wrong-n mutations caught: 600 / 600
shifted psi: 63 graphs, self-dual 63, tau != n in 59 (all rejected as certificates of n)
embedding mutation (B u psi(B'*), B' = B with one S/P node swapped): tau = n always; self-dual check rejects 380, accepts 20 (accepted ones: B' plane-isomorphic to B or other symmetry)
tau mod p cross-check on 500 certificates x 2 primes: mismatches 0
