Interpreting Coding Agents Through Their Behavioral Trajectories A study of 143,209 execution trajectories from 125 agent configurations spanning 117 unique models finds that standard coding-agent benchmarks, which score binary task success, treat execution as a black box and obscure the behaviors that produce outcomes. The analysis aims to enable comparison of agent behavior across configurations rather than judging agents solely on pass or fail. Standard coding-agent benchmarks primarily evaluate binary task success, treating execution as a black box and obscuring the behaviors that produce those outcomes. We analyze a large corpus of 143,209 execution trajectories from 125 agent configurations spanning 117 unique models. To enable comparis