread here, and settled its verdict to healthy. That is a finding about where the experts are arguing, which is the first thing you would need before anyone could resolve it.
Where this goes
That’s the contribution the meta-NPS is reaching for. Not a better single score, but an honest map of the gray area: how much of the supply is contested, where it clusters, which contested foods are worth resolving first, and a way to test what drives each disagreement. Surfacing the gap is the first step toward closing it.
Two honest limits, the same ones the webinar ended on. Harmonizing this universal disagreement is the floor, not the ceiling. Every system here, and the meta-NPS on top of them, rates one food the same way for everyone, and “healthy for whom” is a question none of them ask. The floor is harmonization; the ceiling is personalization, and both are buildable now. And defining healthfulness is as much art as science — there is no proof of the single best system, so the goal isn’t to crown one. It’s to show, food by food, how settled the science actually is.
There’s one more thing the experts’ map can’t tell you: whether anyone outside the panels sees food the same way. Part 3 brings in the public, and what happens when expert verdicts and everyday belief pull apart.
The categories, the food-group shares (about 20% of carb foods contested, near 50% of grains), the 122 foods that make up roughly half of US carbohydrate intake, and the four most contested staples are all from the public IAFNS Insights From the Tool webinar. The numbers are a conservative floor: only five of hundreds of published rating systems were sampled, and a single consensus cutoff drew the lines — a different defensible cutoff flags more, not fewer. The webinars and papers behind the series are on the resources page.
Comments