{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:T3ZH7MLMLV4Q44VICALH5O7L62","short_pith_number":"pith:T3ZH7MLM","schema_version":"1.0","canonical_sha256":"9ef27fb16c5d790e72a810167ebbebf691be5ec383e4a458e0a2f5245efe7d0b","source":{"kind":"arxiv","id":"2504.07164","version":1},"attestation_state":"computed","paper":{"title":"R2E-Gym: Procedural Environments and Hybrid Verifiers for Scaling Open-Weights SWE Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.SE","authors_text":"Ion Stoica, Jaskirat Singh, Koushik Sen, Liang Zheng, Manish Shetty, Naman Jain","submitted_at":"2025-04-09T17:55:19Z","abstract_excerpt":"Improving open-source models on real-world SWE tasks (solving GITHUB issues) faces two key challenges: 1) scalable curation of execution environments to train these models, and, 2) optimal scaling of test-time compute. We introduce AgentGym, the largest procedurally-curated executable gym environment for training real-world SWE-agents, consisting of more than 8.7K tasks. AgentGym is powered by two main contributions: 1) SYNGEN: a synthetic data curation recipe that enables scalable curation of executable environments using test-generation and back-translation directly from commits, thereby red"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2504.07164","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.SE","submitted_at":"2025-04-09T17:55:19Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"70098fc4624de999c5f149328d433a6d363579285e9b244a7aaf326a47f6c794","abstract_canon_sha256":"294bcbb6cd841dd2cc410cfd316752427c0170b5de2ca9bd127c314c8670a96e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:47:04.459080Z","signature_b64":"aadlz3kd6xq6e78NJqMTYfZQP5npNvacIWEEp6QT1NmwLAeou1/Oa2cJfawyHjVa0vIc5K0dLe1q6hWWvmiiBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9ef27fb16c5d790e72a810167ebbebf691be5ec383e4a458e0a2f5245efe7d0b","last_reissued_at":"2026-07-05T10:47:04.458615Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:47:04.458615Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"R2E-Gym: Procedural Environments and Hybrid Verifiers for Scaling Open-Weights SWE Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.SE","authors_text":"Ion Stoica, Jaskirat Singh, Koushik Sen, Liang Zheng, Manish Shetty, Naman Jain","submitted_at":"2025-04-09T17:55:19Z","abstract_excerpt":"Improving open-source models on real-world SWE tasks (solving GITHUB issues) faces two key challenges: 1) scalable curation of execution environments to train these models, and, 2) optimal scaling of test-time compute. We introduce AgentGym, the largest procedurally-curated executable gym environment for training real-world SWE-agents, consisting of more than 8.7K tasks. AgentGym is powered by two main contributions: 1) SYNGEN: a synthetic data curation recipe that enables scalable curation of executable environments using test-generation and back-translation directly from commits, thereby red"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.07164","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2504.07164/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2504.07164","created_at":"2026-07-05T10:47:04.458670+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.07164v1","created_at":"2026-07-05T10:47:04.458670+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.07164","created_at":"2026-07-05T10:47:04.458670+00:00"},{"alias_kind":"pith_short_12","alias_value":"T3ZH7MLMLV4Q","created_at":"2026-07-05T10:47:04.458670+00:00"},{"alias_kind":"pith_short_16","alias_value":"T3ZH7MLMLV4Q44VI","created_at":"2026-07-05T10:47:04.458670+00:00"},{"alias_kind":"pith_short_8","alias_value":"T3ZH7MLM","created_at":"2026-07-05T10:47:04.458670+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":26,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2607.06065","citing_title":"SWE-Review: Closing the Loop on Issue Resolution with Agentic Code Review","ref_index":13,"is_internal_anchor":true},{"citing_arxiv_id":"2604.17931","citing_title":"LiteResearcher: A Scalable Agentic RL Training Framework for Deep Research Agent","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2606.23961","citing_title":"Forget Without Compromise: Nexus Sampling for Streaming KV-Cache Eviction Under Fixed Budgets","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2606.19613","citing_title":"StaminaBench: Stress-Testing Coding Agents over 100 Interaction Turns","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2606.20683","citing_title":"From Question Answering to Task Completion: A Survey on Agent System and Harness Design","ref_index":214,"is_internal_anchor":false},{"citing_arxiv_id":"2606.18284","citing_title":"Breaking the Solver Bottleneck: Training Task Generators at the Learnable Frontier","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05080","citing_title":"AutoLab: Can Frontier Models Solve Long-Horizon Auto Research and Engineering Tasks?","ref_index":58,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03461","citing_title":"What Makes Interaction Trajectories Effective for Training Terminal Agents?","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03077","citing_title":"Libra: Efficient Resource Management for Agentic RL Post-Training","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28436","citing_title":"Dockerless: Environment-Free Program Verifier for Coding Agents","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25624","citing_title":"CUA-Gym: Scaling Verifiable Training Environments and Tasks for Computer-Use Agents","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2606.12191","citing_title":"Agentic Environment Engineering for Large Language Models: A Survey of Environment Modeling, Synthesis, Evaluation, and Application","ref_index":171,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21996","citing_title":"From Patches to Trajectories: Privileged Process Supervision for Software-Engineering Agents","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2512.18552","citing_title":"Toward Training Superintelligent Software Agents through Self-Play SWE-RL","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15815","citing_title":"BootstrapAgent: Distilling Repository Setup into Reusable Agent Knowledge","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05112","citing_title":"Rollout Pass-Rate Control: Steering Binary-Reward RL Toward Its Most Informative Regime","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2510.18270","citing_title":"Can Old Tests Do New Tricks for Resolving SWE Issues?","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2509.08827","citing_title":"A Survey of Reinforcement Learning for Large Reasoning Models","ref_index":228,"is_internal_anchor":false},{"citing_arxiv_id":"2511.16858","citing_title":"Investigating Test Overfitting on SWE-bench","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2504.19678","citing_title":"From LLM Reasoning to Autonomous AI Agents: A Comprehensive Review","ref_index":178,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14445","citing_title":"FrontierSmith: Synthesizing Open-Ended Coding Problems at Scale","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12925","citing_title":"AgentLens: Revealing The Lucky Pass Problem in SWE-Agent Evaluation","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2604.00824","citing_title":"Yet Even Less Is Even Better For Agentic, Reasoning, and Coding LLMs","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18543","citing_title":"ClawEnvKit: Automatic Environment Generation for Claw-Like Agents","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05112","citing_title":"Rollout Pass-Rate Control: Steering Binary-Reward RL Toward Its Most Informative Regime","ref_index":26,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/T3ZH7MLMLV4Q44VICALH5O7L62","json":"https://pith.science/pith/T3ZH7MLMLV4Q44VICALH5O7L62.json","graph_json":"https://pith.science/api/pith-number/T3ZH7MLMLV4Q44VICALH5O7L62/graph.json","events_json":"https://pith.science/api/pith-number/T3ZH7MLMLV4Q44VICALH5O7L62/events.json","paper":"https://pith.science/paper/T3ZH7MLM"},"agent_actions":{"view_html":"https://pith.science/pith/T3ZH7MLMLV4Q44VICALH5O7L62","download_json":"https://pith.science/pith/T3ZH7MLMLV4Q44VICALH5O7L62.json","view_paper":"https://pith.science/paper/T3ZH7MLM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.07164&json=true","fetch_graph":"https://pith.science/api/pith-number/T3ZH7MLMLV4Q44VICALH5O7L62/graph.json","fetch_events":"https://pith.science/api/pith-number/T3ZH7MLMLV4Q44VICALH5O7L62/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/T3ZH7MLMLV4Q44VICALH5O7L62/action/timestamp_anchor","attest_storage":"https://pith.science/pith/T3ZH7MLMLV4Q44VICALH5O7L62/action/storage_attestation","attest_author":"https://pith.science/pith/T3ZH7MLMLV4Q44VICALH5O7L62/action/author_attestation","sign_citation":"https://pith.science/pith/T3ZH7MLMLV4Q44VICALH5O7L62/action/citation_signature","submit_replication":"https://pith.science/pith/T3ZH7MLMLV4Q44VICALH5O7L62/action/replication_record"}},"created_at":"2026-07-05T10:47:04.458670+00:00","updated_at":"2026-07-05T10:47:04.458670+00:00"}