analiza aiqbench
AgentLens exposes a critical flaw in current SWE-agent evaluation by distinguishing between principled solutions and "Lucky Passes" where agents achieve success through chaotic trial-and-error rather than robust logic. By introducing a process-level assessment framework, the research reveals that up to 23.2% of passing trajectories across different models rely on wasteful behaviors like blind retries, which can significantly skew performance rankings. This shift from binary pass/fail metrics to quality-scored trajectory analysis provides a necessary foundation for building more reliable and efficient software engineering agents.
AgentLens: Revealing The Lucky Pass Problem in SWE-Agent Evaluation
arXiv:2605.12925v3 Announce Type: replace-cross Abstract: Evaluation of software engineering (SWE) agents is dominated by a binary signal: whether the final patch passes the tests. This outcome-only view treats a principled solution and a chaotic trial-and-error process as equivalent. We show that this equivalence is empirically false. We evaluate 2,614 OpenHands trajectories from eight model backends on 60 SWE-bench Verified tasks. Of these, 47 have enough passing trajectories to construct task-level process references, yielding a 1,815-trajectory evaluation subset. Among passing trajectories in this subset, 10.7% exhibit behavior we call a Lucky Pass: regression cycles, blind retries, missing verification, or temporally disordered exploration, implementation, and verification. We introduce AgentLens, a framework for process-level assessment of SWE-agent trajectories, and define AgentLens-Bench, a dataset of 1,815 trajectories annotated with quality scores, waste signals, divergence points, and 47 task-level Prefix Tree Acceptor (PTA) references. AgentLens builds PTA references by merging multiple passing solutions for the same task, and uses a context-sensitive intent labeler to assign actions to Exploration, Implementation, Verification, or Orchestration based on trajectory history rather than tool identity alone. On AgentLens-Bench, the quality score separates passing trajectories into Lucky, Solid, and Ideal tiers and further decomposes Lucky Passes into five recurring mechanisms. Across the eight model backends, Lucky rates range from 0.5% to 23.2%, and some models move by as many as five rank positions when ranked by quality score instead of pass rate. We plan to release the project repository soon, including AgentLens-Bench artifacts, the AgentLens SDK, and the analysis tooling.
Abstract:Evaluation of software engineering (SWE) agents is dominated by a binary signal: whether the final patch passes the tests. This outcome-only view treats a principled solution and a chaotic trial-and-error process as equivalent. We show that this equivalence is empirically false. We evaluate 2,614 OpenHands trajectories from eight model backends on 60 SWE-bench Verified tasks. Of these, 47 have enough passing trajectories to construct task-level process references, yielding a 1,815-trajectory evaluation subset. Among passing trajectories in this subset, 10.7% exhibit behavior we call a Lucky Pass: regression cycles, blind retries, missing verification, or temporally disordered exploration, implementation, and verification. We introduce AgentLens, a framework for process-level assessment of SWE-agent trajectories, and define AgentLens-Bench, a dataset of 1,815 trajectories annotated with quality scores, waste signals, divergence points, and 47 task-level Prefix Tree Acceptor (PTA) references. AgentLens builds PTA references by merging multiple passing solutions for the same task, and uses a context-sensitive intent labeler to assign actions to Exploration, Implementation, Verification, or Orchestration based on trajectory history rather than tool identity alone. On AgentLens-Bench, the quality score separates passing trajectories into Lucky, Solid, and Ideal tiers and further decomposes Lucky Passes into five recurring mechanisms. Across the eight model backends, Lucky rates range from 0.5% to 23.2%, and some models move by as many as five rank positions when ranked by quality score instead of pass rate. We plan to release the project repository soon, including AgentLens-Bench artifacts, the AgentLens SDK, and the analysis tooling.
Software Engineering (cs.SE); Artificial Intelligence (cs.AI)
arXiv:2605.12925 [cs.SE]
(or arXiv:2605.12925v3 [cs.SE] for this version)
Text integral sub licența deschisă a publicației · 258 cuvinte
Subiecte conexe
AgentLens: Revealing The Lucky Pass Problem in SWE-Agent Evaluation
arXiv · 7 oct. 2026
Latent reasoning architectures would likely undermine CoT, our strongest oversight tool
AI Alignment Forum · 23 sept. 2026
How Would We Know? Reflections on trying to make AI go well amid deep uncertainty
LessWrong · 5 oct. 2026
How Sonos rebooted itself
The Verge · 3 sept. 2026