Functional Reward by Round
Average functional reward across tasks by round. Missing task-round rewards are excluded from each average.
MeanJudge by Round
Average correctness judge score by round, after averaging replicate scores within each task.
SSR by Round
Stable Solve Rate by round, using the fixed task denominator and the 0.85 correctness threshold.
Tasks
Job-level task outcomes. Select a task to inspect its native multi-turn timeline.