(a) law of Y: triangular on (-1,1), E Y^2 = 1/6, E Y^4 = 1/15 (matches U1-U2)
(a) exact P(|K|<=M) vs local-limit value 2M*sqrt(3/(pi m)):
   m=  1 M=1/100 P=0.019900  LLT=0.019544  ratio=1.0182
   m=  1 M=1     P=1.000000  LLT=1.954410  ratio=0.5117
   m=  2 M=1/100 P=0.013333  LLT=0.013820  ratio=0.9648
   m=  2 M=1     P=0.916667  LLT=1.381977  ratio=0.6633
   m=  4 M=1/100 P=0.009587  LLT=0.009772  ratio=0.9811
   m=  4 M=1     P=0.774752  LLT=0.977205  ratio=0.7928
   m=  8 M=1/100 P=0.006845  LLT=0.006910  ratio=0.9906
   m=  8 M=1     P=0.610129  LLT=0.690988  ratio=0.8830
   m= 16 M=1/100 P=0.004863  LLT=0.004886  ratio=0.9953
   m= 16 M=1     P=0.458038  LLT=0.488603  ratio=0.9374
   m= 32 M=1/100 P=0.003447  LLT=0.003455  ratio=0.9977
   m= 32 M=1     P=0.334302  LLT=0.345494  ratio=0.9676
   m= 64 M=1/100 P=0.002440  LLT=0.002443  ratio=0.9988
   m= 64 M=1     P=0.240272  LLT=0.244301  ratio=0.9835
   max over grid of sqrt(m)*P(|K|<=M)/M = 1.9900  (claimed <= 2*1.18 = 2.36)
(b) finite-window transfer-matrix check (m_k=ceil((|k|+1)^3), R=6):
   seed 0: argmin k0=+0 |K|=0.281; window-flip violations for sigma*/all-sat = 0; other single-wall configs beaten 11/11
   seed 1: argmin k0=+0 |K|=0.313; window-flip violations for sigma*/all-sat = 0; other single-wall configs beaten 11/11
   seed 2: argmin k0=+0 |K|=0.140; window-flip violations for sigma*/all-sat = 0; other single-wall configs beaten 11/11
   seed 3: argmin k0=+0 |K|=0.203; window-flip violations for sigma*/all-sat = 0; other single-wall configs beaten 11/11
   seed 4: argmin k0=-6 |K|=0.135; window-flip violations for sigma*/all-sat = 0; other single-wall configs beaten 11/11
   seed 5: argmin k0=+0 |K|=0.108; window-flip violations for sigma*/all-sat = 0; other single-wall configs beaten 11/11
   (note: walls adjacent to the window edge may not be beatable inside the window; see count)
(c) p=3.0: sum_(|k|<=100) min(1, 2.36/sqrt(m_k)) = 7.673
(c) p=3.0: sum_(|k|<=10000) min(1, 2.36/sqrt(m_k)) = 8.516
(c) p=3.0: sum_(|k|<=1000000) min(1, 2.36/sqrt(m_k)) = 9.601
(c) p=2.0: sum_(|k|<=100) min(1, 2.36/sqrt(m_k)) = 20.451
(c) p=2.0: sum_(|k|<=10000) min(1, 2.36/sqrt(m_k)) = 42.118
(c) p=2.0: sum_(|k|<=1000000) min(1, 2.36/sqrt(m_k)) = 64.854
