Gradient: p_gauss(σ)=e^{−1/2}; p_shep(σ)=e^{−1}; d½_gauss=σ√(2ln2); d½_shep=σ·ln2

Sympy-exact symbolic witness of generalization-gradient canonical forms (Gaussian + Shepard-exponential), their 1-sigma anchors, and their half-response half-widths. Setup: d (stimulus distance from training), σ (generalisation width /…