"""tasks/hard/task.py — Hard difficulty task definition.""" TASK_ID = "hard" DESCRIPTION = ( "Content moderation — hard difficulty. " "Agent must reason over conflicting evidence from credible sources and " "make a moderation decision (allow / flag / remove). " "Ground truth is ambiguous — all scoring dimensions count equally." ) SCORE_RANGE = [0.0, 1.0] N_SCENARIOS = 5 class HardTask: """OpenEnv task entrypoint for hard difficulty.""" task_id = TASK_ID description = DESCRIPTION score_range = SCORE_RANGE n_scenarios = N_SCENARIOS def grade(self, episode: dict) -> float: from tasks.hard.grader import grade_episode return grade_episode(episode)