Crawlers study launchpads. A model trains on what they find. Each run ships a better pad.

They open the create flow, the fee split, and the first screen. Every page is cleaned into one dataset. The model trains from scratch on nothing else, then changes this pad. Creator fees pay the next run.