RoadmapBench Opus47 Strict100 Clean
收藏官方服务：
资源简介：
# RoadmapBench Opus47 Strict100 Clean Split This package contains 251 RoadmapBench tasks with Opus47 rollouts. Criteria: - Opus47 reward in `[0, 0.70]` - At least 100 Opus47 completion files per task - No `exception.txt` in the selected Opus47 trial Completion count: min 100, max 401, average 163.37 Zip contents use a split layout: ```text tasks/<task>/ environment/ instruction.md tests/ solution/ task.toml rollouts/opus47/<task>/ agent/trajectory.json agent/completions/ verifier/reward.txt verifier/reward.json result.json metadata.json ``` SHA256: `a42c6a55116e2394921a8fe3671c6553c0141c3c342f58f48b9afbed571a05bd`
提供机构：
maas创建时间：
2026-07-15
