EPK Platform

Two audiences read the same artifact and must never see the same thing. A promoter deciding a fee needs an artist who looks singular. A DJ deciding whether to buy needs to see the range and picture themselves in it. Show the promoter the range and the artist looks templated at the exact moment their price is being set.

The evidence came from the second kit rather than from research. Building Supreme La Rock's site on the same tier as my own was supposed to be a re staging of identical structure in a different skin. It was not. His music page and mine rendered different section sets, and the reason was a single condition: the editorial music layout only engaged when a collection field was populated, so my kit silently fell back to a simpler variant. Nothing looked broken. The two kits had quietly become two products.

That is the failure mode the whole system exists to prevent, and it changed how correctness is defined here. Structural agreement between two kits on the same tier is not a style preference that can be settled by looking at them. It is a property that has to be checked, by diffing the rendered class sets of the two pages rather than by reading the theme files and believing them.

Should the artist get a builder, or should the controls stay with the designer? I built the system with no builder interface and no exposed values for accent, density, corner style, or image treatment. Variation ships as another named theme instead. The forcing example is small and total: the token for text on the accent color is black because black on that red measures 5.5 to 1, and a color picker that let anyone choose a darker red would make every primary button on the site unreadable, with nobody present to measure it. The tradeoff is real and I am not hiding from it. Refusing the builder means the service does not scale past my own hours.

What I will not claim is speed. That AI made it fast is the least interesting thing about it and the easiest thing to dismiss. What AI actually bought was the ability for one person to hold four applications, a token system, and a deploy pipeline in production at once, which is a scope that would otherwise have needed a team and would therefore never have been attempted. The judgment stayed where it always was: deciding what never gets exposed.

There is no booking conversion data, because the kits are new. No client has reached a renewal. Two of the four tiers are still only defined in my head, which is the honest reason the demo shows two and not four. And there is still no tier field in the code, so structural agreement between kits is checked by diffing rendered output rather than asserted by a type. That last one is a defect, not a design choice, and it is the next thing to fix.