When Experts Disagree, Where Do You Look First?

One carbohydrate food can earn five expert verdicts. Here is the tool we built to make sense of them — and to find the disagreements worth resolving.

Expert food-rating systems disagree — whole-wheat bread scores anywhere from the 12th to the 99th percentile depending on whom you ask. The meta-NPS harmonizes them into a single healthfulness–impact map. Watch it build itself, then explore.

Ask five expert panels whether whole-wheat bread is healthy and you can get five different answers — from the bottom tenth of the food supply to the very top. That is not a measurement error. It is what happens when serious people operationalize the word healthy five different ways.

So in the IAFNS Carbohydrate Quality series I built a meta–nutritional profiling system — a tool that doesn’t pick a winner among the rating systems, but harmonizes them: their middle becomes a starting score (the meta-score), the foods the experts can’t agree on are penalized for it, and every food lands on one map.

The chart below teaches itself. Let it run — it starts with a single loaf and its five verdicts, shows how they become one penalized score, orients each axis with examples, then fills in the map and its regions. Then it hands the map to you.

What you just watched

Each food’s vertical position is its Food Recommendation Index — the meta-system’s single read on healthfulness, which rewards a high median score and agreement between systems. A food with a respectable median can still be pulled down when the experts split on it. Whole-wheat bread is exactly that: its median lands in the low 70s, but its systems range from 12 to 99, so it settles at 53, below where the median alone would put it.

Two foods can share that middle band for opposite reasons — neutral (the experts agree it is middling) or uncertain (they don’t). Solid dots are the agreers; hollow dots are the split. That distinction is the whole point: it separates “we know this food is so-so” from “we don’t actually know.”

The horizontal position is impact — how much of it America actually eats — and this is the lever. A food the experts can’t agree on barely matters if no one eats it; a contested staple is exactly where harmonizing the science would move real diets. So the foods worth a panel’s time sit to the right and hollow: widely eaten and genuinely uncertain. That is where you look first.

One caveat: the quiet, low-impact corner is not worthless. Foods few people eat are often where the method gets tested and the theory sharpened — they teach us how the rating systems behave, even when fixing a single score wouldn’t move the needle on public health. Impact tells you where to act; the whole map is where you learn.

Every dot now carries its real meta-NPS values: the five expert scores, the meta-score, and the FRI, computed across the US food supply (about 6,500 foods in the FNDDS database). The left–right axis is real consumption too. Of the roughly 3,200 carb-rich foods Americans actually eat, the 122 most commonly consumed account for about half of all US carbohydrate intake, and the webinar tool maps those 122. This interactive plots a recognizable subset of them so the idea stays legible; the full tool and dataset sit behind the meta-NPS paper.

This is the first of the carbohydrate-quality tools. The next webinar uses this same map to ask where the disagreements actually cluster — and the answer turns out to be “more places than you’d hope.” See all of them on the resources page.