MindForm — Trait Dimension: A Continuous, Bounded Dynamical Model of OCEAN Personality in LLM Agents

The Big Five (OCEAN) taxonomy is the most extensively validated model of human personality trait structure, yet most attempts to give language-model agents a "personality" reduce it to a static, prompt-time descriptor that never changes in response to what the agent experiences. This paper documents the Trait Dimension node of MindForm, which instead treats OCEAN as a live state vector shaped continuously by experience. We describe why the Big Five was adopted over alternative taxonomies, how traits are represented as continuous scores anchored to a slower-moving temperament baseline, and how trait state is actually updated: an LLM-primary path that judges the directional pressure of an experience directly from text, with a deterministic, appraisal-and-matrix fallback for when the LLM is unavailable. Both paths update traits through the same diminishing-returns mechanism, so no single experience can push a trait past its bounds. We describe how current trait state feeds back into how future experience is interpreted — differently on each formation path — and situate this continuous design against the discrete, threshold-triggered trait model in our earlier StableMind work. We close with an open limitation stated plainly: the direction of primary-path trait change is reasonably well-grounded, but its magnitude is not yet validated against any ground truth, since no dataset we are aware of labels how much a given experience changes a person's traits rather than what those traits already are.