loaded 25959 LP boxes, 3928 NC boxes, 450 cells (1s)
1. Sturm (sympy count_roots) instead of Arb on 400 random LP boxes: 0 failures (1s)
2. semantic: 3000 LP boxes x 12 random points x 2407 y: max(F_true - F_box) = -2.621e-03 (must be <= ~1e-15); min(P - F_true) = 8.700e-03 (must be > 0) (2s)
3. LP corruption P*0.9: 2000/2000 rejected
3. LP corruption c0-1e-03: 19/2000 rejected
3. LP corruption c0-1e-05: 0/2000 rejected
3. LP corruption c0-1e-07: 0/2000 rejected
3. LP corruption c6=0: 1944/2000 rejected
3. LP corruption wrong sign s: 1198/2000 rejected
3. LP corruption r1-1/20: 2000/2000 rejected
3. LP corruption t-range x2: 1680/2000 rejected
3. LP corruption u2+1/20: 2000/2000 rejected
3. LP corruption target sqrt2 -> 1.4142: 1/2000 rejected
3. LP corruption target sqrt2 -> 1.4141: 4/2000 rejected
3. LP corruption target sqrt2 -> 1.414: 12/2000 rejected
3. LP corruption target sqrt2 -> 1.41: 293/2000 rejected
4. float slack over all 25959 LP boxes: dominance min 8.627e-04, 1% quantile 9.893e-04, median 2.206e-03; moment min 4.211e-07, 1% quantile 2.322e-04, median 8.596e-03 (12s)
   tightest dominance: s=-1 u[1.553773,1.560000] t[0.207110,0.258887] r[0.500000,0.550000] dom slack 8.627e-04, moment slack 3.131e-03
   tightest moment: s=-1 u[1.740000,1.750000] t[0.155333,0.168277] r[0.850000,0.900000] dom slack 1.838e-03, moment slack 4.211e-07
   200 tightest boxes: c0 lowered by 0.5 x min slack -> 200/200 still pass; c0 lowered by 2 x dominance slack -> 200/200 rejected
   200 tightest moment margins: c0 raised by 2 x moment slack -> 200/200 rejected
5. NC corruption target sqrt2 -> 1.41: 1/3928 rejected
5. NC corruption target sqrt2 -> 1.4: 133/3928 rejected
5. NC corruption target sqrt2 -> 1.3: 2539/3928 rejected
5. NC corruption r1-1/10: 3928/3928 rejected
5. NC corruption r1-1/40: 2800/3928 rejected
5. NC corruption t-range x2: 187/3928 rejected
6. tiling s=+1: removed: inside True, volume False, overlaps 0 -> REJECTED; duplicated: inside True, volume False, overlaps 1 -> REJECTED; shifted u by 1e-6: inside True, volume True, overlaps 1 -> REJECTED
6. tiling s=-1: removed: inside True, volume False, overlaps 0 -> REJECTED; duplicated: inside True, volume False, overlaps 1 -> REJECTED; shifted u by 1e-6: inside True, volume True, overlaps 1 -> REJECTED
7. one-coordinate cells P*0.9 (depth 0): 450/450 rejected
7. one-coordinate cells c0-1e-5 (depth 0): 450/450 rejected
7. one-coordinate cells a2+1/100 (depth 0): 306/450 rejected
7. one-coordinate cells target sqrt2 -> 1.414 (depth 0): 5/450 rejected
7. one-coordinate cells target sqrt2 -> 1.41 (depth 0): 107/450 rejected
done (42s)
