A tempting design, when modelling a developing agent, is to keep the agent's behaviour fixed and pass age in as a parameter. Maybe some weights get a little softer. Maybe the responses get a little warmer. But the underlying machine is the same.
Paper 2 takes the opposite view. Age isn't a parameter to the machine; it's part of the machine's state. STAB, OPEN, AGREE, ASSERT, SENSE, LEARN, plus an experience axis — seven dimensions that compose into the agent's currently-active behaviour graph.
The practical consequence is that an agent at age 30 days has a different graph, not just different weights, from the same agent at age 300 days. The shape of the developmental curve is the shape of the agent.
The harder consequence — and the part we're still working on — is that the curve isn't fully forwards-only. Regressions are possible under sustained negative feedback. We're modelling that explicitly because pretending otherwise produced an agent that felt wrong even when its outputs were technically correct.
Calm is also resilience. We're trying to formalise both at once.