One question, asked four times

Step back from the wine and look at what this machine is actually capable of saying.

It has four weights, and it uses them the same way every time. Take each taste, multiply it by its weight, add up the results.

Which means the only question the machine can ask is this one:

How much is this taste worth?

It asks it four times, once per taste, and adds the four answers together. That is the entire range of thoughts available to it.

For the last chapter's wines, that was enough, because there the question had an answer. Apricot really was worth 40 dollars a unit, every time, in every bottle. Plum really was worth 16, always. The price genuinely was the four tastes added up, so a machine that adds up four tastes could find it.

For this collection, the question has no answer.

Ask it: what is plum worth? The honest reply is that it depends. In a cabernet, plum is worth a great deal. In a viognier, it is worth less than nothing. Plum on its own, with no grape attached, has no price.

The machine has no way to say that. It has one slot for plum's worth, and the slot holds one number. There is nowhere to write "it depends."

So it writes down the average and carries on, which is what you watched it do.