{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:P7BOBOF3IRNSBWF574YU5UZ36P","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"f9ccd23d416559166eb0e9ebf9be538e1ef2d7578dec7e173f5b2ac2b6f61b03","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T09:10:03Z","title_canon_sha256":"5ddd9a31ac5b93065c171a89b55373d7f557514e651512075531774943f79199"},"schema_version":"1.0","source":{"id":"2605.13217","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13217","created_at":"2026-05-18T03:08:48Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13217v1","created_at":"2026-05-18T03:08:48Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13217","created_at":"2026-05-18T03:08:48Z"},{"alias_kind":"pith_short_12","alias_value":"P7BOBOF3IRNS","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"P7BOBOF3IRNSBWF5","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"P7BOBOF3","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:45fb6d51d5bd37b81f523b7a8849c3a6e9a11af1822f68a78c7908ff3cc744d2","target":"graph","created_at":"2026-05-18T03:08:48Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"GAGPO constructs a non-parametric grouped value proxy from sampled rollouts and uses it to compute TD/GAE-style temporal advantages, recursively propagating outcome supervision backward through time."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the non-parametric grouped value proxy derived from sampled rollouts provides sufficiently low-bias and low-variance estimates of true advantages to support stable policy updates without introducing systematic errors from grouping or sampling choices."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"GAGPO computes step-aligned temporal advantages from grouped rollout samples without a learned critic, enabling stable policy optimization in multi-turn agent environments."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"GAGPO constructs a non-parametric grouped value proxy from sampled rollouts to compute temporal advantages for critic-free policy optimization in multi-turn environments."}],"snapshot_sha256":"87dc1dd425da78478a61b643ada6d2ceefd08dfa988f98f01a00405b414918ca"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Reinforcement learning has become a powerful paradigm for post-training large language model agents, yet credit assignment in multi-turn environments remains a challenge. Agents often receive sparse, trajectory-level rewards only at the end of an episode, making it difficult to determine which intermediate actions contributed to success or failure. As a result, propagating delayed outcomes back to individual decision steps without relying on costly auxiliary value models remains an open problem. We propose Generalized Advantage Grouped Policy Optimization (GAGPO), a critic-free reinforcement l","authors_text":"Chao Yu, Jinjun Hu, Qiwen Chen, Rongxin Yang, Siyuan Zhu, Yibo Zhang, Zongkai Liu","cross_cats":["cs.LG"],"headline":"GAGPO constructs a non-parametric grouped value proxy from sampled rollouts to compute temporal advantages for critic-free policy optimization in multi-turn environments.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T09:10:03Z","title":"GAGPO: Generalized Advantage Grouped Policy Optimization"},"references":{"count":32,"internal_anchors":0,"resolved_work":32,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Training language models to follow instructions with human feedback , author=. 2022 , eprint=","work_id":"5903b651-f252-45b3-9845-98a1abf380ae","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"OpenAI GPT-5 System Card , author=. 2025 , eprint=","work_id":"92a933bb-6195-449a-8230-f8407d9b8afc","year":2025},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities , author=. 2025 , eprint=","work_id":"8fb5d482-cf5e-4141-b15a-3f607aa09f83","year":2025},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Qwen3 Technical Report , author=. 2025 , eprint=","work_id":"26c7b6ed-f86e-4ed8-b9ed-b1783d90255b","year":2025},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Proximal Policy Optimization Algorithms , author=. 2017 , eprint=","work_id":"5a794cfe-c0fb-4191-a347-b4e935d39d00","year":2017}],"snapshot_sha256":"7ef6813ede6c6d38e211834aa920c77a2b9914485b3c23dc16873a708f730b75"},"source":{"id":"2605.13217","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T19:13:32.766124Z","id":"dcfe022f-a8a4-4dc4-860e-f0559b0f4d8a","model_set":{"reader":"grok-4.3"},"one_line_summary":"GAGPO computes step-aligned temporal advantages from grouped rollout samples without a learned critic, enabling stable policy optimization in multi-turn agent environments.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"GAGPO constructs a non-parametric grouped value proxy from sampled rollouts to compute temporal advantages for critic-free policy optimization in multi-turn environments.","strongest_claim":"GAGPO constructs a non-parametric grouped value proxy from sampled rollouts and uses it to compute TD/GAE-style temporal advantages, recursively propagating outcome supervision backward through time.","weakest_assumption":"That the non-parametric grouped value proxy derived from sampled rollouts provides sufficiently low-bias and low-variance estimates of true advantages to support stable policy updates without introducing systematic errors from grouping or sampling choices."}},"verdict_id":"dcfe022f-a8a4-4dc4-860e-f0559b0f4d8a"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:3e7524dcbe319ae50bd9cce01766e821c1ef0b48eb0bb1834e21f77b079a3566","target":"record","created_at":"2026-05-18T03:08:48Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"f9ccd23d416559166eb0e9ebf9be538e1ef2d7578dec7e173f5b2ac2b6f61b03","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T09:10:03Z","title_canon_sha256":"5ddd9a31ac5b93065c171a89b55373d7f557514e651512075531774943f79199"},"schema_version":"1.0","source":{"id":"2605.13217","kind":"arxiv","version":1}},"canonical_sha256":"7fc2e0b8bb445b20d8bdff314ed33bf3c08326cd8314dafa2fd66030ffd0e837","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"7fc2e0b8bb445b20d8bdff314ed33bf3c08326cd8314dafa2fd66030ffd0e837","first_computed_at":"2026-05-18T03:08:48.459191Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T03:08:48.459191Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"ocoO3gBKtgWSxeH0AUFbnhRzfaXcGa+twSvE1SMgZl4gsTTqeT025wrjZfRyRIhhrwucQaXNpLTUd+wqxuAIDQ==","signature_status":"signed_v1","signed_at":"2026-05-18T03:08:48.459904Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13217","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:3e7524dcbe319ae50bd9cce01766e821c1ef0b48eb0bb1834e21f77b079a3566","sha256:45fb6d51d5bd37b81f523b7a8849c3a6e9a11af1822f68a78c7908ff3cc744d2"],"state_sha256":"fae150d7c75baaf099963d1b1f19cb270fd54c97a6f85d20d11f8887287f51ce"}