{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:6A6P7AVRLQJPFUS3V2UWQC3GBY","short_pith_number":"pith:6A6P7AVR","schema_version":"1.0","canonical_sha256":"f03cff82b15c12f2d25baea9680b660e3f6f38c6a9ab5a89e8e6fe74236f05b9","source":{"kind":"arxiv","id":"2502.20380","version":2},"attestation_state":"computed","paper":{"title":"Multi-Turn Code Generation Through Single-Step Rewards","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Alexander M Rush, Arnav Kumar Jain, Gonzalo Gonzalez-Pumariega, Sanjiban Choudhury, Wayne Chen, Wenting Zhao","submitted_at":"2025-02-27T18:55:05Z","abstract_excerpt":"We address the problem of code generation from multi-turn execution feedback. Existing methods either generate code without feedback or use complex, hierarchical reinforcement learning to optimize multi-turn rewards. We propose a simple yet scalable approach, $\\mu$Code, that solves multi-turn code generation using only single-step rewards. Our key insight is that code generation is a one-step recoverable MDP, where the correct code can be recovered from any intermediate code state in a single turn. $\\mu$Code iteratively trains both a generator to provide code solutions conditioned on multi-tur"},"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":"2502.20380","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-27T18:55:05Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"bfb5e0449780e480126b66a8cac4fa6eaaefcafd84d433d75655f53b0313ec2e","abstract_canon_sha256":"8ce0b670209cf4799d250bdc6d69a16473cfad3679f68a0e259890a7595ecd4a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:28:14.669039Z","signature_b64":"TShZL8BI5JX4AkcDC6xvnpBmm8jRZ+kko1zMqEQg/pzgT+3mOj1YzxaSOV1+Qz1GZYq4/boA65Vo4HggiEOUCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f03cff82b15c12f2d25baea9680b660e3f6f38c6a9ab5a89e8e6fe74236f05b9","last_reissued_at":"2026-07-05T11:28:14.668614Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:28:14.668614Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Multi-Turn Code Generation Through Single-Step Rewards","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Alexander M Rush, Arnav Kumar Jain, Gonzalo Gonzalez-Pumariega, Sanjiban Choudhury, Wayne Chen, Wenting Zhao","submitted_at":"2025-02-27T18:55:05Z","abstract_excerpt":"We address the problem of code generation from multi-turn execution feedback. Existing methods either generate code without feedback or use complex, hierarchical reinforcement learning to optimize multi-turn rewards. We propose a simple yet scalable approach, $\\mu$Code, that solves multi-turn code generation using only single-step rewards. Our key insight is that code generation is a one-step recoverable MDP, where the correct code can be recovered from any intermediate code state in a single turn. $\\mu$Code iteratively trains both a generator to provide code solutions conditioned on multi-tur"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.20380","kind":"arxiv","version":2},"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/2502.20380/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":"2502.20380","created_at":"2026-07-05T11:28:14.668664+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.20380v2","created_at":"2026-07-05T11:28:14.668664+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.20380","created_at":"2026-07-05T11:28:14.668664+00:00"},{"alias_kind":"pith_short_12","alias_value":"6A6P7AVRLQJP","created_at":"2026-07-05T11:28:14.668664+00:00"},{"alias_kind":"pith_short_16","alias_value":"6A6P7AVRLQJPFUS3","created_at":"2026-07-05T11:28:14.668664+00:00"},{"alias_kind":"pith_short_8","alias_value":"6A6P7AVR","created_at":"2026-07-05T11:28:14.668664+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.18910","citing_title":"REVES: REvision and VErification--Augmented Training for Test-Time Scaling","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17174","citing_title":"Beyond Execution: Static-Analysis Rewards and Hint-Conditioned Diffusion RL for Code Generation","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2604.27308","citing_title":"BoostLoRA: Growing Effective Rank by Boosting Adapters","ref_index":15,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6A6P7AVRLQJPFUS3V2UWQC3GBY","json":"https://pith.science/pith/6A6P7AVRLQJPFUS3V2UWQC3GBY.json","graph_json":"https://pith.science/api/pith-number/6A6P7AVRLQJPFUS3V2UWQC3GBY/graph.json","events_json":"https://pith.science/api/pith-number/6A6P7AVRLQJPFUS3V2UWQC3GBY/events.json","paper":"https://pith.science/paper/6A6P7AVR"},"agent_actions":{"view_html":"https://pith.science/pith/6A6P7AVRLQJPFUS3V2UWQC3GBY","download_json":"https://pith.science/pith/6A6P7AVRLQJPFUS3V2UWQC3GBY.json","view_paper":"https://pith.science/paper/6A6P7AVR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.20380&json=true","fetch_graph":"https://pith.science/api/pith-number/6A6P7AVRLQJPFUS3V2UWQC3GBY/graph.json","fetch_events":"https://pith.science/api/pith-number/6A6P7AVRLQJPFUS3V2UWQC3GBY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6A6P7AVRLQJPFUS3V2UWQC3GBY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6A6P7AVRLQJPFUS3V2UWQC3GBY/action/storage_attestation","attest_author":"https://pith.science/pith/6A6P7AVRLQJPFUS3V2UWQC3GBY/action/author_attestation","sign_citation":"https://pith.science/pith/6A6P7AVRLQJPFUS3V2UWQC3GBY/action/citation_signature","submit_replication":"https://pith.science/pith/6A6P7AVRLQJPFUS3V2UWQC3GBY/action/replication_record"}},"created_at":"2026-07-05T11:28:14.668664+00:00","updated_at":"2026-07-05T11:28:14.668664+00:00"}