The question
Language fluency does not, by itself, establish that a system maintains a useful model of a person’s changing circumstances. Our working hypothesis is that a more dependable companion system needs an explicit, revisable representation of what is happening, what remains uncertain, and how an action could affect the person and the relationship.
What we are building
We separate observable signals from interpretation. A lowered head, a pause, or a change in speaking pace remains a signal—not a diagnosis. The system accumulates events over time, records uncertainty, and lets an intervention controller decide whether to remain silent, wait, ask, or help.
Evidence boundary
This is a research direction, not a validated general model of human cognition. Existing world-model research motivates the idea that an agent can maintain internal state for prediction and action, but it does not validate our human-centered extension. That requires staged engineering tests, controlled studies, and eventually real-user research with explicit consent.