Foodpairing, the Belgian company that turned aroma chemistry into a subscription product, offers this on its explainer page. "Cherry and asparagus, for example, are a perfect pairing because they both share similar floral and green aromas."

Nobody has ever eaten that. Not because the chemistry is wrong. Because the chemistry was never the point.

The claim under the example is the food pairing hypothesis, and it is stated cleanly. "Two (or more) ingredients are complementary when they share key aroma components." It is an appealing idea with a genuinely interesting research literature behind it, and it has produced one of the best-known dishes of the last thirty years. What it has never produced is a room full of people who were handed the pairings and agreed.

Three of the page's load-bearing claims are worth pulling apart, because the way each one goes wrong is different, and the third one is the interesting failure.

Every study above counts recipes. Recipes record what people cooked, which is a product of trade routes, refrigeration, religion, price, and what the neighbors did. Co-occurrence is not liking.

Wender Bredie and Ditte Hartvig at Copenhagen did the obvious thing and asked people. As Bredie presented it at a 2011 molecular gastronomy seminar, and as Martin Lersch reported it at the time, they built 53 binary mixtures from 19 food odors and put them in front of a sensory panel, rating intensity, pleasantness, complexity, harmony and novelty. If the hypothesis were right, pleasantness should climb with volatile overlap.

It did not. There was no correlation between the hedonic score of a pairing and the overlap of its volatiles. In 91% of the mixtures the perceived intensity landed between that of the two components rather than above both, which is ordinary mixture suppression and had been known for decades. Pleasantness of the individual components generally went down when they were combined.

One panel, one set of odorants, one lab, reported through a seminar rather than a paper. That is not fatal on its own. But it is the test the hypothesis most needed to pass, it has been sitting there since 2011, and nobody appears to have run a better one. The marketing has not slowed down.

The deepest problem is not statistical. It is that the method counts the wrong molecules.

Gas chromatography coupled to mass spectrometry will find hundreds of volatiles in almost anything. Roasted coffee yields well over eight hundred. Almost none of them matter. A compound only contributes to what you smell if it is present above its own odor threshold, and thresholds vary across something like twelve orders of magnitude between compounds. Sharing a molecule that sits below threshold in both ingredients is sharing nothing.

Peter Schieberle, Thomas Hofmann and colleagues at Munich spent decades doing this properly, identifying the odorants that actually carry a food's smell and then proving it by rebuilding the aroma from pure compounds. Their 2014 review in Angewandte Chemie put a number on the whole enterprise.

Two percent of the known volatiles do the work. Everything that smells of anything is built from a vocabulary of about two hundred and thirty words, which means that any two foods will share compounds, and that the overlap is mostly background chemistry rather than signal. Counting it is counting noise with high precision.

There is a related problem with where the compound lists come from. Ahn and colleagues built their network out of Fenaroli's Handbook of Flavor Ingredients, a flavor industry reference, which records that a compound has been detected in an ingredient. It does not record how much, and a presence-absence table cannot tell a character-defining molecule from a trace artifact. Comparing the third and fifth editions, the ingredient count runs 916 to 1,507 and the compound count 629 to 1,107, which means a large part of what the network measures is how much analytical attention a food has received.

Foodpairing's own sentence, to be fair, says "key aroma components." That is the right idea. The trouble is that a chromatogram does not tell you which peaks are the key ones. A person has to do that, food by food, and it takes years.

The origin story is better than the theory. In the early 1990s François Benzi, a flavor chemist at Firmenich in Geneva, noticed that white chocolate and caviar both carry amine notes. Heston Blumenthal put the pairing on the menu at the Fat Duck, it worked, and it became the founding anecdote of an entire movement. Blumenthal and Benzi went on to jasmine and pork liver, which share indole. Martin Lersch later showed that strawberry and coriander both carry (Z)-3-hexenal, which is why that combination keeps turning up on tasting menus.

These are good dishes. What is not established is that the chemistry explains them. White chocolate and caviar is also fat against salt, sweet against umami, smooth against granular, and warm against cold, every one of which is a principle cooks had before anyone owned a mass spectrometer. Harry Klee, the tomato geneticist at Florida who has spent his career on which volatiles people can actually taste, put it to Chemical & Engineering News without much diplomacy. The whole flavor-pairing business, he said, is a gimmick by a chef practicing biology without a license.

Blumenthal, it should be said, never claimed the chromatogram was doing the cooking. He tasted the dish. The theory got built on top of him afterward.

Quite a lot, once the universal claim is dropped.

Aroma really does carry most of flavor, and treating a pairing as a smell problem rather than a taste problem is a genuine improvement over intuition. The 2011 network paper found a real cultural signature, and the fact that it points in opposite directions in Milan and in Maharashtra is more interesting than a universal law would have been. Shared-compound analysis is a plausible way to generate candidates, which is a real use. It is a hypothesis generator that got sold as a physical constant.

Foodpairing is still trading. It now calls itself a flavor intelligence company and sells AI agents to marketing and innovation teams at consumer packaged goods firms. Which is roughly where a theory ends up when it never has to pass a taste test.