Agent Task Utility and Admission Integrity on Code-agent exact-effect benchmark
231STC CountCXI
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CXIUnit / N=400 repo episodes, Accounting note=73 parse failures; 61 no effect authorization; 35 task-quality failures2026.07 | 231 | — | — | — | — | 0 | 0 | 0 | 73 | 61 | 35 |