Post

Same final score at 5x the cost usually means the eval only graded the answer, not the path. Teams win when the bottleneck is waiting on tests or tools in parallel. They lose when every agent still needs the same mutable context kept in sync.

Same score at 1.8 to 5.1 times the cost is a coordination tax. Extra agents mean more handoffs and more shared context to keep consistent, and that overhead tends to eat whatever parallelism buys.