nx_superopt_govern_test.nx

nx_superopt_govern_test.nx -- the team builds END TO END, no human in the rule: the GENERATOR authors a lowering for each constant (by search), the VERIFIER proves it 1:1, the COUNCIL (3->2->1) governs admission, and the winners are BANKED into a rewrite library the team grew ITSELF. I wrote the searcher and the council; the team wrote + selected the rules. GENERATE (so_best) -> VERIFY (so_verify) -> GOVERN (cc_council) -> BANK A genuine improvement (cheaper than mul) is ADMITTED; a constant with no cheaper form (the LCG prime -> mul) is "not worth it" and the council does NOT bank it -- the team only keeps real wins. Known answer: every reducible constant banked, the non-reducible one rejected. exit 0.