nx_benchmark_release_candidate_t179.nx

nx_benchmark.nx -- TRUE competitive benchmarking, not cherry-picking (operator: a recurring flaw -- pick a few random features and declare "we suck" or "we exceed", ignoring the hundreds of features the competitor has AND ignoring what we built). The fix is a METHODOLOGY that REFUSES to issue a verdict until the comparison is actually COMPLETE: 1. ENUMERATE the competitor's full known feature surface (a COUNT you cannot hide behind). 2. COMPLETENESS GATE -- if the comparison covers too small a fraction of that surface, it is CHERRY-PICKED and NO verdict is allowed (the exact flaw, caught). 3. WEIGHTED COVERAGE both ways -- features weighted by importance, our support of theirs AND their support of ours (so we neither pretend to match their breadth nor erase our unique strengths). 4. HONEST VERDICT -- only when valid: AHEAD / PARITY / BEHIND, or DIFFERENTIATED when each side covers little of the other (we play different games -- the honest answer that is neither "we suck" nor "we exceed"). The Critic/Examiner ENFORCE bm_is_valid before any competitive claim is published. license_tier: ORIGINAL.