{"paper":{"title":"AgentLens: Revealing The Lucky Pass Problem in SWE-Agent Evaluation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Binary pass rates in SWE-agent tests equate chaotic trial-and-error successes with systematic ones.","cross_cats":["cs.AI"],"primary_cat":"cs.SE","authors_text":"Benjamin Steenhoek, Gaurav Mittal, Pingping Lin, Priyam Sahoo, Shengjie Ma, Xiaomin Li, Yu Hu","submitted_at":"2026-05-13T03:00:57Z","abstract_excerpt":"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 Pa"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"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.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That Prefix Tree Acceptor references built by merging multiple passing solutions accurately represent principled behavior without incorporating lucky elements from the source trajectories.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"10.7% of passing SWE-agent trajectories are Lucky Passes with chaotic behaviors, and a quality score based on process references changes model rankings across eight backends.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Binary pass rates in SWE-agent tests equate chaotic trial-and-error successes with systematic ones.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"adce0b450020fee18eaa30704ab28f998777265f989745261d06c17949418e61"},"source":{"id":"2605.12925","kind":"arxiv","version":1},"verdict":{"id":"65a2b27d-f897-4cc8-8786-4ee98826b4a1","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T18:46:32.801828Z","strongest_claim":"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.","one_line_summary":"10.7% of passing SWE-agent trajectories are Lucky Passes with chaotic behaviors, and a quality score based on process references changes model rankings across eight backends.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That Prefix Tree Acceptor references built by merging multiple passing solutions accurately represent principled behavior without incorporating lucky elements from the source trajectories.","pith_extraction_headline":"Binary pass rates in SWE-agent tests equate chaotic trial-and-error successes with systematic ones."},"references":{"count":61,"sample":[{"doi":"","year":null,"title":"arXiv preprint arXiv:2410.20285 , year=","work_id":"e26e6ec9-37e3-47d8-b494-e0a227e77e36","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Swe-rebench: An automated pipeline for task collection and decontaminated evaluation of software engineering agents","work_id":"c819f5fe-32b6-41d1-aeb5-2d80c6fa8474","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Islem Bouzenia, Premkumar Devanbu, and Michael Pradel","work_id":"9584953a-2ac8-49b2-bcd3-44924077f86d","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2026,"title":"2026 , note =","work_id":"ad7bad76-b98f-4522-8476-a382ae24a4b8","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.5281/zenodo.19357078","year":null,"title":"doi:10.5281/zenodo.19357078 , url =","work_id":"5aab64ed-4762-4aa1-922b-18ae4b5b8b27","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":61,"snapshot_sha256":"66df3e8e7819f1fb70a7bba3a29167044596023591235646b3882339a9f38ab8","internal_anchors":14},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}