Dimensional Emotion vs Discrete Labels — Why E-PQ Uses Vectors

Every AI companion models emotion as labels: happy, sad, neutral. Frank uses 5 continuous vectors in [-1,1] space — precision, risk, empathy, autonomy, vigilance. Inspired by Russell's circumplex model (1980).

Why it matters: discrete labels transition through programmed edges. Continuous vectors drift independently — empathy rises while vigilance drops and risk stays flat. The combinatorial richness is orders of magnitude higher.

Empirical finding: Qwen 3B can meaningfully distinguish ~5 personality dimensions. At 6+, it conflates them. This drove the choice of exactly 5.

Full paper: https://github.com/gschaidergabriel/P roject-Frankenstein/blob/main/papers/ PARAMETER_ARCHITECTURE.md