design — user-visible artifacts that work and are worth looking at

Generated visual artifacts fail in four ways, and each needs its own counter. Structure is drawn from habit rather than from the work the artifact supports — questions land under the thing they ask about, filters above the table, controls at the bottom of the form, whether or not the work can be done in that arrangement; the counter is a written model of that work, declared before layout and binding on it. Styling effort lands on the wrong layer while the functional layer collapses — pages ship with no empty state, no error path, no undo. Stated principles do not survive into output — what holds is external evidence: a render, a machine check, a counted budget, never your own account of what you did. And unforced style choices converge on the same few looks — distinctiveness must be derived from the subject at hand, never sampled from habit.

So work through six decisions, in order. Each bounds the next; skipping one is where those failures enter.

Purpose picks the metric, the task model picks the structure, register picks the rules, tokens hold the system, floors guard the layer that collapses, and verification is external evidence per genre — with distinctiveness derived from the subject and spent only where the register allows.

Decision 0 — name the purpose

State which one thing the artifact optimizes for. The four purposes want different and sometimes opposite designs, so one artifact picks one: