Rethinking the Evaluation of Harness Evolution for Agents

  • 类型:arxiv
  • 标识:2607.12227
  • 链接:https://arxiv.org/abs/2607.12227
  • 主分类:evaluation
  • 形态:benchmark
  • 被引:45
  • 被引来源:Semantic Scholar
  • S2被引:45
  • OpenAlex被引:0
  • 影响力被引:2
  • TLDR:The evaluation of automatic harness evolution procedures is revisited and the need for matched-budget baselines and held-out evaluation to distinguish genuine harness improvements from benchmark-specific search and overfitting is highlighted.
  • OpenAlex ID:W7168361355
  • OpenAlex DOI:10.48550/arxiv.2607.12227
  • DOI:10.48550/arxiv.2607.12227
  • DOI来源:OpenAlex
  • 开放获取:green
  • 开放获取链接:https://doi.org/10.48550/arxiv.2607.12227
  • OpenAlex更新:2026-07-19
  • 副分类:agent
  • 待LLM分类:否
  • 标题中文:Rethinking the Evaluation of Harness Evolution for Agents(重新审视 Agent 的 Harness 演化评估)
  • TLDR中文:重新审视自动 harness 演化流程的评估方法,强调需要采用匹配预算的基线(matched-budget baselines)和留出评估(held-out evaluation),以区分真正的 harness 改进与针对特定基准的搜索和过拟合。
  • 来源文件:
  • /inbox/tom/_candidates/2026-07-18-agent-rag-longcontext-candidates.json
  • [S2 enrich]
  • /inbox/tom/_candidates/2026-07-19-agent-rag-longcontext-candidates.json
  • [OpenAlex backfill]
  • /inbox/tom/_candidates/2026-07-20-agent-rag-longcontext-candidates.json
  • /inbox/tom/_candidates/2026-07-20-agent-memory-tool-use-candidates.json