# mutations of fsearch_production.c (scratch copies only)
m1_noinvthr: okc[c] no longer requires cinv[c] >= THR (only removes pruning: classes unchanged, nodes up)
m2a_thr_minus1: threshold t-1 instead of t (only removes pruning: classes unchanged, nodes up)
m2b_thr_plus1: after the root gluing all further gluings need type > t (too strong: loses solutions; count must drop)
m3_dropsub: sub-slices I=0..S-1 select index = I mod (S+1): candidates with index = S mod (S+1) are lost (count must drop)
m4_noloop: loop classes allowed during the search (leaf test full_check must reject: classes unchanged, leaves > solutions)
m5_noloop_nofree: loop classes allowed AND orbit-size test of the leaf check disabled (non-free actions accepted: count must rise)
m6_noprop: no ridge-cycle propagation and no quick test (pure weakening: classes unchanged, nodes up enormously; run with a node cap)
m1b_noinvthr_both: inverse-type threshold test removed in okc AND in glue() (only the type seen from the first facet is tested): weakening, classes unchanged, nodes up
m7_ell_only_c: elliptic filter tests only the class itself, not the inverse class (weakening or no-op)
m8_quick_strict: quick test (filter_domain) additionally rejects about half of the candidates by a parity of flag labels (too strong: must lose solutions)
