The Valence Curve: Measurement Design and Falsification Criterion for Thesis B of the Axiomatica Universalis

This paper formalises Thesis B of Beyer 2026g: the valence function V(t) = −dH/dt has a characteristic universal form in real learning processes. A formal derivation of the expected shape from the AU framework is given (Theorem M1). A concrete measurement protocol is specified for three systems: LLM training curves, child language acquisition (CHILDES), and market entropy. The falsification criterion is precise: if normalised valence curves V(t/τ)/V_max do not collapse to a universal curve across systems, Thesis B is falsified. Two open questions remain: the derivation of the shape parameter β from the axioms, and the treatment of negative valence (catastrophic forgetting) under Axiom 3.