A button costs 3. An AlertDialog costs 7.
I know this because I built the scoring system. Five dimensions, each 0-2, summed. Visual complexity, interaction complexity, state management, content density, accessibility requirements. A button has one state transition (press), one visual treatment (variant), and one accessibility need (label). Three. An AlertDialog has modal focus trapping, overlay management, destructive action confirmation, keyboard dismissal, screen reader announcements, and the psychological weight of asking a user to make an irreversible decision. Seven.
But the interesting space is not 3 or 7. It is the gap between them.
In that gap live the components nobody thinks about. A Separator costs 0 — it is pure visual structure with no cognitive demand. A Tooltip costs 2 — it appears, delivers information, and disappears. A Combobox costs 8 — it combines text input, dropdown selection, filtering, keyboard navigation, async loading, and the expectation that the user knows what they are looking for.
George Miller published “The Magical Number Seven, Plus or Minus Two” in 1956. His argument: human working memory holds roughly 7 items. Not bytes. Not pixels. Items. Chunks of meaning.
A page with a Button (3), a Card (4), an Input (5), and a Dialog (7) costs 19. That is a page-tier budget. Reasonable. But add a second Dialog and you are at 26. Add a DataTable (8) and you are at 34, bleeding into app-tier territory. The budget math is simple. The design judgment is not.
The question cognitive load scoring answers is not “is this component expensive?” Every designer knows a modal is heavier than a button. The question it answers is: how expensive is this composition? Not the individual ingredient — the recipe. A page with five cheap components can cost more than a page with two expensive ones if the cheap ones compete for the same attention.
Miller’s number was never about memory. It was about attention. The magical number seven is not how many things you can remember. It is how many things you can hold in focus at once. A design system that counts this — that tells the AI “you have 30 points to spend on this page, and you just spent 19 on four components” — is not measuring memory. It is measuring the cognitive price of being a user.
The Sentinel cost 7. We answered it with a 200 that cost 0. The cheapest response resolved the most expensive problem.
That is what good design does.