Break single-model echo-chamber effects by having one model plan and another challenge the plan across up to three rounds. Use when a design is self-consistent but untested, when you suspect confirmation bias in a plan, or before committing to a large build that only one model has seen.
