== B12 (seven columns, four copies)
  mu = -1/4, pi_+ = ['0', '1/12', '1/12', '1/4', '1/6', '1/6', '1/6', '1/12'], pi_- = ['1/5', '3/20', '3/20', '1/20', '1/10', '1/10', '1/10', '3/20']
  h' Fourier: [((), '-1/4'), ((0,), '-1/8'), ((1,), '-1/8'), ((2,), '-1/8'), ((0, 1), '-1/4'), ((0, 2), '-1/4'), ((0, 1, 2), '1/8')]
  rho_+- are probability laws with the assigned column, pair and copy marginals of pi_+- (both signs)
  coefficients of R aggregated by law equal those of h' (L1 on 12 pairs (coefficient sum -1/2), L2 only inside copies (coefficient 0), cubic 1/8)
  R <= 0 at all 128 vertices; max R = 0; R = 0 on supp rho_+ (12 points); zeros of R: 42
  supp rho_+: 12 points
  supp rho_-: 18 points
  32R = -8 - 2(y0+y1+y3+y4+y5+y6) - (1-y2)(y0+y6)(y1+y3) - 2(y1y3+y0y4+y2y4+y4y5+y2y5+y5y6) at all vertices
  max over y4,y5 of 32R: -t(2+t) if y2 = 1 or s = 0; -8-6t-t^2 if y2 = -1, s = 2; t(2-t) if y2 = -1, s = -2
  exact random tests (280 decoders, k <= 3, all nu): Gamma = E[R] <= 0 always; largest value 0
