(D1) P(K_0 <= 1) for m_0 = 3 paths of length 2 = 47/90 = 0.5222222222222223
     Monte Carlo (evidence, 4e5 samples): 0.5233
(D2) for 2 <= j <= 3000: Var K_k <= 3/j^4 and Cantelli <= Chebyshev <= 3/j^2 (exact): OK
(D3) P(|G(J)|=2) = P(E) >= 0.076808 (Chebyshev),  >= 0.105965 (Cantelli)
     P(|G(J)|=4) >= P(K_0 <= 1) = 47/90 = 0.522222
     (for comparison, product over k != 0 alone: Chebyshev 0.160761, Cantelli 0.221786)
(D4) Monte Carlo (evidence, |k| <= 12, 4000 reps): P(|G(J)|=4) ~ 0.543
