Example. (Reduction loses a factor). Let k=Fp, K=k((t)) and L=k((u)) with up=t. Keep ∣t∣=c with 0<c<1, so ∣u∣=c1/p. Take R=k[[t]], R′=k[[u]], and regard X=Pk[[v]]d as an R-model via t=vp, where d≥2. Grouping power-series exponents modulo p identifies k[[v]] with the free R-module on 1,v,…,vp−1. Thus X is integral, proper and flat over R, and its generic fibre is normal integral over K.
Take exponent one throughout, (L0,σ0)=(O,v) with vertical support, and use d coordinate hyperplane sections of O(1) for the other factors. Their intersection with div(v) is one k-point of length one, so JR(X)=λK. The full base change is Y=PR′[v]/(vp−up)d≃PR′[ϵ]/(ϵp)d,v=u+ϵ. It has generic length p along its one reduced component. After the hyperplane operations the quotient by v is R′/(up), of length p; multiplication by v is injective because vp=up and the algebra is R′-free. Equivalently, compute on the full fundamental cycle p[PR′d]. Thus degk′βY=p,JR′(Y)=(λK/p)p=λK. Replacing Y by its reduction before forming its fundamental cycle gives degree one and the incorrect value λK/p. Its reduction is a different scheme, not a permissible deletion of generic multiplicity.