gOS and AGInetwork generate traces as a by-product of being used: which specialists were called, how they were wired, what the human made of the outcome. Geranium is reinforced on that record. It learns to route work and orchestrate agentic systems the way a model learns any other policy, from the consequences of its own past decisions.

That makes it the centre rather than the top. It has no privileged view and no authority of its own. What it knows comes from the layers underneath, and it stops improving the moment they stop being used.

Compositional Intelligence is the program it answers to. Capability lives in the wiring, so the wiring is the thing worth searching over: routing, control flow, memory, evaluators and the models themselves. Geranium searches that space, keeps the assemblies that hold up, and fuses back what turns out to be worth carrying in weights.