{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:5T3JCLYXK3UK6MSXLGSJEUWPXL","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":"f74a957ad172eea0f389658e2d4e70c908f04e77b1b0ae047729822bbb14d558","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T12:45:03Z","title_canon_sha256":"69378e129fc2f4f34aa8ffab9c29b8251f76273548881c5cac738d2b94b74a9a"},"schema_version":"1.0","source":{"id":"2605.12058","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.12058","created_at":"2026-05-22T01:04:06Z"},{"alias_kind":"arxiv_version","alias_value":"2605.12058v2","created_at":"2026-05-22T01:04:06Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.12058","created_at":"2026-05-22T01:04:06Z"},{"alias_kind":"pith_short_12","alias_value":"5T3JCLYXK3UK","created_at":"2026-05-22T01:04:06Z"},{"alias_kind":"pith_short_16","alias_value":"5T3JCLYXK3UK6MSX","created_at":"2026-05-22T01:04:06Z"},{"alias_kind":"pith_short_8","alias_value":"5T3JCLYX","created_at":"2026-05-22T01:04:06Z"}],"graph_snapshots":[{"event_id":"sha256:e2df0c368eca116dfcd5d0c57d556bae0e8d5ba248af75c130b3ea46b8c85352","target":"graph","created_at":"2026-05-22T01:04:06Z","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":"our approach achieves a state-of-the-art average accuracy of 54.9% across multiple mathematical benchmarks, yielding a substantial 7.2% relative gain over standard GRPO and secures an exceptional 93.8% success rate on ALFWorld."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That modulating the single scalar p via annealing will reliably resolve the concentration-stability trade-off across different model sizes, tasks, and sampling budgets without introducing new failure modes not captured by the reported experiments."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"HölderPO unifies token aggregation in GRPO via the Hölder mean with dynamic p annealing, reporting 54.9% average math-benchmark accuracy and 93.8% ALFWorld success."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"HölderPO resolves GRPO's aggregation trade-off by using a tunable Hölder mean with annealed parameter p to control gradient concentration and variance."}],"snapshot_sha256":"921a14289d15c434d5c9bf698699e05d167af22f773ed5a22ff67dea8289679d"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-20T17:01:26.385247Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-20T11:25:11.599221Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-19T23:01:58.411205Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-19T11:33:48.696831Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.12058/integrity.json","findings":[],"snapshot_sha256":"5f5ffa52975d0f31002e74ed111dc23a82e862fd4809b4cbbae06f482fcc3429","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Group Relative Policy Optimisation (GRPO) enhances large language models by estimating advantages across a group of sampled trajectories. However, mapping these trajectory-level advantages to policy updates requires aggregating token-level probabilities within each sequence. Relying on a fixed aggregation mechanism for this step fundamentally limits the algorithm's adaptability. Empirically, we observe a critical trade-off: certain fixed aggregations frequently suffer from training collapse, while others fail to yield satisfactory performance. To resolve this, we propose \\textbf{H\\\"{o}lderPO},","authors_text":"Chenyang Le, Dingli Liang, Jiachen Zhu, Jianghao Lin, Jun Wang, Lingyu Yang, Weinan Zhang, Yihang Chen, Yuxiang Chen, Zhaokai Wang, Ziqin Gong","cross_cats":["cs.AI"],"headline":"HölderPO resolves GRPO's aggregation trade-off by using a tunable Hölder mean with annealed parameter p to control gradient concentration and variance.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T12:45:03Z","title":"Holder Policy Optimisation"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.12058","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-13T05:58:47.841499Z","id":"1c02ce04-4baa-4742-adac-638daa1a1dd5","model_set":{"reader":"grok-4.3"},"one_line_summary":"HölderPO unifies token aggregation in GRPO via the Hölder mean with dynamic p annealing, reporting 54.9% average math-benchmark accuracy and 93.8% ALFWorld success.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"HölderPO resolves GRPO's aggregation trade-off by using a tunable Hölder mean with annealed parameter p to control gradient concentration and variance.","strongest_claim":"our approach achieves a state-of-the-art average accuracy of 54.9% across multiple mathematical benchmarks, yielding a substantial 7.2% relative gain over standard GRPO and secures an exceptional 93.8% success rate on ALFWorld.","weakest_assumption":"That modulating the single scalar p via annealing will reliably resolve the concentration-stability trade-off across different model sizes, tasks, and sampling budgets without introducing new failure modes not captured by the reported experiments."}},"verdict_id":"1c02ce04-4baa-4742-adac-638daa1a1dd5"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:646925b9cd2aadf48bee358bf9cc4a1b8c478f6fe4ab5db5815370479afcc6aa","target":"record","created_at":"2026-05-22T01:04:06Z","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":"f74a957ad172eea0f389658e2d4e70c908f04e77b1b0ae047729822bbb14d558","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T12:45:03Z","title_canon_sha256":"69378e129fc2f4f34aa8ffab9c29b8251f76273548881c5cac738d2b94b74a9a"},"schema_version":"1.0","source":{"id":"2605.12058","kind":"arxiv","version":2}},"canonical_sha256":"ecf6912f1756e8af325759a49252cfbaef349a6d5a9e0e4e2b22fa8b61802a22","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"ecf6912f1756e8af325759a49252cfbaef349a6d5a9e0e4e2b22fa8b61802a22","first_computed_at":"2026-05-22T01:04:06.183005Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-22T01:04:06.183005Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"eG1QT2ujXloQFnGpqYq2myxlZtXcfqBpsQ+qdQQSzZHAQNPw3Donp7V2noNKc8urlwNWLjkIKBZFoEtMZ4T4Cg==","signature_status":"signed_v1","signed_at":"2026-05-22T01:04:06.183707Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.12058","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:646925b9cd2aadf48bee358bf9cc4a1b8c478f6fe4ab5db5815370479afcc6aa","sha256:e2df0c368eca116dfcd5d0c57d556bae0e8d5ba248af75c130b3ea46b8c85352"],"state_sha256":"e9f01be85163617930a593385d7e37cb1c60c4460e1bb186b74e641fb67ee6b8"}