Solves hard problems with a multi-model consensus: several AI advisors propose, cross-review and sign one plan, tested against fixed pass/fail numbers.