Hmm. First I thought my math was all wrong. Then I thought it was wrong just when δ(1-σ) ≥ 1/8. Then I thought about a way of doing something entirely different for that case, and turned the crank for a while and got exactly the same answer as for δ(1-σ) ≤ 1/8. This is either incorrect or spookily correct. In fact, if it's correct, it's sort of very correct, since it shows that, quantitatively, things oughtn't be nonsmooth across that boundary, even if conceptually they are.