The price is in the pairing

Try to find the price anywhere in a single taste, and you will not.

Line up the four bottles by whether they are cabernet:

• cabernet, plummy, 40 dollars
• cabernet, apricotty, 10 dollars

Knowing it is a cabernet tells you nothing about the price. It could be either.

Line them up by whether they are plummy:

• cabernet, plummy, 40 dollars
• viognier, plummy, 10 dollars

Knowing it is plummy tells you nothing either.

Do it for viognier and for apricotty and the story is identical. Every single taste, on its own, is exactly uninformative. Each one appears on an expensive bottle as often as a cheap one.

This is why the weights cancelled. A weight is a statement about what one taste is worth on its own, and in this collection no such statement is true.

But now put two together, and the price appears immediately:

• cabernet and plummy, expensive, every time
• viognier and apricotty, expensive, every time
• the other two combinations, cheap, every time

There is nothing subtle in the data. The rule is completely reliable. It simply lives in the pairs rather than in the parts.

A machine that only ever looks at tastes one at a time is looking in the one place the answer is not.