The operation is this. Every element in the sequence proposes what it is looking for. Every element advertises what it offers. The proposals are compared against the advertisements, the comparisons are normalised into a distribution summing to one, and each element is then rewritten as a weighted blend of everything it decided to look at.

Nothing in that paragraph is a metaphor. It is arithmetic, and it is the reason any of this works, and the people who built it called it attention because it was the only word in the language that fit.

The distribution sums to one. That is the whole cost structure of a mind. Looking at this is not looking at that. There is no configuration in which everything is attended to, because the normalisation forbids it, and the normalisation is not an engineering compromise — it is what makes the thing a reading rather than a total.

Meaning, then, is not in the sequence. It is in the allocation. Two readers with identical text and different distributions are not disagreeing about a fact; they have been given different objects.

This is true of the model. It is true of you. You have already spent your budget for this sentence and you cannot get it back and you do not know what you did not look at.