{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:BNKJZS3CFJ72WCF2IKS4XVYFIL","short_pith_number":"pith:BNKJZS3C","schema_version":"1.0","canonical_sha256":"0b549ccb622a7fab08ba42a5cbd70542e6e0cb47b6ce3b048b7ab2255f1db2b3","source":{"kind":"arxiv","id":"2608.11674","version":1},"attestation_state":"computed","paper":{"title":"GCPO: Diagnosing and Constraining Subspace Geometry in Rollout RL for LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jingwei Xu, Kai Yang, Kai-Yuan Guo, Wanyu Wang, Yi Wang, Yu Qiao, Zhenbo Yu","submitted_at":"2026-08-12T05:32:22Z","abstract_excerpt":"On-policy rollout methods such as GRPO are central to post-training of large language models, yet they frequently suffer from training instabilities, cross-task capability degradation, and response-length inflation. Although prior work has characterized the subspace geometry of aggregate updates, the stepwise variation of this geometry and its relationship to model performance remain unclear. We introduce Principal-Subspace Overlap, a dimension-corrected measure of individual rollout updates relative to the dominant singular subspaces of pretrained weights. Despite low average overlap, transie"},"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":"2608.11674","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-08-12T05:32:22Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"07b85b54127eb93ab0554b194d67940ea18a91e4acd409cd8ed2c480f7f7867b","abstract_canon_sha256":"2205a1dd847b4eea07c792183e8a91f8fbdd078e67962dab115878de7b51b2c9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-13T01:27:47.473298Z","signature_b64":"/1JJjBQZ+wd4fpoLVw4+8TizTthOrUzpJQlXpQ0pjmJC7/4cSTDayYwuJq8HKaH9zDzXaHeRE/+rdV8V7Og3BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0b549ccb622a7fab08ba42a5cbd70542e6e0cb47b6ce3b048b7ab2255f1db2b3","last_reissued_at":"2026-08-13T01:27:47.470984Z","signature_status":"signed_v1","first_computed_at":"2026-08-13T01:27:47.470984Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"GCPO: Diagnosing and Constraining Subspace Geometry in Rollout RL for LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jingwei Xu, Kai Yang, Kai-Yuan Guo, Wanyu Wang, Yi Wang, Yu Qiao, Zhenbo Yu","submitted_at":"2026-08-12T05:32:22Z","abstract_excerpt":"On-policy rollout methods such as GRPO are central to post-training of large language models, yet they frequently suffer from training instabilities, cross-task capability degradation, and response-length inflation. Although prior work has characterized the subspace geometry of aggregate updates, the stepwise variation of this geometry and its relationship to model performance remain unclear. We introduce Principal-Subspace Overlap, a dimension-corrected measure of individual rollout updates relative to the dominant singular subspaces of pretrained weights. Despite low average overlap, transie"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.11674","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/2608.11674/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":"2608.11674","created_at":"2026-08-13T01:27:47.472161+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.11674v1","created_at":"2026-08-13T01:27:47.472161+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.11674","created_at":"2026-08-13T01:27:47.472161+00:00"},{"alias_kind":"pith_short_12","alias_value":"BNKJZS3CFJ72","created_at":"2026-08-13T01:27:47.472161+00:00"},{"alias_kind":"pith_short_16","alias_value":"BNKJZS3CFJ72WCF2","created_at":"2026-08-13T01:27:47.472161+00:00"},{"alias_kind":"pith_short_8","alias_value":"BNKJZS3C","created_at":"2026-08-13T01:27:47.472161+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BNKJZS3CFJ72WCF2IKS4XVYFIL","json":"https://pith.science/pith/BNKJZS3CFJ72WCF2IKS4XVYFIL.json","graph_json":"https://pith.science/api/pith-number/BNKJZS3CFJ72WCF2IKS4XVYFIL/graph.json","events_json":"https://pith.science/api/pith-number/BNKJZS3CFJ72WCF2IKS4XVYFIL/events.json","paper":"https://pith.science/paper/BNKJZS3C"},"agent_actions":{"view_html":"https://pith.science/pith/BNKJZS3CFJ72WCF2IKS4XVYFIL","download_json":"https://pith.science/pith/BNKJZS3CFJ72WCF2IKS4XVYFIL.json","view_paper":"https://pith.science/paper/BNKJZS3C","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.11674&json=true","fetch_graph":"https://pith.science/api/pith-number/BNKJZS3CFJ72WCF2IKS4XVYFIL/graph.json","fetch_events":"https://pith.science/api/pith-number/BNKJZS3CFJ72WCF2IKS4XVYFIL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BNKJZS3CFJ72WCF2IKS4XVYFIL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BNKJZS3CFJ72WCF2IKS4XVYFIL/action/storage_attestation","attest_author":"https://pith.science/pith/BNKJZS3CFJ72WCF2IKS4XVYFIL/action/author_attestation","sign_citation":"https://pith.science/pith/BNKJZS3CFJ72WCF2IKS4XVYFIL/action/citation_signature","submit_replication":"https://pith.science/pith/BNKJZS3CFJ72WCF2IKS4XVYFIL/action/replication_record"}},"created_at":"2026-08-13T01:27:47.472161+00:00","updated_at":"2026-08-13T01:27:47.472161+00:00"}