{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:Z54I7K4FRLCQCXV4SMMFLFTDZ7","short_pith_number":"pith:Z54I7K4F","canonical_record":{"source":{"id":"2603.00963","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-01T07:40:12Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"5d388401b511a50152114e4bf1410ebd7a6c81ba09f9ed3b14a0cad9942f3b38","abstract_canon_sha256":"69c137aff0e1e46d6cbba0ab413da813755ab6c966f528c2fc221e0a190c1428"},"schema_version":"1.0"},"canonical_sha256":"cf788fab858ac5015ebc9318559663cfd2141eec5644e2158752bd6d0bc37e2d","source":{"kind":"arxiv","id":"2603.00963","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2603.00963","created_at":"2026-06-02T02:04:15Z"},{"alias_kind":"arxiv_version","alias_value":"2603.00963v2","created_at":"2026-06-02T02:04:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.00963","created_at":"2026-06-02T02:04:15Z"},{"alias_kind":"pith_short_12","alias_value":"Z54I7K4FRLCQ","created_at":"2026-06-02T02:04:15Z"},{"alias_kind":"pith_short_16","alias_value":"Z54I7K4FRLCQCXV4","created_at":"2026-06-02T02:04:15Z"},{"alias_kind":"pith_short_8","alias_value":"Z54I7K4F","created_at":"2026-06-02T02:04:15Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:Z54I7K4FRLCQCXV4SMMFLFTDZ7","target":"record","payload":{"canonical_record":{"source":{"id":"2603.00963","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-01T07:40:12Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"5d388401b511a50152114e4bf1410ebd7a6c81ba09f9ed3b14a0cad9942f3b38","abstract_canon_sha256":"69c137aff0e1e46d6cbba0ab413da813755ab6c966f528c2fc221e0a190c1428"},"schema_version":"1.0"},"canonical_sha256":"cf788fab858ac5015ebc9318559663cfd2141eec5644e2158752bd6d0bc37e2d","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T02:04:15.457242Z","signature_b64":"EorKwjoshu1NHTVJeBDUhONOdvTzRGwPnUsNcTlNxD3ooICsnMxBZLD03W2D5L0IE3hRiWZ0iUcmI7EgN5E5Dw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cf788fab858ac5015ebc9318559663cfd2141eec5644e2158752bd6d0bc37e2d","last_reissued_at":"2026-06-02T02:04:15.456706Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T02:04:15.456706Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2603.00963","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-02T02:04:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"8yMelu5hRZk3nkgE0cNXKmQCpluhbl8DZzEQD7TnchpKSayvuNLvnzjbgdZPxv/F+a7gvf+VZcbIqJXY+eVdCg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-03T20:16:31.971001Z"},"content_sha256":"f9cf7da9af9f53941fb62d0caa5eb1f7ef0901cfa16c41ad04e19e44980572cc","schema_version":"1.0","event_id":"sha256:f9cf7da9af9f53941fb62d0caa5eb1f7ef0901cfa16c41ad04e19e44980572cc"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:Z54I7K4FRLCQCXV4SMMFLFTDZ7","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Stabilizing Policy Optimization via Logits Convexity","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Hongzhan Chen, Shiping Gao, Tao Yang, Ting Yao, Xiaojun Quan, Yuhua Zhu","submitted_at":"2026-03-01T07:40:12Z","abstract_excerpt":"While reinforcement learning (RL) has been central to the recent success of large language models (LLMs), RL optimization is notoriously unstable, especially when compared to supervised fine-tuning (SFT). In this work, we investigate the stability gap between SFT and RL from a gradient-based perspective, and show that the convexity of the SFT loss with respect to model logits plays a key role in enabling stable training. Our theoretical analysis demonstrates that this property induces favorable gradient directionality during optimization. In contrast, Proximal Policy Optimization (PPO), a wide"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.00963","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/2603.00963/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-02T02:04:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"tcYJj46ymGvXlMBy47wcDAFPNpWp/kLbt1CFS2/5ITfKY6XaatkODAHkzLtZ+jiyJn+9j5CkGYyllx8vEQJwDw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-03T20:16:31.971396Z"},"content_sha256":"f14c7a9e74d4febe61ec33807cb73705d903d8015bccac7e20f71de69b7fd4f8","schema_version":"1.0","event_id":"sha256:f14c7a9e74d4febe61ec33807cb73705d903d8015bccac7e20f71de69b7fd4f8"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/Z54I7K4FRLCQCXV4SMMFLFTDZ7/bundle.json","state_url":"https://pith.science/pith/Z54I7K4FRLCQCXV4SMMFLFTDZ7/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/Z54I7K4FRLCQCXV4SMMFLFTDZ7/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-03T20:16:31Z","links":{"resolver":"https://pith.science/pith/Z54I7K4FRLCQCXV4SMMFLFTDZ7","bundle":"https://pith.science/pith/Z54I7K4FRLCQCXV4SMMFLFTDZ7/bundle.json","state":"https://pith.science/pith/Z54I7K4FRLCQCXV4SMMFLFTDZ7/state.json","well_known_bundle":"https://pith.science/.well-known/pith/Z54I7K4FRLCQCXV4SMMFLFTDZ7/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:Z54I7K4FRLCQCXV4SMMFLFTDZ7","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":"69c137aff0e1e46d6cbba0ab413da813755ab6c966f528c2fc221e0a190c1428","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-01T07:40:12Z","title_canon_sha256":"5d388401b511a50152114e4bf1410ebd7a6c81ba09f9ed3b14a0cad9942f3b38"},"schema_version":"1.0","source":{"id":"2603.00963","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2603.00963","created_at":"2026-06-02T02:04:15Z"},{"alias_kind":"arxiv_version","alias_value":"2603.00963v2","created_at":"2026-06-02T02:04:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.00963","created_at":"2026-06-02T02:04:15Z"},{"alias_kind":"pith_short_12","alias_value":"Z54I7K4FRLCQ","created_at":"2026-06-02T02:04:15Z"},{"alias_kind":"pith_short_16","alias_value":"Z54I7K4FRLCQCXV4","created_at":"2026-06-02T02:04:15Z"},{"alias_kind":"pith_short_8","alias_value":"Z54I7K4F","created_at":"2026-06-02T02:04:15Z"}],"graph_snapshots":[{"event_id":"sha256:f14c7a9e74d4febe61ec33807cb73705d903d8015bccac7e20f71de69b7fd4f8","target":"graph","created_at":"2026-06-02T02:04:15Z","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":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2603.00963/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"While reinforcement learning (RL) has been central to the recent success of large language models (LLMs), RL optimization is notoriously unstable, especially when compared to supervised fine-tuning (SFT). In this work, we investigate the stability gap between SFT and RL from a gradient-based perspective, and show that the convexity of the SFT loss with respect to model logits plays a key role in enabling stable training. Our theoretical analysis demonstrates that this property induces favorable gradient directionality during optimization. In contrast, Proximal Policy Optimization (PPO), a wide","authors_text":"Hongzhan Chen, Shiping Gao, Tao Yang, Ting Yao, Xiaojun Quan, Yuhua Zhu","cross_cats":["cs.CL"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-01T07:40:12Z","title":"Stabilizing Policy Optimization via Logits Convexity"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.00963","kind":"arxiv","version":2},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:f9cf7da9af9f53941fb62d0caa5eb1f7ef0901cfa16c41ad04e19e44980572cc","target":"record","created_at":"2026-06-02T02:04:15Z","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":"69c137aff0e1e46d6cbba0ab413da813755ab6c966f528c2fc221e0a190c1428","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-01T07:40:12Z","title_canon_sha256":"5d388401b511a50152114e4bf1410ebd7a6c81ba09f9ed3b14a0cad9942f3b38"},"schema_version":"1.0","source":{"id":"2603.00963","kind":"arxiv","version":2}},"canonical_sha256":"cf788fab858ac5015ebc9318559663cfd2141eec5644e2158752bd6d0bc37e2d","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"cf788fab858ac5015ebc9318559663cfd2141eec5644e2158752bd6d0bc37e2d","first_computed_at":"2026-06-02T02:04:15.456706Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-02T02:04:15.456706Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"EorKwjoshu1NHTVJeBDUhONOdvTzRGwPnUsNcTlNxD3ooICsnMxBZLD03W2D5L0IE3hRiWZ0iUcmI7EgN5E5Dw==","signature_status":"signed_v1","signed_at":"2026-06-02T02:04:15.457242Z","signed_message":"canonical_sha256_bytes"},"source_id":"2603.00963","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:f9cf7da9af9f53941fb62d0caa5eb1f7ef0901cfa16c41ad04e19e44980572cc","sha256:f14c7a9e74d4febe61ec33807cb73705d903d8015bccac7e20f71de69b7fd4f8"],"state_sha256":"1c6a30ab3c92d8e5f68dd08a851af5333e4349fefc473796721e5ffe4dd14a07"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"y+jMKinZK40VoZDZYRh+YSDTl9lPhFm44l6XXrXU3FmZhYRHPsOVshUGSCemgJw7NFipdXX8ZQfehF296y/eCw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-03T20:16:31.973468Z","bundle_sha256":"52652cf14ead23698588cf87394003d9ced19dd0d2fdfe9ce2f9a85d915b2cba"}}