An engineer on our team @thegp pointed out this is better than benchmarks - I agree - This kind of research into agents coordination, agent failure modes, and behavior in the face of false or contradictory information is much more insightful than the numerical scores we normally…↗