Every theory starts with a toymodel.
Most scientific tools force a choice. A notebook reaches anywhere and remembers nothing. A specialist package reaches deep and stops at its edge. Toymodel removes the choice. One composable part builds everything in it.
That part is a component, a small typed program that does one readable job behind a stated interface. Components connect into a sector, one graph that holds a whole piece of work. Seal a sector you trust into a Theory, a packaged computation that keeps its interface, and carry it into the next problem. Depth accumulates. Breadth costs nothing extra, because a new field asks for new components, not for a new tool.
Components also learn. Point one at data, train it in place, and it keeps the interface it had before, so a trained piece composes like every other piece — beside your solvers, your tables, and your charts, not inside a separate framework. The Toymodel Generative AI harness works within that structure rather than in place of it. You can read every value along the way.