omp+0x-alpha beat both cc+fable and codex-sol in creating/refactoring a big eval setup. the former just knows where things should belong and completed the task all the way while the other two failed on both metrics.